1.

GIZ - India Green Energy Corridors

IGEN-GEC

Report on Forecasting, Concept of

Renewable Energy Management
Centres and Grid Balancing

Published by
Consortium Partners

University of Oldenburg, Germany

Fraunhofer IWES, Germany

FICHTNER GmbH & Co. KG, Germany

Ernst & Young LLP, India

Table of Contents

Table of Contents 1
Table of Figures 3
List of Tables 4
List of Abbreviations 5
Executive Summary 6

1 Introduction 9

2 Overview of the Work Package 10

3 Wind & Solar Power Forecasting Infrastructure and Requirements 11

3.1 Existing Infrastructure and Practices for RE Forecasting in India 11
3.1.1 Operational Wind and Solar Power Forecasting in India Today 11
3.1.2 Operational Numerical Weather Prediction System of IMD 12
3.1.3 Role of Forecasting in New Regulation issued by CERC 14
3.1.4 Comments on Current Practice of Load Forecasting 15
3.2 State of the Art Operational Wind and Solar Power Forecasting 17
3.2.1 Overview 17
3.2.2 Numerical Weather Prediction 18
3.2.3 Measure of Accuracy of Wind and Solar Power Forecast 19
3.2.4 Wind Power Forecasting 21
3.2.5 Solar Power Forecasting 28
3.3 The Australian Wind Energy Forecasting System as reference implementation for
India 38
3.3.1 Anemos wind power prediction platform 39
3.3.2 Extreme event warnings 41
3.3.3 Experience 41
3.4 Wind and Solar Power Forecasting Practice in Germany 44
3.5 Recommendations for Wind & Solar Power and Load Forecasting in India 46

4 Establishment of Renewable Energy Management Centres (REMC) 49

5 Balancing Capability Enhancement 52

5.1 Methodology and Introduction 52
5.1.1 Introduction 52
5.1.2 Methodology 53

6 Enhancing Balancing Capacity 54

6.1.1 Wrap-up of interview phase in India 54
6.1.2 Introduction 54
6.1.3 State perspective 54
6.1.4 Central perspective 59

1|P a g e
 6.1.5 Assessment of existing balancing capacity 61
6.1.6 Flexibility of power plants – literature overview 61
6.1.7 Turndown capability in India 61
6.1.8 Theoretical thermal balancing and ramping potential 62
6.1.9 Theoretical hydro balancing potential 65
6.1.10 Conclusion 65
6.1.11 Short-term solutions 66
6.1.12 Mid-term solutions 70
6.1.13 Long-term solutions 75

7 Cost analysis of balancing options 78

7.1.1 Regional balancing 78
7.1.2 Retrofitting 78
7.1.3 Storage options 79

8 Summary and Recommendations 82

8.1.1 Main outcomes and recommendations 82

9 Overall Strategy Roadmap & Recommendations 84

9.1 Summary of the strategy and the recommendations of all three work packages 84

Annexures 86

References 87

2|P a g e
Table of Figures

Figure 1 - Outer and inner domain of the WRF model at 27 km and 9 km ........................................... 13
Figure 2 - WRF model domains with 3 km horizontal resolution at Regional Centres ......................... 14
Figure 3 - Typical pattern of geographical diversity for the daily load curve in the N, W, and S states of
India on a single day (29 March 2014) .................................................................................................. 16
Figure 4: Wind power forecasting time scale ........................................................................................ 21
Figure 5: Temporal development of the one-day forecast error in the German control area of ‘E.On
Netz’(blue) and for Germany (red) ........................................................................................................ 23
Figure 6: Very high resolution model domains, left: UM-4km (grey shaded area, UK Met Office, 70
vertical layers), right: COSMO-DE (DWD, 50 vertical layers). .............................................................. 24
Figure 7: Table of Wind power software models with international operation ...................................... 26
Figure 8 : Example of a ramp event following a shut-down due to high wind speeds .......................... 28
Figure 9: Forecasting methods used for different spatial and temporal scales .................................... 29
Figure 10: Overview of a regional PV power production scheme ......................................................... 29
Figure 11: Shortest-term forecasting scheme using cloud index images. ............................................ 32
Figure 12: Example of nested domains used in the WRF model......................................................... 34
Figure 13: Derivation of global irradiance on tilted surfaces from global horizontal irradiance. ........... 35
Figure 14: Forecast of global irradiance ............................................................................................... 37
Figure 15: RMSE of five forecasting approaches and persistence for three German stations for the first
three forecast days. (1)–(3): different global models plus post-processing, (4)–(5): ............................ 37
Figure 16: Absolute (left) and relative (right) forecast errors ................................................................ 38
Figure 17: Indian wind farm example .................................................................................................... 39
Figure 18: Simplified flow chart of the Anemos short-term model chain (0-72 h) ................................. 40
Figure 19: Anemos. Live GUI for forecasts visualization ...................................................................... 40
Figure 20: Visualization example .......................................................................................................... 42
Figure 21: Example for the platform surveillance processes: Monitoring of SCADA data feed quality.
.............................................................................................................................................................. 43
Figure 22 - Proposal of a load forecasting framework .......................................................................... 48
Figure 23: Focus of report and distinction between short-term (frequency control) and long-term
balancing (scheduling) .......................................................................................................................... 52
Figure 24: Installed capacity and capacity penetration of RE in the analyzed states in India in 2014 . 54
Figure 25: Minimum load of power plants in Gujarat ............................................................................ 62
Figure 26: Total theoretical balancing potential for each state and comparison to installed RE capacity
.............................................................................................................................................................. 63
Figure 27: Theoretical thermal balancing potential in RE rich-states compared to the potential of regions
and all India ........................................................................................................................................... 63
Figure 28: Theoretical state wise ramping potential of all thermal power plants (left) and ramping
demand in Gujarat (right) ...................................................................................................................... 64
Figure 29: Installed capacity ................................................................................................................. 65
Figure 30: Theoretical thermal balancing capability of RE-rich states today and up to 2022 ............... 66
Figure 31: Forecasting quality of three selected DISCOMs in 2014 ..................................................... 67
Figure 32: Overview of natural gas sector in India................................................................................ 70
Figure 33: Storage capacity .................................................................................................................. 71
Figure 34: Thermal Balancing Potential – Comparison between states, regions and India ................. 72
Figure 35: Potential and installed capacity of pump hydro storage in India ........................................ 73
Figure 36: Increase of transmission capacity under the Green Energy Corridor Project Plans ........... 74
Figure 37: Smoothing effects of wind energy supply from RE due to geographical diversification ...... 75
Figure 38: Overview of storage options and their typical storage capacity and possible cycle
durationSource: Sterner/Stadler 2014 .................................................................................................. 76
Figure 39: Cost estimates of retrofitting gas turbines in single cycle and combined cycle plants ........ 79
Figure 40: Cost of electric load shifting for different storage options .................................................... 80

3|P a g e
List of Tables
Table 1: Challenges for balancing and integrating RE in India – state perspective ............................. 58
Table 2: Problems in respect to grid operation and challenges of RE integration – central perspective
.............................................................................................................................................................. 60
Table 3: Overview of flexibility parameters of power plants to be found in literature (international practice
today / state-of-the-art) ......................................................................................................................... 61
Table 4: Starting capabilities of power plants ....................................................................................... 68
Table 5: Improvement potential ............................................................................................................ 70
Table 6: Overview of storage options for scenarios with high shares of RE ........................................ 77
Table 7: Different retrofitting measures, estimated costs and benefits for coal-fired power plants ...... 79
Table 8: Measures to increase balancing capability in RE-rich states in India and qualitative evaluation
of priorities, costs and impact................................................................................................................ 83

4|P a g e
List of Abbreviations

CEA Central Electricity Authority

CERC Central Electricity Regulatory Commission
GFS Global Forecasting System
GoI Government of India
GTS Global Telecommunication System
GW Giga Watt
HPCS High Performance Computing System
HPSEBL Himachal Pradesh State Electricity Board Limited
IMD Indian Meteorological Department
NCAR National Center for Atmospheric Research
NCEP National Centre for Environmental Prediction
NIWE National Institute of Wind Energy
NLDC National Load Dispatch Center
PGCIL Power Grid Corporation of India
RE Renewable Energy
REMC Renewable Energy Management Center
RLDC Regional Load Dispatch Center
SLDC State Load Dispatch Center
STU State Transmission Utility
UTC Universal Time Coordinated
WRF Weather Research and Forecasting

5|P a g e
Executive Summary
The electricity system in India faces several challenges as the energy demand is expected to grow
significantly within the next decades while the domestic energy resources in terms of fossil fuels are
limited. It is important to increase electricity production in order to keep pace with the demand. Primary
objective of the government is to build and efficiently deploy renewable energy for supplementing the
energy requirements of the country. This will also thereby enable the government to reduce greenhouse
gas emissions. The Indian grid has a grid connected RE capacity of 31.69 GW (January 2015).
Integration of large quantities of RE power in the grid has significant challenges both technical and
economic in nature.

The project aims to provide a comprehensive analysis of the current challenges that RE faces and those
that will arise out of significant capacity addition in RE. The Indian grid is currently the 5 th largest in the
world. Maintaining grid stability and power quality is a herculean task with its own legacy of issues.
Variable generations from RE such as wind and solar plants together are posing significant technical
difficulties of grid management. Gauging the future projections of higher share of renewables, it is
imperative to have a good forecast and appropriate balancing action. This report focuses upon
forecasting tools, methods to enhance balancing and the concept of Renewable Energy Management
Centre. Forecasting focuses on the tools used and methods followed to determine accurately the
amount of RE power that will be produced in a scheduling time block. Balancing of the grid lays
emphasis on the tools used and methods followed (current and suggested) to mitigate the effects of
wind and solar variability for one day ahead and for 4 time blocks ahead in a day.

The first section of the forecasting chapter evaluates the current RE forecasting infrastructure and
methods prevalent in the country. Later in the chapter the pros and cons of various forecasting
techniques globally have been elucidated. Forecasting primarily is a necessity to minimize deviations
between schedule and actual dispatch at the SLDC level and if appropriate the same can be undertaken
at the RLDC level. Moreover, the need for forecasting for a grid operator can be different from those
for farm owners/traders. For example, a grid operator with grid balancing perspective will require
forecast at large spatial region and at smaller time frame, however farm owner/traders will require
forecast at smaller spatial region and at day ahead time frame. Different approaches are preferable for
differing time frames to produce the best forecast for each time period and spatial scale. However, it
has been found that most accurate forecasts can be obtained by using many local and global scale
models and combining them to form a single multi model ensemble. This section of the chapter is further
supported by a detailed explanation on the state of the art forecasting techniques undertaken for wind
and solar power across the globe along with the different kinds of accuracy or error measure techniques.

The analysis of the current forecasting scenario in the country depicts that generation forecasting is at
its infancy. However IMD has significant resources and experience in traditional weather forecasting.
Numerous multi model ensembles can be developed and adapted to wind and solar power forecasting
in the country. Pilot forecasting projects were implemented in Gujarat. However, the projects were not
successful due to a various factors and an analysis upon the same is highlighted in later sections of this
chapter. However, from stakeholder consultations, a strong consensus unilaterally was observed for a
need of robust and reliable forecasting systems for the deployment of large RE capacities in India.

With deregulated electricity markets getting common with its high costs of over or under contracting and
buying or selling power in the balancing market, load forecasting has become an integral process in the
planning and operation of electric utilities, system operators and other market participants. In this
chapter, load forecasting methodologies prevalent and practiced globally have also been highlighted.
Methodology practiced in three states of the country, namely Gujarat, Rajasthan and Himachal Pradesh
have been mentioned along with the accuracy levels. Further, this has also been analyzed with the
inputs garnered from POSOCO. It was observed that the deviation due to incorrect load forecast and

6|P a g e
conventional power plants not adhering to schedule is higher than the variability due to renewable
energy sources.

Last section of the forecasting chapter provides recommendations suitable for Indian scenario. It is
recommended that due to the extraordinarily large uncertainty in forecasting, forecasting should not be
done at the wind farm level. There are two very specific reasons identified. One reason is due to spatial
smoothening of the prediction that occurs over large geographical areas. Second reason is due to
forecasting level which corresponds with the spatial scale on which decisions regarding scheduling,
balancing and grid control are usually taken. Further detailed recommendations can be found in later
sections of the report.

This report also contains a brief overview of the REMC concept. However a separate detailed report
covering REMCs has been prepared.

This second section of the report addresses the challenges of balancing generation from RE in India.
The focus is on balancing demand, RE and conventional generation. The overall objective is to avoid
frequency deviations arising out of RE integration. Accuracy of the schedule and dispatch process in
tandem with grid discipline is imperative for optimal performance of the national grid. The balancing
capacity of states using hydro and conventional plants has been evaluated. Measures to improve these
capacities with respect to their technical and economic considerations have been suggested. The
balancing section of the report is divided into four major parts such as stake holder consultation held in
India, assessment of existing balancing capacity, enhancing of balancing capacities and qualitative cost
analysis of suggesting balancing options.

In the stakeholders’ consultation, there were a lot of mixed views garnered from the state and central
perspective. However, there were certain issues such as lack of available capacity of hydro and gas for
balancing due to technical and economic considerations which were repeatedly pointed out by the
stakeholders at both the levels. India has a limited ability to back down conventional generation due to
a variety of technical and economic considerations. Hydro power available for balancing is low in
capacity and also not completely at the disposal of grid operators. Gas availability is a key issue for
thermal plants which can used for secondary as well as tertiary balancing. Concern of managing
variability of RE sources was highlighted as one of the key concerns due to several considerations. It
was pointed out that lack of regional balancing plays a very important role in maintaining the grid stability
in control areas of the grid.

The central grid operator POSOCO highlighted the need for control reserves. The lack of control
reserves puts the onus of frequency regulation on the level of grid discipline. There is a regulation which
provides for 5% control reserve to be maintained by all generators, However compliance and
enforcement of this regulation is low.

With an emphasis upon the variation of RE sources as one of the key concern, it was imperative to
assess the existing balancing capacity prevalent in the states. The second section of this chapter
focuses on the balancing potential that could be theoretically available to the grid operator. A plethora
of issues were identified while conducting the assessment, such as the flexibility of the conventional
plants, ramping potential of plants, availability of storage facilities of power and many more. Indian
conventional generation plant portfolio has plants of a variety of make and age. Their flexibility of
operations varies significantly. It was identified that in the public domain there is no data defining the
actual operating limits of the plants available. Thus estimating the balancing potential available is
difficult and only indicative of the actual potential. The available data indicated that Indian power plants
have a low turn down capability when compared to international standards. This is attributed to a variety
of factors ranging from age to technical configurations of the plants.

7|P a g e
Ramping potentials of plants vary significantly and may need retrofitting to achieve the desired
performance levels. Hydro balancing potential of the states has been evaluated. It was found that on a
state to state basis it varies significantly from sufficient to highly insufficient. Hydro balancing potential
was found to be further restricted by the control and use of plants by the irrigation department. It was
also observed that most hydro power plants in the country are not reservoir based hence cannot be
used for balancing.

Upon conducting the assessment of balancing potential, it has driven an inquisitiveness to study about
the methods of enhancing the balancing capacities. This section of chapter outlines strategies in three
phases such as short term, medium term and long term, which can be implemented to achieve
enhanced balancing capacities.

In the Short Term Solutions it is suggested that improvement in load forecasting would give the grid
operator an improved perspective of the scheduling requirements. RE generation forecasting is critical
to improvement of schedule and dispatch correlation. System operations and plant flexibility need to be
enhanced significantly. The use of central thermal plants for balancing needs to be explored. Revision
of mandatory generation flexibility for new plants is needed. Retrofitting of existing power plants is
required to improve flexibility. Allocation of gas to RE rich states will be helpful to ensure the balancing
needs.

In the Medium Term Solutions it is suggested that use of hydro power plants with storage reservoirs for
intraday balancing needs to be explored and developed. The control areas where balancing is done
need to be increased in geographical size. This would reduce the balancing requirement for the said
control area. There needs to be a regulatory framework to promote regional balancing between the
individual control regions. There is a requirement for development of large scale pumped storage type
hydro-electric plants. Demand side management needs to be regularized and streamlined in the
country.

In the Long Term Solutions it is suggested that the wind generation is to be dispersed over a large
geographical area. Geographical dispersion of WEGs is known to reduce the overall balancing
requirement of the system. It is suggested that power storage options need to be explored and a
significant push towards the R&D of these technologies is required.

These approaches and solutions will help to salvage the immediate concerns of balancing RE sources
and also enables the country to plan for its future outlook. In the last section of this chapter, a qualitative
cost analysis has also been conducted along with prioritization of tasks. Regional balancing, retrofitting
and storage options are some of the balancing options for which the cost analysis was performed. It is
suggested that the options of regional balancing and retrofitting need to be explored to their complete
potential before storage projects are undertaken. This helps the reader to decide upon the priorities of
actions with qualitative comparison of various parameters.

8|P a g e
1 Introduction
As of January 2015, 31.69 GW of grid connected generation capacity was from RE sources (excluding
large scale hydro power plants), which is around 12% of the overall installed capacity of 258.7 GW.
Deployment of RE is an important way for India to meet its future energy demand and ensure national
energy security. The GoI has set extensive RE capacity addition targets for the remaining period of the
12th Five Year Plan and the 13th Five Year Plan.

The integration of this planned RE generation capacity with the national grid requires expansion and
modernization of the intra- and interstate distribution as well as transmission grid. This is mainly due to
the geographical distance between centers of generation and consumption as well as due to the
intermittent availability of RE sources and the necessary means for grid stabilization. The requirements
for efficient transport of power and strengthening of the grid has been identified in the comprehensive
transmission plan called "Green Energy Corridors" prepared by Power Grid Corporation of India
(PGCIL).

The Government of India and the Government of the Federal Republic of Germany share a long
standing and successful development cooperation. Sustainable access to energy is a major focus of
this bilateral cooperation. Based on the existing experience and knowledge of Germany in the field of
the exploitation of RE, technical assistance for realizing the "Green Energy Corridors" plan including
technical assistance for forecasting, balancing, market design and network management in connection
with grid integration of renewable power has been considered under Indo- German Technical
Cooperation. This project is realised through "Deutsche Gesellschaft für Internationale Zusammenarbeit
(GIZ) GmbH".

Six Indian states, Andhra Pradesh, Gujarat, Himachal Pradesh, Karnataka, Rajasthan and Tamil Nadu,
have been identified for this project. The main stakeholders are the Ministry of New and Renewable
Energy (MNRE) and implementing institutions of the Ministry of Power i.e., Central Electricity Authority
(CEA), Central Electricity Regulatory Commission (CERC), Power Grid Corporation of India (PGCIL),
National Load Dispatch Centre (NLDC), Regional Load Dispatch Centers (RLDCs), State Load
Dispatch Centers (SLDCs), and State Transmission Utilities (STUs).

9|P a g e
2 Overview of the Work Package
The recent increase in variable wind and solar power generation, future projections of higher share of
RE in the total generation portfolio and associated challenges of grid management make wind and solar
power forecasting a mandatory task for the Indian electricity grid. Owing to higher penetration of variable
wind and solar resources, appropriate balancing actions are becoming increasingly complex.
This work package addresses the following challenges,

• Necessary measures for grid stabilization

• Implementation of appropriate forecasting techniques and balancing capabilities
• Establishment of an effective control infrastructure.
•
The above challenges have been addressed by conducting a detailed analysis of the Indian electricity
sector as a whole and on an individual basis in selected states. This analysis provides a reliable
inventory of the current electricity sector and its potential to meet the needs for an increased RE
integration. A special focus of this report is on the question of whether the state-of-the-art instruments
for forecasting and balancing are appropriate for the Indian situation and which specific changes should
be applied to them to make them suitable for the Indian scenario.

Based on this analysis, recommendations for the implementation of forecasting techniques and
balancing actions and for the establishment of an effective control infrastructure are provided.
The work package comprises of the following four tasks,

1. Analysis of wind and solar power forecasting infrastructure and requirements

2. Proposal for establishment of Renewable Energy Management Centers
3. Balancing capability enhancement
4. Overall strategy and roadmap

10 | P a g e
3 Wind & Solar Power Forecasting Infrastructure and Requirements
For the implementation of adequate wind and solar power forecasting schemes in India, an inventory
of currently applied techniques including the available data from the power system and the
meteorological department, and the communication structure for the dissemination of this data is
necessary. The partners of the consulting consortium have visited the selected Indian states and
collected information – mainly through personal communication with relevant stakeholders in the states
–necessary for the analysis of the current situation.

The existing systems and practices on wind, solar and load forecasting in these states have been
reviewed and potential improvements have been identified. Forecasting practices which are –
successfully operating globally - and specifically in Germany are presented and analysed with respect
to their applicability for the selected Indian states. On this basis, a recommended framework for the
implementation of a sophisticated forecasting system for wind and solar electricity generation is
described. This framework includes both, the forecasting techniques (software, data, etc.) needed for
obtaining relevant information, and the infrastructure necessary for the generation and communication
of this information.

3.1 Existing Infrastructure and Practices for RE Forecasting in India

3.1.1 Operational Wind and Solar Power Forecasting in India Today

At the time of the field visits (in February 2015) no mature operational wind or solar forecasting systems
were observed in the states.

A pilot project for wind power forecasting was started in Gujarat in October 2014 and ended in February
2015. The following commercial providers of wind power forecasting services participated (on non-
commercial basis) in this project

• AWS Truepower
• Meteologica
• 3 Tier
• DNV-GL (former Garrad Hassan)
• Earth Networks

The preliminary results of the pilot project were less satisfactory according to the Gujarat SLDC. The
achieved accuracy was reported to be in +/- 30 % range for approximately 60-65% of the times per 15-
minute time block during high wind scenarios. In low wind scenarios, even this accuracy was difficult to
achieve.

These results do not necessarily represent the overall capabilities of the participating forecast providers.
The pilot project was initiated without past measurements and forecast data resulting in a ‘un-trained’
forecast system since it is expected that all systems perform significantly better in a well prepared
operational environment. In addition, the forecasts were provided on a pooling station basis.
In Tamil Nadu, Spanish RE forecast provider Vortex will start the operation of a wind power forecast
system in cooperation with the National Institute of Wind Energy (NIWE) on behalf of the Indian Wind
Power Association. No results of this project are available yet.

11 | P a g e
3.1.2 Operational Numerical Weather Prediction System of IMD
Since 2009, the India Meteorological Department (IMD) operates a High Performance Computing
System (HPCS) and is running a version of the Global Forecast System - GFS T574/L64. This high
resolution global forecast model and the corresponding assimilation system are adopted from the
National Centre for Environmental Prediction (NCEP), USA. The spectral model has a horizontal
resolution over ~ 22 km and runs twice a day (00 UTC and 12 UTC). In addition to this, the meso-scale
forecast system WRF (Weather Research and Forecasting) is being operated twice in a day, at 27 km,
9 km and 3 km horizontal resolutions for forecasting up to 3 days using initial and boundary conditions
from the GFS-574/L64 model. At ten other regional centres, very high resolution mesoscale models
(WRF at 3 km resolution) are operational. NWP-based forecast products from IMD are mainly prepared
to support cyclone warning service. All NWP products are routinely made available on the IMD web site
(www.imd.gov.in).

IMD operates a full data management cycle comprising reception of data through the Global
Telecommunication System (GTS), processing of observations for various operational uses,
dissemination of final products, and archival of model outputs. The entire observational data stream is
received at IMD along with processes for the generation of the initial conditions of the NWP models, the
generation of the global and regional forecasts, and for the generation of numerical guidance for the
operational forecasting offices.

IMD implemented a Multi-Model Ensemble (MME) model using the output of five NWP models - IMD
GFS T574, ECMWF T799, JMA T899, UKMO, and NCEP GFS. The model outputs were interpolated
at a grid resolution of 0.25° x 0.25° latitude/longitude. A rainfall-based scheme was applied to determine
the weights for each model at each grid. The ensemble forecasts (up to day 5) were then generated at
the 0.25° x 0.25° resolution. District level forecasts were then generated from the ensemble forecast
fields by averaging all grid points in a particular district.

For wind and solar power forecasting the most significant forecasting model is the regional mesoscale
analysis system - WRF supported by the National Center for Atmospheric Research (NCAR) but
developed as an open source model by several US research entities.

WRF takes the first guess of the initial field from the GFS global analysis and produces a modified
mesoscale analysis for each specified time (00 and 12 UTC) of the operational run, performed with 27
km horizontal resolution and 38 vertical levels. For this step all conventional observations in the region
(20°S to 45°N and 40°E to 115°E) are additionally considered. The boundary conditions from the GFS
forecast are updated each time for consistency with the improved mesoscale analysis. WRF is then
integrated for 75 hours with a nested configuration (27 km mother and 9 km child domain, see Fig.1)
and configured with cloud microphysics, cumulus, planetary boundary layer and surface layer
parameterization, etc. The whole WRF forecasting process has been scheduled to provide forecasts at
00 UTC and 12 UTC daily.

The forecasts from the 9 km domain are further downscaled to 3 km to prepare the initial and boundary
conditions for WRF to be run at a higher resolution of 3km. The nest down component of the system
can be utilized as many times as possible to generate the initial and boundary condition for many sub-
domains inside the area covered by the 9 km domain. The high-resolution 3-km WRF model can be
separately configured for forecasts over sub-domains around (currently four) regional centres of IMD
(Fig. 2).

The operational weather forecasting structure of the IMD shows a good potential for contributing to RE
forecasting system in India. The products which are best suited for application in this context are the
global-scale Multi-model Ensemble (MME) model and the meso-scale Weather Research & Forecasting

12 | P a g e
(WRF) Model. WRF already provides worldwide solar and (mainly) wind power forecasts of good
quality. It is expected that a high level of operational experience with WRF can be found at IMD.

It is highly recommended to include inputs from IMD in solar and wind power forecasting activities due
to IMD’s competence in meteorological forecasting and numerical modeling. However, IMD does not
have experience in energy-related issues (e.g., forecasting needs by the energy sector, energy specific
statistical post-processing of the forecast results, etc.). Therefore, financial resources should be
provided – primarily for training – to enable IMD to fulfil the role of a key partner in future solar and wind
power forecasting in India.

Source - IMD

Figure 1 - Outer and inner domain of the WRF model at 27 km and 9 km

Source: IMD

13 | P a g e
 Figure 2 - WRF model domains with 3 km horizontal resolution at Regional Centres

3.1.3 Role of Forecasting in New Regulation issued by CERC

In March 2015, CERC published its proposed ‘Framework for Forecasting, Scheduling & Imbalance
Handling for Wind & Solar Generating Stations at Inter-State Level’ according to which RE forecasting
needs to be done both by the RE power producer and the concerned RLDC. While the forecast by the
RLDC would be done with the objective of securing grid operation, the forecast by the RE power
producer would be plant-centric and form the basis of scheduling. The RE power producer may choose
to utilize its own forecast or the regional forecast given by the concerned RLDC (via its REMC).

At this point it is necessary to distinguish between technically motivated needs and economically
desired mechanisms. That is, if market mechanisms are excluded, there is generally no need for plant
specific (i.e., single-site) forecasts and forecast information is necessary only on control area level
(responsibility of RLDC). This is because at the RLDC level, grid stability, security of supply, etc. are
potentially affected by the availability of high-quality forecasts.

Site-specific forecasting does not contribute to improvements of overall forecast quality on the more
important regional level. This is because site-specific forecasting uses post-processing techniques that
are adapted (or tuned) to the specific single site case. Therefore, much effort is spent on the optimization
of site-specific forecasts without any benefit for the regional level. This leads to a misallocation of
personnel and financial resources.

14 | P a g e
Similarly regional forecasts should not be used for forecasting the output of single plants. Regional
forecasts are optimised for the regional level and do not take into account any local information and
therefore will result in inferior quality forecasts compared to explicit site-specific forecasts.
The question of whether RE producers should contribute to the costs of forecasting (and possibly its
improvement) should be separated from the technical concerns.

3.1.4 Comments on Current Practice of Load Forecasting

Load forecasting involves the accurate prediction of total system load and peak system load in various
time scales within the electrical utility’s planning horizon and for different geographical regions
depending on the grid topology. Forecasts for different time horizons are important for different
operations of an electrical utility. Accordingly, load forecasting is classified as short-term, medium-term
and long-term load forecasting.

Short-term load forecasting has become increasingly important since the rise of competitive energy
markets. It can help to estimate load flows, make decisions that can prevent overloading and reduce
occurrences of equipment failures. Consequently, hourly and daily forecasts up to a few days ahead
are of primary interest in everyday market operations.

Since high costs of over- or under-contracting and selling or buying power on the balancing market are
common in deregulated electricity markets, load forecasting has become a central and an integral
process in the planning and operation of electric utilities, energy suppliers, system operators and other
market participants.

Scientific approaches to load forecasting are less developed compared to other sectors which apply
forecasting techniques. This is mainly because of the huge amount of exogenous variables like weather
conditions, social events that influence the load and the need of modeling human behavior as part of
load forecasting. The famous half-time peak during TV broadcastings of soccer matches is an example
of one such peak loading event. Load time series also exhibit seasonality at many different time scales.
Keeping this diversity in mind, many different methods for load forecasting have been developed and
applied, with varying degrees of success.

Current methods may be classified into two categories (Weron, 2006) - statistical methods (like similar-
day, exponential smoothing, regression r, and time series methods) and methods based on artificial
intelligence techniques (like neural networks, fuzzy logic, expert systems, and support vector
machines).

Statistical methods predict the current load value by using a mathematical combination of the previous
loads and/or previous or current values of exogenous factors, e.g., weather and social variables.
Although of statistical nature, they allow some physical interpretation making understanding load
behavior easier.

Artificial intelligence-based methods are inherently flexible and capable of dealing with non-linearity.
They do not require any prior modeling experience and the employed algorithms automatically classify
the input data and associate it with the respective output values. They are ‘black box’-type tools and to
a large extent do not allow for including knowledge of physical relationships between model
components.

It needs to be mentioned that so far no single load forecasting model or algorithm has been established
to be superior for all markets and market participants. This is because, depending on the region with its
climatological and socio-economic characteristics, electricity demand comes in differing mixtures of
industrial, commercial and residential load (see also Fig. 3).

15 | P a g e
Figure 3 - Typical pattern of geographical diversity for the daily load curve in the N, W, and S states
of India on a single day (29 March 2014)

Source: S.K. Soonee, POSOCO.

The most suitable load forecasting method can usually only be identified empirically. Often, several load
forecasting methods are used in parallel or hybrid. Such solutions combine the best features of different
models. Despite the huge economic impact of load forecasting, manual techniques which are largely
based on the personal skill of a single load forecaster in the utility’s control center are in practice.

This situation is reflected in current European practice of operational load forecasting. A variety of
methods based on the above mentioned techniques including manual forecasting are applied by
different utilities. This is done by in-house staff as well as outsourced to third party companies that
specialise in load forecasting. A huge amount of adaptation has to be done to the very specific needs
of a certain utility with its specific supply area and customer mix. Load forecasting accuracy is generally
high and is not considered a bottleneck in power forecasting (compared to RE power forecasts).

The available information on load forecasting in the considered Indian states is limited. In Himachal
Pradesh, load forecasting is done manually based on data from the previous years. No external load
forecasting software tools are applied. Short term load forecasting is done by HPSEBL for each 15
minutes time block for the next day on a daily basis. Efforts are made to keep the deviation of demand
within 3–4% of the estimated demand.

In Rajasthan, load forecasting is currently outsourced to a consulting service run by a former employee.
However, the accuracy of the load forecasting services is below expectations and a tender is ongoing
for further activities on load forecasting. Several forecasting activities are reported with respect to the
long-term time scale. This task has to be completely separated as it provides necessary information for
grid and generation planning purposes only.

16 | P a g e
In Gujarat, all six DISCOMs regularly submit their area’s demand in 96 blocks to SLDC, Gujarat. The
DISCOMs provide day ahead and real time revision of demand to the SLDC. The following factors are
kept in mind for load forecasting

 Category wise load - Jyoti Gram Yojana (JGY-24 hr power feeder),Urban, Agricultural and
Industrial loading
 Agricultural timing
 Agricultural group operation
 Special crop
 Irrigation methods
 Industrial stragger holiday
 Public holiday
 Special event like elections, bandh , world cup / matches etc.
 Office time/recess time

The SLDC combines the demand of all DISCOMs and calculates the final total forecasted demand by
adding pool loss and auxiliary loss. The accuracy of load forecast varies from 3 % to 5 %. However,
during rainy season and any abnormal atmospheric change, the forecasting error is higher for some
blocks. There is no standardised load profile for state. However there can be day wise, season wise,
festival wise profile of past years and these are used as the basis for future forecasting.

According to a statement from POSOCO, load forecasting has been identified as one of three major
reasons for deviations in power scheduling, besides conventional power plants not adhering to schedule
and variable generation from renewables. A study on these factors in Tamil Nadu and Gujarat showed
that the deviation due to incorrect load forecasting and conventional power plants not adhering to
schedule is larger than the deviation due to wind power variability. This put emphasis on the huge
importance of load forecasting in a scenario with high-quality RE forecasting.

An example of the regional diversity of load in India is given in Fig. 3. Any non-manual (i.e. software
based) load forecasting system needs to automatically consider the respective characteristics of the
profile resulting from climatological, economic, and social conditions.

3.2 State of the Art Operational Wind and Solar Power Forecasting

3.2.1 Overview

The availability of solar and wind energy is largely determined by the prevailing weather conditions and
therefore characterized by strong variability. Consequently, power generation from these sources
cannot easily be matched to the electricity demand like power generated with conventional plants. This
introduces new challenges not only for the operation of single wind and solar power plants, but with the
expected integration of large shares of fluctuating RE it will have important consequences for
organization, structure, and management at all levels of electricity supply system. In this context, any
adaptation within the power system, ranging from intelligent power plant scheduling, demand side
management or on the long-term restructuring of grid topology, the introduction of storage capacities,
and for renewable energy trading. This will require very detailed information on the expected power
production from these sources on various temporal and spatial scales.

Accurate forecasts of renewable power production therefore are an essential factor for a successful
integration of large amounts of renewable power into the electric supply system, aiming at precise
information on timing and magnitude of power generation from these variable sources. Forecasting of
renewable power generation is a rather new subject compared with system load forecasting. System
load forecasting traditionally is a significant concern in the operation of electric power systems. As a

17 | P a g e
consequence, accuracies of wind and solar power forecasts are today generally not as accurate as load
forecasts; however, they are catching up very fast.

Generally, wind and solar power forecasts derive future power generation through either numerical
weather prediction models or statistical approaches, and more often than not relying on both.
Experience suggests that the overall shape of wind energy production can be predicted most of the
time, but significant errors in the level and timing of wind energy production can occur. Therefore,
electric power system operators are interested in both the uncertainty of a particular forecast and the
overall accuracy of forecasts in general.

Generally, wind and solar power forecasts for the near term tend to be more accurate than forecasts
for longer terms. They also show strong geographical aggregation benefits. Aggregating wind and solar
power plants over a region significantly reduces the forecasting error by up to a factor of two.
Combining different forecasting models into an ensemble wind forecast can also improve the
forecasting accuracy by up to 20%, as measured by root mean square error. For example, typical
accuracies for German wind power forecasts show 10-15% root mean square error of installed wind
capacity for a single wind project, drop to 5-7% for day-ahead forecasts for a (regional) control area,
and reduce to 4-6% for day-ahead wind forecasts for complete Germany. More importantly, with
aggregation, the impact of forecast errors on individual plants is not as severe because the aggregate
forecast of all plants drives the generation scheduling.

Forecasting of solar power faces issues similar to those of wind. However, a significant difference is the
highly predictable component given by the deterministic path of the sun across the sky – but still leaving
a high degree of stochastic variability mainly due to clouds. Solar resource forecasting is not yet as
mature due to a delay of around one decade in the introduction of scientific forecasting methods.

3.2.2 Numerical Weather Prediction

Any wind and solar power forecast which is produced in time scales of more than several hours is based
on the results of Numerical Weather Prediction (NWP). The NWP model is basically a computer
simulation of the Earth's atmosphere. All its processes including the ones on the land surface and in
the oceans will affect the weather. Knowing the current state of the atmosphere (i.e., the weather
conditions), future changes in the weather are predicted by the model. NWP is based on a set of
mathematical equations which describes all of the relevant atmospheric processes. These equations
are solved for each grid box to predict the values at that point a certain time later. Model calculations
start with a description of the current atmospheric conditions at discrete points on a three-dimensional
grid. This numerical process is repeated many times using a model-specific time step, producing a
forecast –as long as the desired forecast horizon is reached.
NWP allows for the assimilation of observations to initialize the model to begin with a best initial state.
These weather observations are provided by ground weather stations, radiosondes, weather satellites,
ships, and more. At a basic level, standard operational NWP models run at national centers can provide
an estimate of expected changes in weather, and thus wind speed, clouds, and solar irradiance due to
large-scale forces. Customized NWP, however, allows prediction of the wind or solar resource that is
tuned to the specific application and location of the wind or solar plant and also provides an opportunity
to assimilate specialized local observations in and around the power plant.
Numerical weather prediction (NWP) systems have been historically designed to benefit major public
sectors and thus concentrated primarily on optimized predictions of temperature and precipitation. The
NWP forecasts are usually not optimized for forecasting wind and solar power. Recent research
programmes have started to improve the NWP models e.g. by assimilating data derived from wind and
solar power production into them. However, this development is still in its infancy and has not been
transferred into forecasting practice.

18 | P a g e
NWP models mainly differ in the spatial model domain and the resolution of the model grid. Global (i.e.
worldwide) and regional (limited area, meso-scale) forecast models are run with typical spatial
resolutions of the model grid between 50 km (global) and approx. 2 km (regional). Global forecast
models are designed to predict large scale synoptic weather patterns. But the continuous increase in
computing power allows the global models to overtake the current role of the regional models down to
about 10 km horizontal resolution. Currently most national European meteorological services run
models at 5-10 km resolution with a tendency in the next few years towards 1-3 km grid resolution.
These high resolution models then most likely will be ‘nested’ into a global model at around 25km
resolution providing the necessary boundary conditions for the calculation.
The value of NWP forecasts not only depends on the quality of the observations, the data assimilation
system and the physical models, but to a large extent on the knowledge on how certain (or uncertain)
a particular forecast is and what possible alternative developments might occur. In simple words, a
good forecast is of low value, if it is not known that it is good. Forecast errors result from a combination
of initial analysis errors and model deficiencies, the former typically dominating during the first few days.
To compensate for these shortcomings there has been a trend in the last two decades toward ensemble
forecasting, which means the realization of a number of forecast model runs using perturbed initial
conditions. With this strategy one can estimate the probability of various events and also the uncertainty
associated with a particular forecast. Analysis errors amplify most easily in the sensitive parts of the
atmosphere, in particular where strong baro-clinic systems develop. These errors then move
downstream and amplify and thereby affect the large-scale flow. To estimate the effect of possible initial
analysis errors and the consequent uncertainty of the forecasts, small changes to the initial analysis
are made, creating an ensemble of many (currently 50) different, “perturbed”, initial states. Model
deficiencies then are represented by a stochastic process. This is the principle of ensemble forecasting.
In addition, it is beneficial to combine the results of multiple NWP models where available. Typically,
even this ‘poor man’s ensemble' approach produces a better forecast than any single model.

3.2.3 Measure of Accuracy of Wind and Solar Power Forecast

A thorough evaluation of the performance of an RE power forecast system needs the utilisation of a
standardized methodology. Generally, the forecast error is defined as the difference between the
measured and the forecasted value. For a forecast lead time k, the forecast error for the lead time (t+k)
is
𝑒(𝑡 + 𝑘)𝑡 = 𝑃(𝑡 + 𝑘) − 𝑃̂ (𝑡 + 𝑘)𝑡

where P(t+k) is the measured power at time (t+k), P^(t+k)t is the power forecast for time (t+k) made at
time t, and e(t+k)t is the error corresponding to time (t+k) for the forecast made at time t.
It is convenient to introduce the normalized (or relative) forecast error ε for comparing forecast
performances of differently sized wind or solar power systems by relating the error to the installed
capacity Pinst
1 1
𝜀(𝑡 + 𝑘)𝑡 = 𝑒(𝑡 + 𝑘)𝑡 = [𝑃(𝑡 + 𝑘) − 𝑃̂(𝑡 + 𝑘)𝑡 ]
𝑃𝑖𝑛𝑠𝑡 𝑃_𝑖𝑛𝑠𝑡

Any prediction error is composed of a systematic (μe) and a random (xe) contribution

𝑒 = 𝜇𝑒 + 𝑥𝑒

Here μe is a constant and xe is a random variable with zero mean.

19 | P a g e
The systematic error is described by the model bias, which is the average error over the complete
evaluation period (with number N of data used for the model evaluation) and is calculated for each
horizon k as,
𝑁
1
𝐵𝐼𝐴𝑆(𝑘) = 𝜇̂ 𝑒 = 𝑒̅𝑘 = ∑ 𝑒(𝑡 + 𝑘)𝑡
𝑁
𝑡=1

Two basic criteria for the description ofthe forecast systems’ performance are common - the Mean
Absolute Error (MAE) and the Root Mean Squared Error (RMSE). The Mean Absolute Error is
𝑁
1
𝑀𝐴𝐸(𝑘) = ∑|𝑒(𝑡 + 𝑘)𝑡 |
𝑁
𝑡=1
The most commonly used measure is the Root Mean Squared Error (RMSE) based on the Mean
Squared Error (MSE),
𝑁
1
𝑀𝑆𝐸(𝑘) = ∑ 𝑒 2 (𝑡 + 𝑘)𝑡
𝑁
𝑡=1
The Root Mean Squared Error then is

𝑁 1/2
1
𝑅𝑀𝑆𝐸(𝑘) = √𝑀𝑆𝐸(𝑘) = ( ∑ 𝑒 2 (𝑡 + 𝑘)𝑡 )
𝑁
𝑡=1

Systematic and random errors contribute both to the MAE and RMSE. An alternative to the use of the
RMSE is the Standard Deviation of Errors (SDE)
𝑁 1/2
1
𝑆𝐷𝐸(𝑘) = ( ∑( 𝑒(𝑡 + 𝑘)𝑡 − 𝑒̅𝑘 )2 )
𝑁−1
𝑡=1

SDE describes the standard deviation of the error distribution and therefore includes only the random
error.
As the BIAS and MAE measures are associated with the first statistical moments of the forecast error,
they are directly related to the produced energy. The RMSE and SDE values are associated with the
second statistical moment, i.e., related to the variance of the forecast error. In the latter case, large
prediction errors show a larger influence.
These error measures can be calculated using both, the absolute prediction error e(t+k)t or the
normalised prediction error ε(t+k)t. A normalised error measure yields results independent of the size
of the considered energy system. In this case, the error measures are referred to as relative or
normalised BIAS (rBIAS, nBIAS), etc.
In practice, the installed capacity Pinst is mostly used for normalisation. Historically, electric utilities have
preferred this convention because the installed power is a more robust value denoting the maximum
achievable production from RE sources and always giving a constant non-zero value. However, this
obviously does not allow for a description of the error as a percentage of measured (or predicted) power.
To overcome the division-by-zero problem, the forecast evaluation may be done over a longer period
and then normalising by the average measured power production over the complete period.
In any case, we strongly recommend the implementation of a proper forecast system evaluation strategy
including the choice of a standardised set of criteria. This procedure should be documented, for
example, in an evaluation handbook which then is mandatory for all instances of RE power forecasting.

20 | P a g e
3.2.4 Wind Power Forecasting

When looking at the needs for wind power forecasting from a practical perspective the diverse
requirements become evident. Grid operators require information on very short time scales of minutes
to hours while the energy traders have traditionally operated on a day-ahead schedule. Long term
operation planning and system maintenance even need longer lead forecasts. Today, no single weather
forecasting system is able to deliver sufficiently high quality on all these temporal scales. Statistical and
very short-term forecasting methods are most beneficial for the very short term scale below around
three hours, and NWP will best perform beyond a few hours up to several days. Spatial scales have
also to be considered when designing a forecast system. A wind farm operator may be only interested
in local phenomena and thus needs forecasts tuned for a particular wind farm or turbine. Grid balancing
requirements, on the other hand, need to consider wind power availability from all wind farms and
utilities in their service area.
Different approaches are preferable for differing time frames to produce the best forecast for each time
period and spatial scale and ideally these technologies are combined in an optimal manner. Fig. 4
illustrates at which time scale different forecasting techniques show optimal performance. At the
shortest time scales, say below 5 minutes, persistence forecasts are the first choice. Thus, predicting
no change in wind power at these very short timescales is preferred. However, real-time observations
are necessary for this approach. With increasing forecast lead time beyond one hour predictive models
become essential. For the shorter modeled scales (1 to 3 h), it is most important to combine real-time
observations with predictive models. These ‘‘now-casting’’ models assimilate local data and use known
dynamics and physics relationships to intelligently extrapolate the solution forward in time according to
the local weather conditions. From around 3 h onward, conventional NWP models become the
prediction technology of choice. Typically NWP models require time to spin-up or come to a balanced
state, but with the utilization of local data, using dynamic data assimilation methods such as the real-
time four-dimensional data assimilation (RTFDDA) method described below, the spin-up period can be
significantly shortened resulting in useful predictions in the time range of 3–12 hours. Beyond about 6
hours, more distant weather phenomena influence the local weather conditions more and more.
Therefore, many NWP modellers choose to use data assimilation methods that leverage spatial
correlations over larger spatial scales so that observational data and atmospheric boundary conditions
derived from global weather models further away begin to influence the NWP model solution.
Furthermore, combining the results of multiple NWP models improves the overall result. Typically, this
ensemble approach produces better forecasts than any single model.

Figure 4: Wind power forecasting time scale

Persistence
0-1 h

NWP + FDDA
3-12 h Climatology
>14 d
Rapid cycle
models
0-2 h
NWP + 3DVAR
12 h - 2 weeks

1h 3h 6h 12h 24h 7d 20d

Forecast time

21 | P a g e
 Figure shows the benefit of combinations of various technologies with their specific time scales.
FDDA - four-dimensional data assimilation
3DVAR - three-dimensional variation methods
Adapted from S.E. Haupt et al., 2014.

The Model Chain

By far most forecast systems employed by utilities in Europe and U.S are based on NWP forecasts.
This is dictated by their need for information in time scales longer than several hours which only can be
provided by physical modelling of the relevant atmospheric processes.

For a long time, wind power forecast models have been divided into physical and statistical models.
Physical models use physical considerations as long as possible to reach to the best possible estimate
of the local wind speed before using Model Output Statistics (MOS) or other statistical techniques to
reduce the remaining error. Statistical models try to find the relationship between measured wind power
data and a set of relevant variables including NWP output. This is commonly done by applying ‘black-
box’ models, for example Artificial Neural Networks (ANN). Frequently, statistical models use some
knowledge of the wind power properties for tuning the models to the specific domains, turning into more
‘grey-box’ models. Today, mostly all operational forecast systems apply a combination of statistical and
physical models – using the best of both worlds.

For a physical model, i.e. which explicitly has to calculate the wind speed at the point of interest, the
stages of the forecast process (or in the model chain) are the calculation of the wind speed at the hub
height of the turbine (downscaling), the conversion to power output using the turbine’s power curve
(relating wind speed to power output), and the up-scaling of the results for single turbines to regional or
national aggregates.

Wind speed and direction from the relevant NWP model grid point needs to be transferred to the hub
height of the turbine. For this purpose the appropriate level of the NWP model has to be selected –
having in mind typical hub heights of modern wind turbines of around 100 m. For instance, this has
recently caused the European Centre for Medium Range Weather Forecasts (ECMWF) to provide wind
speed values at 100 m height as output. The NWP model results can be obtained for the exact location
of the wind farm or turbine as well as for a grid of surrounding points.

In the physical approach, a meso or micro-scale model is used for downscaling. Differences between
both are due to domain size, resolution, parameterizations and numerical solver. Generally, micro-scale
models benefit from their capability to resolve very small scales down to meters.
The downscaling process yields a wind vector (speed, direction) at the turbine hub height which then is
converted to electric power with a power curve. Taking a power curve from the manufacturers’ data
sheet is straight forward, however, for a large number of wind turbines or farms more generalized power
curves have been used with success. Also, Power curves statistically estimated from the forecasted
wind vector and measured power has been applied. The calculation of the wind farm power finally needs
the summation over the individual turbines and to take into account the wake effects between the
turbines.

This conversion process may benefit from the inclusion of measured power data. This is likely to reduce
the residual errors in statistical post-processing. Of course, this makes the availability of online data
mandatory. In case that only offline data is available, model calibration can only be done in hindsight.
This is the reason that many system operators prescribe to receive online data from wind farms to be
integrated in their online forecasting tools.

Up-scaling from single wind turbine or wind farms to the total area of interest is in most cases is the
final step in the forecasting chain, since utilities usually want a prediction for the total area. The ‘trick’ in

22 | P a g e
this up-scaling step is not to use predictions for all the wind turbines available, but for a representative
subset only. What helps in this respect is that the error of distributed turbines/farms is reduced
compared to the error of a single turbine/farm.

By using the statistical approach only, a direct estimation of regional wind power from the input
parameters is possible. This is done via a combination of input such as NWP output of wind speed,
direction, temperature etc., even from various model levels together with on-line measurements such
as wind power, speed, direction etc. Mostly, a combination of physical and statistical approaches is
beneficial, using physical considerations as far as necessary to capture the air flow in the region
surrounding the turbines, and using advanced statistical modelling to make full use of all information
from the physical models.

Most of the errors in wind power forecasting arise from the NWP model. Two types of error are mostly
observed: amplitude errors and phase errors. Amplitude errors occur when the level of wind power is
not matched in the forecast, although for example the timing of an upcoming storm event is perfectly
reflected. Phase errors behave different in that they show a mismatch between the real and the
forecasted timing of the storm, although they may forecast the amplitude perfectly.

The example in Fig. 5 illustrates the accuracy of one-day forecasts in the Germany. The obvious
improvement resulted mainly from (i) considering atmospheric stability in the models (reduces forecast
RMSE by more than 20%), and (ii) a combination of different models. Today, forecasting errors of below
5% RMSE for the day-ahead forecast for complete Germany can be expected to be standard for
operational forecasting.

Figure 5: Temporal development of the one-day forecast error in the German control area of
‘E.On Netz’(blue) and for Germany (red)
Source: EWEA

Very Short-term Forecasting

For very short time horizons, the relevant time scales are given by system-inherent factors. One main
factor is the mechanics of the wind turbine (generator, gearbox, yaw mechanism, blade pitch
regulation), with typical time constants of seconds, i.e., the time scales of turbulence. The active control
of wind turbines benefits from forecasting information in this time scale which may be provided using a
LiDAR system viewing upstream and a simple advection scheme of the measured wind field a few

23 | P a g e
seconds ahead of the rotor. If wind power is integrated in small or medium isolated systems the relevant
time scale is given by the type of conventional units and the functions for which the forecasts are
required, e.g., start-ups of Diesel generators with quite short time scales. Also, market mechanisms
may govern the need for very short-term forecasting. If, for example, the market is a 5-min ahead
market, as in some countries, wind power predictions for 5 minutes ahead are of major importance.

The typical approach for very short-term forecasting is to use time series analysis techniques or neural
networks. The easiest technique and a common benchmark is persistence, assuming no change of
wind conditions in the respective time scale. Forecasting with autoregressive algorithms and Artificial
Neural Networks (ANN) both use data from online measurements as input and have been widely
applied.

Numerical Weather Prediction

The main error in the final forecast comes from the meteorological input. It is the wind speed input from
the NWP model that is decreasing the accuracy significantly. Therefore, it is logical to try to improve the
NWP input in order to come up with significant improvement in forecasting accuracy.

Various global forecasting NWPs exist, designed to predict large scale synoptic weather patterns. But
the increase in computer resources during the next years will allow the global models to overtake the
current role of the limited area models (LAM) down to about 10 km horizontal resolution. At the moment
most countries run their models for overlapping European areas at 12-7 km grid resolution. In the next
few years they will move towards 4-1 km grid resolution and therefore will not run an intermediate nested
European grid area anymore. They plan to directly nest their very high resolution models, which then
will cover only the national area, into a global model at 25km or less. For very high resolution
requirements of a European wide short range NWP coverage a need arises for close cooperation and
exchange of NWP products. There are already operational suits running at very high resolution in most
European weather services. Fig.6 shows some examples of model domains.

Figure 6: Very high resolution model domains, left: UM-4km (grey shaded area,
UK Met Office, 70 vertical layers), right: COSMO-DE (DWD, 50 vertical layers).

A large effort to the aim of meteorological forecasts for wind energy purposes has also been made by
the original ANEMOS project. Downscaling techniques with micro-scale, meso-scale and CFD models
have been implemented. Two-way nesting between the different domains has shown to be inferior. Both
approaches, physical modelling as well as statistical models have been used and reported increased
accuracy down to two kilometers grid spacing, another one using an advanced the influence of
horizontal grid spacing on forecast quality in general and especially on phase errors has been

24 | P a g e
investigated. Also an increase of the vertical resolution in the lowest part of the atmospheric boundary
layer improved the results. The increase in horizontal grid resolution not necessarily improves the
forecast quality and one option therefore is to use improved computer capabilities for ensemble
forecasting.

Short-term Prediction Models

This section lists models which use NWP data as input and can be found, directly or indirectly, on the
international wind power forecast market. In the meantime, most if not all model providers have
extended their models to also provide solar power forecasts. Fig.7 presents a non-exhaustive list of
wind power software models developed internationally. The table has been published 2012 and lacks
some currently active providers of wind and solar power forecasts.
Providers which are operating on the international market and are not listed in Fig.7 are Meteologica in
Madrid and Vortex in Barcelona, two rather new Spanish companies providing wind and solar power
forecasts, the latter with forecasting solutions based on the meso-scale numerical model WRF. 3Tier
Environmental Forecast Group (U.S.) works with a nested NWP and statistical techniques for the very
short term mainly in the U.S., but also internationally.
Spin-off companies from university groups which have been early adaptors of wind power forecasting
are ENFOR, a DTU spin-off, and energy & meteo systems GmbH, which came out of Oldenburg
University. ENFOR now runs the Wind Power Prediction Tool (WPPT) a statistical model to find the
optimal relationship between NWP predicted wind speeds and measured power for each forecast
horizon. Energy & meteo systems now operate the physical short-term forecasting system Previento, a
meanwhile multi-model forecasting system making extensive use of on-line measurements of power
production.
The table shown above summarises the wind power software models that are currently in operation.
.

25 | P a g e
 Figure 7: Table of Wind power software models with international operation

Model name Developer(s), Operating company Method Selected locations of applications

Spain, Denmark, Ireland, Northern

Prediktor L. Landberg, DTU Risø, Denmark Physical Ireland, France, Germany, USA, UK,
Japan

Eltra/Elsam collaboration with Informatics and Mathematical Denmark, Australia, Canada, Ireland,
WPPT Modeling at Danmarks Tekniske Universitet (DTU), now: Statistical Holland, Sweden, Greece, Northern
ENFOR, Denmark Ireland

Zephyr Risø & IMM ay DTU, Denmark Hybrid Denmark, Australia

Oldenburg University / energy & meteo systems GmbH, Germany, Northern Ireland, Canada,
Previento Hybrid
Germany Australia

eWind True Wind Inc., AWS Truepower, USA Hybrid USA

Sipreólico University Carlos III, Madrid, Spain & Red Eléctrica de Espana Statistical Spain

Institut für Solare Energieversorgungstechnik (ISET), now

WPMS Statistical Germany
Fraunhofer IWES, Germany

WEPROG /
J. Jørgensen & C. Möhrlen at University College Cork Hybrid Ireland, Denmark and Germany
MSEPS

GH
Garrad Hassan Statistical Greece, Great Britain, USA
Forecaster

AWPPS École des Mines, Paris Statistical Crete, Madeira, Azores, Ireland

M. Perez at Centro Nacional de Energias Renovables

LocalPred &
(CENER) and Centro de Investigaciones Energéticas, Hybrid Spain and Ireland
RegioPred
Medioambientales y Tecnalógicas, Spain (CIEMET)

Aleasoft at the Universitat Polytécnica de Catalunya, Spain

Alea Wind Statistical Spain
(UPC)

SOWIE Eurowind GmbH, Germany Physical Germany, Austria, Switzerland

Instituto de Engenharia de Sistemas e Computadores do

Porto (INESC), Instituto de Engenharia Mecânica e Gestão
EPREV Statistical Portugal
Industrial (INEGI) and Centro de Estudos de Energia Eólica e
Escoamentos Atmosféricos (CEsA) in Portugal

EPREV Aeolis Forecasting Services, Netherlands Hybrid Netherlands, Germany, Spain

Source: Foley et al., 2012

26 | P a g e
 Up-scaling

In most cases the power forecast for a region (as given by a TSO) is more important than that for a
single turbine or farm. In only very few cases there is online data available for all turbines in this region.
Therefore, a relationship has to be established between a set of wind farms delivering online data within
a region and the total regional production. Since the errors in the NWP are not uniformly distributed in
time and within a region, forecasting errors for a region are smaller than for a single site. Looking at the
spatial correlation between wind power generation and forecasting errors, it becomes evident that the
uncorrelated part of the error is responsible for the error reduction due to spatial smoothing.
The variability of an averaged time series (e.g., in form of its relative standard deviation) depends on
the variability of the single time series and on the correlation between the different series. Now there
are two effects which reduce the forecast error for a region compared to the one of a single wind farm.
First, the generation as such is already smoother for a region due to the uncorrelated frequencies of
the single wind farm generation profiles, making it thereby more easily predictable, and second, the
forecast errors are uncorrelated on an even smaller length scale.

Ramp and Variability Forecasting

In the early days of wind power generation in Europe, installations were mostly small and well distributed
leading to a quite smooth wind power feed. In recent years though wind farms are installed in clusters
of 100 MW and more. This leads to a much larger probability of ramps, i.e., fast changes in wind speed.
Wind power can suddenly decrease several GW, making control much more difficult. Ramps typically
occur due to sudden gradients in wind speeds (passing of fronts, etc.) or due to the power
characteristics of wind turbines and their respective shut-down behaviour (fig. 8). There is some
evidence that hourly predictions are not sufficient to deal with ramps and shorter timings of the forecast
yield smaller deviations. However, making use of a ramp forecast and integrating it the control structure
of electric utilities is not straight forward and needs a separate treatment than conventional power
forecasts.
Ramps are a challenge for any electric supply system, not only for India. Therefore, any state-of-the-
art RE forecasting system should have a solid ramp forecasting capability.
While ramp forecasting and variability forecasting bear some resemblance, the two are actually quite
different. Variability forecasting refers to large amplitude, periodic changes in wind speed, and has only
recently come into the focus of research. One approach is to define an index of variability based for
example on the standard deviation of a meteorological variable. Potential predictors for variability are,
for example, atmospheric boundary layer height, vertical velocity-wind speed component, geo-potential
height, cloud water content. In general, variability is clearly depending on the occurrence of specific
meteorological conditions.
Two main methodologies for uncertainty forecasting have been established. Statistical approaches
working on single NWP forecasts, and uncertainties derived from ensembles of predictions. While
statistical models already have an estimate of the uncertainty explicitly integrated in the method,
physical models need some additional processing to yield an uncertainty result. As an appropriate tool
for online assessment of the forecast uncertainty confidence intervals have been introduced. Typical
confidence interval methods, developed for models like neural networks, are based on the assumption
that the prediction errors follow a Gaussian distribution. This however is often not the case for wind
power prediction where error distributions may exhibit some skewed characteristics, while the
confidence intervals are not symmetric around the spot prediction due to the form of the wind farm
power curve. On the other hand, the level of predicted wind speed introduces some nonlinearity to the
estimation of the intervals; e.g. at the cut-out speed, the lower power interval may suddenly switch to
zero.

27 | P a g e
 Figure 8 : Example of a ramp event following a shut-down due to high wind speeds

3.2.5 Solar Power Forecasting

Depending on the application and its corresponding time scale, different forecasting approaches have
been introduced. Time series models using on-site measurements are adequate for the very short term
time scale from minutes up to a few hours. Intra-hour forecasts with a high spatial and temporal
resolution may be obtained from ground-based sky imagers. Forecasts based on cloud motion vectors
from satellite images show a good performance for a temporal range of 30 minutes to 6 hours. Grid
integration of PV power mainly requires forecasts up to two days ahead or even beyond. These
forecasts are based on numerical weather prediction (NWP) models.
Methods used for solar power forecasting depend on the application of interest and the relevant time
scale associated with this application (Fig.9). This overview concentrates on bulk solar power
generation and its integration into power grids and consequently covers mainly NWP-based forecasting
with time scales of one day and more.
Also, only photovoltaic solar power generation is considered. However, the introduction of concentrated
solar thermal power technologies (CSP) is similarly in need of high-quality forecasting information. As
much of the methodology described here is applicable as well, the need of direct solar irradiance (DNI)
in these devices involves an additional step in the generation of solar power forecasts with an additional
source for uncertainties.

28 | P a g e
 Figure 9: Forecasting methods used for different spatial and temporal scales

TS - Time series modeling

CM-SI - cloud motion forecast based on sky-imagers

CM-sat - cloud motion forecast based on satellite images
NWP - numerical weather prediction

Design of a PV Power Prediction System

Power prediction of PV systems usually involves several modeling steps in order to obtain the required
forecast information from different kinds of input data. A typical model chain of a PV power forecasting
system comprises the following basic steps (Fig. 10):
• Forecast of site-specific global horizontal irradiance
• Forecast of solar irradiance on the module plane
• Forecast of PV power.
• Regional forecasts need an additional step for up-scaling.
• Forecast of regional power production.

Figure 10: Overview of a regional PV power production scheme

Cloud motion
from satellite

Numerical
weather
prediction

PV power
measurements Solar irradiance Simulation of Regional PV
forecast PV power power forecast

29 | P a g e
These steps may involve physical or statistical models or a combination of both. Not all approaches for
PV power prediction necessarily include all modeling steps explicitly. Several steps may be combined
within statistical models, for example, relating power output directly to input variables like measured
power of previous time steps or forecast variables of NWP systems.
Forecasting of global horizontal irradiance is the first and most essential step in almost any PV power
prediction system. Depending on the forecast horizon, different input data and models may be used.
In the very short-term time scale from minutes to a few hours, on-site measured irradiance data in
combination with time series models are appropriate. In short-term irradiance forecasting, information
on the motion of clouds which largely determine solar surface irradiance may be used. Forecasts based
on satellite images show a good performance for up to 6 hours ahead. From subsequent images
information on cloud motion can be extracted and extrapolated to the next few hours. For the sub-hourly
time scale, cloud information from ground-based sky imagers may be used to derive irradiance
forecasts with much higher spatial and temporal resolution compared with satellite data. Forecast
horizons are limited here through the spatial extension of the monitored cloud scenes and
corresponding cloud velocities.

From about 4–6 h onward, forecasts based on NWP models typically outperform the satellite-based
forecasts. Some weather services, for example, the European Centre for Medium-Range Weather
Forecasts (ECMWF), directly provide surface solar irradiance as model output. This allows for site-
specific irradiance forecasts with the required temporal resolution produced by downscaling and
interpolation techniques. Statistical models may be applied to derive surface solar irradiance from
available NWP output variables and to adjust irradiance forecasts to ground-measured or satellite-
derived irradiance data.
From horizontal irradiance, the irradiance on the plane of the PV modules has to be calculated next.
Different installation types have to be considered. Systems with a fixed orientation require a conversion
of the forecasts of global horizontal irradiance to the specific orientation of the modules based on
information on tilt and azimuth of the PV system. For one- and two-axis tracking systems, these models
have to be combined with respective information on the tracking algorithm. Concentrating PV systems
require forecasts on direct normal irradiance. The procedure then is the same as with any concentrating
system, e.g., solar thermal power plants.
The PV power forecast then is obtained by feeding the irradiance forecast into a PV simulation model.
Generally, two models are used in this step: One for the calculation of the direct current (DC) power
output and another for modeling the inverter characteristics. Both models are widely available in the PV
sector with various degrees of complexity. For PV power prediction, rather simple models show a
sufficient accuracy. Additional input data are module temperature, which can be inferred from available
temperature forecasts, and the characteristics of the PV module (nominal power etc.), usually taken
from the module data sheets.
In the final stage towards an optimized power forecast for a single PV system, the power forecast may
be adapted to measured power data by statistical post-processing techniques. Self-calibrating recursive
models are most beneficial if measured data are available online. Off-line data are successfully used
as well for model calibration.
Prediction of bulk PV power usually addresses the cumulative PV power generation for a larger area
rather than for a single site. This is achieved by up-scaling from a representative set of single PV
systems to the regional PV power production. This approach leads to almost no loss in accuracy when
compared to the addition of the complete set of site-specific forecasts if the representative set properly
represents the regional distribution of installed power and installation type of the systems. In addition
to the power prediction, a specification of the expected uncertainty of the predicted value is important
for an optimized application. As the correlation of forecast errors rapidly decreases with increasing

30 | P a g e
distance between the systems, the uncertainty associated with regional power prediction is generally
much smaller than for single PV systems.

Models for Solar Irradiance Forecasting

Solar power production is essentially determined by the incoming solar irradiance – forecasting
therefore has first of all to concentrate on this variable. Depending on the application and the
corresponding requirements with respect to forecast horizon and temporal and spatial resolution,
different data and forecast models are commonly applied, as seen in Fig. 10 above.

Time Series Modelling

Time series models provide solar irradiance forecasts using only measured values of solar irradiance
as input. Further measurement values related to solar irradiance, for example, cloud cover, may be
included as well. Time series models make use of high autocorrelation characteristics of irradiance time
series for short time lags. For shortest-term time scales, typically up to 1 or 2 h ahead, time series
forecasts based on accurate on-site measurements will be advantageous. Examples of the time series
modeling approach are Kalman filtering, autoregressive (AR), and autoregressive moving average
(ARMA, ARIMA) models, artificial neural networks (ANNs). Time series approaches may include not
only on-site measured data but also additional input from NWP models. This allows for an extension of
the forecast horizon from some hours to some days.

Cloud Motion Vectors

With increasing forecast horizon, time series models are unable to account for any development of
clouds which is the major influence on the temporal and spatial variability of irradiance. Any technique
which is able to detect the horizontal cloud motion in sufficient detail provides valuable information for
a cloud motion forecast in the corresponding time scales. Currently, satellite images and ground-based
sky imagers are common sources for this data.
Geostationary satellites with their high temporal and spatial resolution offer the potential to derive the
required information on cloud motion. A common scheme of solar irradiance forecasting based on cloud
motion vectors derived from satellite images consists of the following steps (Fig. 11): Cloud information,
in form of normalized cloud index1 data, is derived from satellite data and motion vector fields are
calculated from consecutive cloud index images. Future cloud situations then are estimated by
extrapolating the motion of clouds, i.e., by applying the calculated motion vector field to the actual image
adding a noise-filtering smoothing filter. Surface solar irradiance then is derived from this forecast cloud
index images using commonly available methods to derive solar irradiance from satellite images.
Information on cloud motion as a basis for shortest-term forecasting may also be derived from ground-
based sky imagers. They offer a much higher spatial and temporal resolution compared to satellite data,
including the potential for capturing sudden changes in the irradiance, i.e., ramps on a temporal scale
of less than 1 minute. The maximum possible forecast horizon strongly depends on the cloud condition
and is limited by the time the monitored cloud scene has passed the location or area of interest, which
depends on the spatial extension of the monitored cloud scenes in combination with its velocity. The
forecasting procedure, involves similar steps as for satellite-based forecasts.

1Cloud index is a normalised value relating the digital satellite count to a measure of cloudiness. Its range is from
0 (clear sky) to 1 (overcast).

31 | P a g e
 Figure 11: Shortest-term forecasting scheme using cloud
index images.

NWP-based Irradiance Forecasts

As physical modeling is essential for any forecast more than several hours ahead only NWP-based
forecasts are candidates for forecasts of sufficiently high quality in the time domain of many hours and
above.
Today, more than ten global NWP models are available worldwide, most prominent the European
Centre for Medium-Range Weather Forecasts' Integrated Forecast System (ECMWF-IFS) with a
horizontal grid spacing of 16 km, and the Global Forecast System (GFS) of the US National Center for
Environmental Prediction (NOAA/NCEP) operated at a spatial resolution of 50 km. Several of these
NWP models offer global horizontal irradiance (GHI) as direct model output, and some also provide
forecasts of direct and diffuse irradiance. Though this output can in principle be used for solar power
forecasting, practically additional post-processing techniques have to be applied to yield sufficient
forecast accuracy. To achieve a higher spatial resolution, local or regional models are employed for
downscaling the output from global forecast models for specific regions to a finer grid of typically less
than five km with hourly resolution.
A global model which has proven its high quality as a basis for solar power forecasts and surface solar
irradiance and different cloud parameters in several investigations is the IFS model of ECMWF. It
provides solar irradiance as a direct prognostic variable and its model performance is constantly
increasing. The increasing grid resolution and the improvements with respect to radiation and cloud
calculation schemes are especially important for solar irradiance forecasting.
A second class of numerical models apart from global models are meso-scale models which are
restricted to limited areas, but generally solve the governing equations on a much finer grid which

32 | P a g e
enables them to resolve smaller atmospheric phenomena as local circulations, thunderstorms, and
topographically forced wind flow. Although mostly developed as special-purpose tools, they are
nowadays widely used by weather services for everyday weather forecasting. Prediction of surface
solar irradiance should in principle benefit from using meso-scale models as they have the potential to
reflect small-scale features like broken cloud fields and heterogeneous surface characteristics which
strongly influence the spatial and temporal dynamics of surface solar irradiance.
The most commonly used meso-scale model in the scientific world is the Weather Research and
Forecasting Model (WRF) of the US National Center for Atmospheric Research (NCAR) which has been
developed as a open source model in a collaborative effort of several institutes. It has been shown its
potential for wind and solar forecasting in many research projects, but is also used as operational
weather forecasting model in many places as, for example, by the India Meteorological Department.
As limited-area models, meso-scale models require input from global NWP models for initialization and
boundary conditions. Usually, the input data chosen strongly influences the performance of a meso-
scale model. To achieve the intended high spatial resolution in a meso-scale model with reasonable
computing time, a stepwise increase of the spatial resolution is achieved by a nesting procedure as
shown in Fig. 3.12, where domains are ‘nested’ into each other thus receiving their boundary conditions
from the higher level domain.

Post-processing of NWP Model Output

Different concepts may be applied to account for local effects not resolved by the given grid size of the
numerical model and to derive improved site-specific forecasts. One is the downscaling to higher spatial
resolution by meso-scale models, as shown in Fig. 3.11. Also, statistical post-processing methods, for
example, model output statistics (MOS), can be used to represent local effects. They also allow for the
correction of systematic deviations in dependence on different meteorological parameters and for
inferring solar irradiance from other model output parameter of a NWP model. Generally, post-
processing is applied directly to the output of a global or meso-scale numerical model.
An established technique for post-processing in meteorology is Model output statistics (MOS).It is based
on a statistical regression and relates e.g. observations of the wanted forecast variable (predictand) to
model forecast variables (predictors). The set of predictors may include any relevant information, for
example, prior observations and climatological values. Most important are high-quality measurements
from local surface stations. In case of solar irradiance, also satellite-derived values may be used. MOS
has successfully been applied for solar irradiance forecasting.
NWP output shows frequently systematic deviations depending on the meteorological situation. In these
cases, a bias correction based on statistical analyses can improve the results. For example, ECMWF
irradiance forecasts showed a significant overestimation of solar irradiance for intermediate cloud cover
which could be reduced by a bias correction depending on the predicted cloud situation.
A common post-processing task is the temporal interpolation of global model forecast output, which
typically is provided with a temporal resolution of 3 hours. Solar power forecasting with one hour time
resolution therefore needs an adequate interpolation techniques. Commonly, for solar irradiance these
are based on a clear-sky model to account for the typical diurnal course of solar irradiance.
Another technique for improving forecast quality is spatial averaging. In variable cloud situations, this
reduces fluctuations in the irradiance forecast values, but in homogeneous (clear-sky or overcast)
situations it does not harm the forecast quality. For example, ECMWF irradiance forecasts show best
results when averaging over 4×4 grid points, corresponding to a region of 100km×100km, is applied.

33 | P a g e
 Figure 12: Example of nested domains used in the WRF model

Solar PV Power Forecasts

A major step in solar power forecasting not included in the meteorological models is the derivation of
electric power output from solar irradiance forecasts. Most precisely, explicit physical models describe
the conversion process in solar cells and subsequent inverters from irradiance to electric output
including the conversion of solar irradiance from the horizontal to the array plane. Here, temperature
forecasts may be integrated and improve the conversion to electric power. For most utility-scale
applications, an up-scaling of the power output to regional power is necessary as a final step of the
conversion.

Irradiance on the Plane of Array

The first step in any PV power forecast is the calculation of the proper irradiance values for the given
module plane from the forecasts of horizontal global irradiance. Systems with fixed tilt angles, tracked
systems, and concentrating systems have to be treated differently at this stage. A large number of
empirical models for this conversion is available. These models are mostly based on a decomposition
of global irradiance into its direct, diffuse, and ground-reflected components, which are derived
separately from the corresponding horizontal values (Fig. 3.13). It should be noted that this
straightforward principle implies the use of two empirical relationships which add further uncertainties
to the forecast.
PV simulation

PV simulation models are widely used in the context of planning and sizing of PV systems and for yield
estimation. Their accuracy generally meets the requirements for PV power forecasting. However, as
necessary input information the specification of module and inverter characteristics, and the geometric
orientation of the modules has to be known. This implies a practical problem, because usually detailed
system information is generally not available for all PV systems within the framework of, e.g.,a regional
PV power forecast.

34 | P a g e
 Figure 13: Derivation of global irradiance on tilted surfaces from global horizontal
irradiance.

Up-Scaling to Regional Power Prediction

For most forecasts used in grid operation and management the cumulative power production for the
particular control area is required. This is done by up-scaling from a representative set of single sites.
The quality of the up-scaling process depends on the choice of the representative data set, which needs
to be representative for the full ensemble in terms of the response to the irradiance conditions. A correct
representation of the spatial distribution of the nominal power is most important in this respect. Also,
the distribution of PV system orientations influences the ensemble power production and the subset
should represent the ensemble distribution. Finally, the mix of module technologies has to be
considered due to the different part-load behavior of different module types.

Evaluation of Solar Irradiance and PV Power Forecasts

Analyzing and specifying the forecast accuracy is an essential step towards a valuable solar power
forecast. A good knowledge of forecast accuracy is the basis for any decision and assists in choosing
between different forecasting products. In research, forecast evaluation is necessary for model testing
and further model development. To assess the forecast‘s accuracy, it is compared with the
corresponding measurements ofsolar irradiance or solar power. In the following evaluation results are
shown for a data set consisting of mmeasured hourly global irradiance data foroneGerman site
(Mannheim, 49.2° N, 9.56° E, height 96 m a.g.l., DWD). Evaluations are done for the period January 1
to October 31 in 2007. Forecast data are based on the 0:00 UTC model run of the ECMWF deterministic
global model with a spatial resolution of 25 km x 25 km and a temporal resolution of 3 h in combination
with an appropriate post-processing procedure.

35 | P a g e
 Statistical Error Measures

A quantitative description of forecast accuracy is obtained using statistical error measures and it is
common practice to use the root mean square error (RMSE) as a main score. As additional measures
the standard deviation of the error and the mean absolute error (MAE) are used. The MAE is appropriate
for applications with linear cost functions, i.e., when the costs associated with a poor forecast are
proportional to the forecast error. The RMSE is more sensitive to large forecast errors, i.e., more
appropriate when large errors cause disproportionately high costs, as in case of utility applications. For
applications where decisions are related to threshold values, the consideration of the frequency
distribution of the forecasts is useful. For comparisons of distribution functions of measured and
predicted time series the Kolmogorov-Smirnov test integral has been used.
Statistical error measures for solar irradiance or solar power forecasts differ significantly depending on
whether they are based on daylight hours only or on all hours of a day. Within the solar resource
assessment community, usually only daytime values (i.e., with non-zero irradiance) are considered for
accuracy assessments. However, parts of the electric utility sector are used to evaluate electric power
forecasts including all hours of the day, resulting in lower RMSE values. Evaluation results shown here
are based on daylight hours only.
Error measures for solar power predictions are usually normalized and the relative RMSE is used. Both,
the mean solar power and the installed capacity (i.e., the maximum solar power) are used for reference
in practice. The latter is used frequently in the electric utility sector – obviously resulting in a significantly
different error measure. The results shown here are based on the mean value as reference.

Uncertainty Information

Specifying the expected uncertainty of solar irradiance or power predictions is a highly recommended
addition to any forecast. As a basis for quantifying the expected uncertainty of a forecast, the probability
distribution function of either forecast errors or power predictions have to be known. From this
information, confidence or error intervals can be derived indicating the range in which the actual value
is expected to appear with a quantified probability. Two different approaches are common to provide
uncertainty information: (i) ensemble prediction systems and (ii) analysis of simultaneous historical time
series of forecasts and observations. As pointed out earlier, the output of ensemble prediction systems
is usually interpreted in terms of a probability density function, i.e., as a direct measure of uncertainty.
For the second approach weather-specific error intervals are determined. Assuming normally
distributed errors the distribution function of errors is completely described by the standard deviation of
the forecast errors. Values of the standard deviation of the error are determined and confidence
intervals with a given uncertainty level then can be given. Fig. 14 shows forecasted irradiance values
with 95% confidence intervals derived by this method in comparison with observed irradiance.
Another approach to determine the probability distribution of forecasts is quantile regressions to
determine the probability distribution of normalized power depending on predicted normalized power.
The advantage of this method is that no assumption on distribution functions is necessary.
In figure 14, forecast of global irradiance with confidence intervals of an uncertainty level of 95%
compared with measured irradiance for six days in May 2007 for a single site (left) and for two hundred
measurement stations in Germany (right) is depicted. In clear-sky situations the forecast quality is high
and prediction intervals are narrow. On days with variable cloudiness, large deviations are to be
expected for forecasts with hourly resolution for single stations (left). As shown in this figure, regional
forecasts (right) show an improved agreement between forecasts and measurements, with narrow
confidence intervals for different weather situations.

36 | P a g e
Figure 14: Forecast of global irradiance

Evaluation of Different Approaches to Irradiance Forecasting

In the framework of the International Energy Agency‘s SHC Task 36 ‘Solar Resource Knowledge
Management’ a common benchmarking procedure for solar irradiance forecasting has been developed
and applied to seven different solar irradiance forecasting procedures. The algorithms used in the
different forecasting methods can be grouped into three categories: (i) combination of a global NWP
model with a post-processing technique involving historical surface observations or satellite-derived
irradiance data, (ii) combination of a meso-scale NWP model and a post-processing technique based
on historical surface observations, and (iii) forecasts of the meso-scale model WRF without any
integration of observation data. A common one-year data set of measurements of hourly irradiance data
from four different European climatic region was chosen. The different forecasting approaches are all
based on global NWP model predictions, either ECMWF global model or GFS data.
A strong dependence of the forecast accuracy on the climatic conditions was found. For Central
European stations the relative RMSE of the NWP based methods ranges from 40% to 60%, for Spanish
stations relative RMSE values are in the range of 20% to 35%. Irradiance forecasts based on global
model numerical weather prediction models in combination with post-processing showed best results.
All proposed methods perform significantly better than persistence. For short term horizons up to about
six hours the satellite based approach leads to best results. Selected results are shown in Fig. 15 and
16.

Figure 15: RMSE of five forecasting approaches and persistence for three German stations for the
first three forecast days. (1)–(3): different global models plus post-processing, (4)–(5):

(1)
(2)
(3)
(4)
(5)

37 | P a g e
 Figure 16: Absolute (left) and relative (right) forecast errors

RMSE (solid lines with circles) and bias (dashed lines) of five different forecasting approaches and persistence in dependence
on the month for the first forecast day.

3.3 The Australian Wind Energy Forecasting System as reference implementation for India

To provide an example of an operating forecast system, the Australian Wind and Solar Energy
Forecasting systems, AWEFS and ASEFS, are presented which are run by the Australian Energy
Market Operator AEMO (see http://www.aemo.com.au/Electricity/Market-
Operations/Dispatch/AWEFS). The systems’ availability is 100 % since its implementation in 2008. It
could serve as a potential reference implementation for an India-specific forecasting system.
AWEFS and ASEFS are based on the wind and solar power prediction system Anemos, one of the best
known wind power prediction systems with high reputation. The system is run on this prediction
platform. This platform is a commercial activity as a spin-off of several Anemos research projects
devoted to state-of-the-art wind power prediction in Europe. As a prediction platform, the system allows
for the incorporation of several RE forecast models from different providers and needs at least one of
them.
This Anemos-based Australian forecasting system AWEFS/ASEFS shows several characteristics which
makes it a candidate for a reference implementation in India:
- AWEFS is running locally in the dispatch centers, so it doesn’t depend on internet availability
and access to external prediction providers.
- AWEFS includes algorithms to deal with poor quality or missing measurement/SCADA data.
The implemented prediction models are very robust to missing or poor quality data.
- Grid operation in Australia is – similar to India – delicate due to the large distances covered by
transmission lines and locally high penetration by renewables.
- Like in India, in Australia grid congestions play a major role in handling renewable energy
production. AWEFS/ASEFS handles historic and future turbine and substation unavailability
and grid congestions/wind farm down regulation.
- AWEFS is a high-availability system with 100 % availability since its first installation.

38 | P a g e
This reference implementation can provide significant insights concerning many technical constraints
of a forecasting system for India. Nevertheless, different meteorological and climatological conditions
in both continents do not allow for an a priori equalisation of forecasting performance.

3.3.1 Anemos wind power prediction platform

Basis for the AWEFS implementation is the Anemos platform, which hosts the prediction models and
services. It is a highly standardized software that permits secured handling of information related to
wind power predictions. Data exchange is handled in a Service Oriented Architecture based on web
services. This eases the implementation of server mirroring, load balancing and providing on-site and/or
remote services. In addition, a high degree of availability is made possible.
Due to its modular design and having originated from research, the Anemos platform is ideal for
comparing and optimizing different modelling approaches. It is also suitable for including the predictions
from third party service providers in its comparisons. The system contains a reporting tool, which
supports the automatic and manual evaluation of model performance.
Combined forecasts could be post-processed to apply modifications for any known future scheduled
maintenance events (WPP outage plans) and projected down-regulation or line/substation restrictions
(power limitations). Forecasts for the required aggregations will then be calculated on the combined
and/or post-processed forecasts.
The Anemos platform is designed to be expandable to operations for a large number of wind farms,
new prediction models and advanced customized prediction services by using the same basic
infrastructure.

. Figure 17: Indian wind farm example

Fr
eq
ue
nc
y

Number of turbines available

Demonstrating the problem of limited availability for both, grid (very left; zero turbines available) and turbines.
Limited availability leads to the need for an advanced handling of missing or low-quality data.

39 | P a g e
 Figure 18: Simplified flow chart of the Anemos short-term model chain (0-72 h)

The post-processing module is taking care of grid outages, down regulation and unavailability of wind farms.

Figure 19: Anemos. Live GUI for forecasts visualization

Main window with map view and two selected plots and tabular view window.
The predictions plot shows the forecasted mean values (blue line) and the prediction uncertainties (green area).

40 | P a g e
3.3.2 Extreme event warnings

Managing extreme events such as storm fronts is becoming an ever more important task in light of
increasing amounts of wind energy in the electricity grid. This challenge is tackled with special models
for ramp prediction, coupled with an alarming system for extreme situations, which immediately informs
the user of expected power surges, power declines or turbine shut-down events, also on aggregated
regional/national levels. These warnings are available in the web GUI. This system is based on a rules
engine, Anemos.Rulez, which provides the analysis and alarming part. The engine is generic and can
be operated with custom rule sets, which are developed with respect to the business processes of the
customer and the regional meteorological conditions. Details like time horizons to consider, thresholds
on different levels etc. are configurable.

3.3.3 Experience

The Anemos platform is successfully operating in many operational systems since more than 10 years.
For several installations, the system runs on-site and is integrated directly into the customer’s IT
infrastructure. AWEFS and ASEFS, the prediction system for the Australian energy Market run by the
Australian Energy Market Operator, is a reference for an on-site installation in a high-availability, highly
secured IT environment.
The Anemos system is currently in commercial operation in Europe and Australia. It provides highly
accurate wind and solar power forecasts as well as on-line assessments of the forecast uncertainty in
form of prediction intervals and a reliable operation and maintenance of the system. The ensemble of
this information supports optimal decisions in power system management with respect to fluctuating RE
power generation.
Summarizing, the Anemos approach as implemented for the Australian Energy Market Operator could
serve as a reference for the Indian case and is characterized by:
- Representing the European state of the art in wind and solar power prediction
- Combining long term experience in R&D and commercial RE power predictions world wide
- Extensive experience in implementation and operation of on-site systems in the dispatch center
in business critical high availability IT systems
- Potential extensions for extreme alerting, now-casting to replace missing measured data feeds,
solar power prediction, and many more features.
- Implemented locally at dispatch centers, hence not depending on permanent internet access
to external prediction providers.

41 | P a g e
Figure 20: Visualization example
Figure includes deterministic forecast, 10% and 90% POE forecast uncertainty interval, detected extreme events and forecast

risk index indicator (coloured boxes at the bottom of the plot)

42 | P a g e
Figure 21: Example for the platform surveillance processes: Monitoring of SCADA data feed
quality.

43 | P a g e
3.4 Wind and Solar Power Forecasting Practice in Germany

In the German power sector currently more than 75 GW of wind and solar power are subject to
operational RE forecasting. These renewable power generators are must-run-plants according to the
Renewable Energy Act (Erneuerbare-Energien-Gesetz, EEG) which was established in 2000 with
guaranteed feed-in tariffs. TSOs are required to preferentially feed-in this electricity into the grid over
electricity from conventional sources. The system has only recently been modified to also include a
market premium system.
The German power transmission system is subdivided into four areas, each of them run by one of the
TSOs Tennet, Amprion, 50Hertz, and TransnetBW. As they are responsible for grid operation within
their respective control area they also are demanding for high-quality RE forecasts for these given
areas. Due to the EEG system, single power producers are not in need of having high-quality forecasts,
to be precise, of no forecasts at all. The only exception is if they decide not to supply power according
to the EEG rules but to act in the direct marketing domain. However, this need for forecasting then is
purely due to economic constraints.
This need for regional forecasting for the areas of the four TSOs resulted in the establishment of several
different providers of forecasting services on the German market. In the beginning, no meteorological
expertise was found at the level of the TSOs and they used the power forecasts provided by the services
without further treatment. In the meantime, the TSOs – as well as several of the larger DSOs – have
built up their own forecasting expertise (mainly through educated meteorologists) and now are able to
perform extensive evaluations of forecast performance and to give valuable feedback to the forecast
providers. Furthermore, this capacity lets them deploy own post-processing schemes based on a set of
different power forecasts delivered by different providers. This can be seen as an additional post-
processing at the TSOlevel combining these different power forecasts. Purchasing several RE power
forecasts from different providers has become common practice as it increases knowledge about
forecast uncertainty at relatively low costs2. In this respect, forecast service providers and TSOs more
and more interact and in future the integration of full forecasting services into special divisions within
the TSO structure could be possible.
The following list includes typical characteristics and functionalities of a state-of-the-art forecasting
system operated for German TSOs
- Wind and solar power forecasts for the four control areas of the TSOs
- Forecast horizons of typically up to three days (although RE forecasting can be easily extended
up to 7 days)
- Temporal resolution of the forecasts 15 minutes to one hour
- Capability of additional very-short term forecasts of up to six hours
- Updates on an intra-day time scale
- Forecasts for ramps (time of occurrence, duration, magnitude, ramp rate)
- Detailed information on forecast uncertainty (mostly resulting from probabilistic forecasts)
- Including available on-line measurement data in the forecasting workflow
- Continuous evaluation of the forecasts according to community-accepted accuracy measures
Although all the RE forecast systems are capable of delivering forecasts on any spatial scale down to
single generation plants, the majority of todays’ services – and all forecasts serving control zone

2
As a consequence of the market situation with several suppliers of RE forecasts the market price for wind and solar power
forecasts came down during the recent years by a large amount. Purchasing RE forecasts thus is generally a minor item
compared to e.g. infrastructure (of REMC), personnel, …

44 | P a g e
operation – is providing regional forecast products. Single site forecasting – as has been outlined before
– yields much poorer performance figures in terms of accuracy.
Within that scheme, any forecasting system includes a set of post-processing stepsaiming at
- Reducing systematic forecast errors,
- Accounting for local effects (e.g., topography, surface),
- Accounting for wind farm effects (wakes) in wind power
- Accounting for the influence of selected variables in more detail (e.g., aerosols in solar power),
- Deriving parameters that are not provided as direct NWP model output (e.g., wind speed in hub
height, direct solar irradiance)
- Combining the output of different models.

45 | P a g e
3.5 Recommendations for Wind & Solar Power and Load Forecasting in India

For the purpose of this report, it is expected that the primary need for wind and solar forecasting is to
ensure the stability of the electricity supply in general and in particular of the grid operation. The authors
are aware of further needs and applications of RE forecasting, for example in the domain of market
mechanisms or accounting. Within this report, the requirements for ensuring grid stability are considered
to be of top priority.

We assume a strong consensus among all stakeholders in the Indian electricity sector on the fact that
expected future deployment of RE strongly needs to be supported by state-of-the-art forecasting
schemes for the fluctuating wind and solar power generation. This forecasting functionality shall be a
major component of the Renewable Energy Management Centers (REMCs) to be established in or
attached to the existing regional and state dispatch centers. This consensus was expressed also during
the workshop ‘Enhanced RE Grid Integration with Emphasis on Forecasting, REMC and Balancing
Capacity’, held April 22-23, 2015 in Delhi.

Here we present our recommendations on how forecasting services can be implemented in the
framework of the REMCs to be established.

• Due to the large uncertainty of RE forecasts on a local level, i.e. for single sites, we strongly
suggest not to concentrate on this spatial scale. Larger areas considered in forecasting at SLDC
level result in a smoothing due to the spatial averaging and therefore lead to lower uncertainties.
This forecasting level also corresponds to the spatial scale on which decisions with respect to
grid control, balancing, and scheduling are usually taken.

• When forecasting is done on the SLDC level, there is no need for single power providers to
forecast their own (local) production – except for economic reasons given by the market
mechanisms.

• The recently proposed ‘Framework for Forecasting, Scheduling & Imbalance Handling for Wind
& Solar Generating Stations at Inter-State Level’ foresees that forecasting needs to be done by
both the RE producer and the concerned RLDC. The RE power producer may choose to utilize
its own forecast or the regional forecast given by the concerned RLDC (via its REMC). Here
again, we do not recommend site-specific forecasting and envisage the responsibility for
forecasting on the RLDC level. This issue has been elaborated in Chapter 3.1.3.

• The RE forecast system should at least provide the following functionalities:

- Wind and solar power forecasts on state (i.e., SLDC) level
- Forecast horizons of up to two days
- Temporal resolution of the forecasts 15 minutes
- Updates on an intra-day time scale
- Option for forecasts in the time scale of up to six hours
- Ramp forecasting (time of occurrence, duration, magnitude, ramp rate)
- Detailed information on forecast uncertainty (mostly resulting from probabilistic forecasts)
- Capability to make use of on-line measurement data
- Continuous evaluation of the forecasts according to community-accepted accuracy
measures

• Forecasts in different states may be provided by different forecast service providers. This could
be beneficial as this offers the opportunity of exchanging information on the performance of the
various systems and to compare them with respect to their capability of targeting the Indian
specific meteorological and technical conditions. At a later stage this may be expanded to a

46 | P a g e
 central organisation of forecasting making use of several RE forecasts to yield an optimised
forecasting scheme for each state.

• A high-quality forecast system for wind and solar power needs to be supplemented by a load
forecasting scheme of at least the same accuracy. Available information from the Indian power
sector indicates that this is not yet achieved and load forecasting is primarily done on a manual
and intuitive basis and not using science-based software support. It is therefore highly
recommended to put efforts on the establishment of state-of-the-art load forecasting
techniques. It is likely that different solutions have to be applied for different states. A joint effort
for setting up a common framework for load forecasting – best to be organised by a national
authority – is needed for all states. Within that framework, support could be given to the
individual states to establish state specific operational load forecasting approach. Also, training
on load forecasting techniques should be provided on the national level (Fig. 22).

• Successful implementation of RE forecasting is not only based on high quality forecasting

models but also on the availability of well-trained staff that are familiar with the RE forecasting.
It is recommended to start an educational program including

(i) Basic meteorological concepts

(ii) Post-processing techniques
(iii) Probabilistic methods
(iv) Statistical evaluation of forecast performance.

Training activities by external forecast providers to be offered regularly to staff personnel should be
mandatory.

• Any forecast system includes statistical components (mainly in its post-processing part) which
need some time to adjust to the specific configuration of the application. To optimise this
process, RE forecasts need to be continuously evaluated. At the REMCs, a standardised
evaluation process should be implemented and the results should be communicated to the
forecast providers regularly. A complete evaluation process not only helps to improve
forecasting but also enables the forecast user to monitor forecast quality. A possible option to
develop a standardised process could be a centrally organised Evaluation Handbook which is
continuously updated.

• It is recommended to include IMD’s expertise in future solar and wind power forecasting
activities. As this is a new field of activity for IMD, appropriate resources should be provided.
The link between IMD and the electricity sector needs to be strengthened by bilateral
consultations and training on the specific needs of the sector. IMD may contribute significantly
in training staff people in the REMCs on meteorological forecasting.

47 | P a g e
Figure 22 - Proposal of a load forecasting framework

48 | P a g e
4 Establishment of Renewable Energy Management Centres (REMC)

In view of the expected increase in RE penetration, there is a need to equip the power system operators
with state-of-the-art tools along with real time data of RE generation. The establishment of Renewable
Energy Management Centres (REMC) equipped with advanced forecasting tools, smart dispatching
solutions, and real time monitoring of RE generation, closely coordinating with SLDCs/RLDCs has been
envisaged as a primary requirement for grid integration of large scale RE. Renewable Energy
Management Centres (REMCs) at State, Regional and National level should be co-located with
respective Load dispatch centres (LDC) and integrated with real time measurement and information
flow. There should be a hierarchical connection between the State Load Dispatch Centre, Regional
Load Dispatch Centre and National Load Dispatch Centre.

In an attached contribution to this report, a conceptual recommendation for the Renewable Energy
management Centres (REMC) is provided. REMC is envisaged as the “hub” for all information regarding
RE power generation in its area of responsibility which could be on SLDC-, RLDC- or even on NLDC
level.

REMC should have a dedicated team for managing forecasted RE generation, its despatch and real-
time monitoring to ensure safe, secure and optimal operation of the grid. REMC acts as the RE Single
Point of Contact for the main Grid Operations team. In order to facilitate better coordination between
REMC and the main xLDC teams, it is essential that REMC team should be collocated with the main
LDC team. Recommended functionalities of REMCs are
 Forecasting of RE generation (day ahead and intra-day, ramp prediction etc)
 Online geospatial monitoring of RE Generation – at the transmission grid boundaries & at RE
pooling Stations (through direct Data Acquisition OR through interface with RE Developer
monitoring Systems)
 Responsible for quality and reliability of RE data
 Propagate RE related data to its partner xLDC, Forecasting, scheduling and balancing systems.
 Coordinating with xLDC for dispatching and balancing RE power
 Central Repository for RE generation data for MIS and commercial settlement purposes
 Coordination agency on behalf of xLDC for interacting with RE Developers
 Training and Skill building for RE integration into the grid.

REMC shall therefore comprise the following features

 Data Acquisition, Monitoring and Control

 Geo-Spatial Visualization of RE Generation within the area of responsibility
 Information exchange with xLDC, forecasting, scheduling and control reserve
monitoring/balancing tools
 Data Engineering
 Data Archiving and Retrieval
 Reports and MIS to support commercial settlements
 Future Readiness for advanced functions such as Virtual Power Plants, Storage etc.

Guiding Principles for REMC System

The following requirements are mandatory for a REMC system:

 State of the art SCADA – Functionality

49 | P a g e
  Support real-time monitoring at refresh rates of at least 2-4 seconds and capability to perform
remote control operations
 High availability, based on fully redundant hot standby configuration
 Utilization of existing market standards
 Compliance with all IT-Security standards
 Prevention of cost intensive new developments with high error rates
 Potential for later extension and upgrading without difficulty
 Security of the investment for the next 10 - 15 years

Generally there are two main approaches to establish the REMC system.

1. Integration of new software features into the existing SCADA/EMS - Systems

Most of the existing SCADA/EMS systems are from well-known international Control-System
manufacturer and they are the mainline products of these vendors. Some parts of the features
needed for the REMC are already available in the SCADA/EMS software platform. Apart from the
standard SCADA functionality which is also needed for the REMC, also parts of the EMS
functionalities can be used after adoptions to fulfil the requirements for dispatching the generation
of renewable energy. However, the SCADA/EMS Systems currently in use are procured from
various vendors; therefore the structure is not homogeneous. A significant number of systems is
already approaching the end of their life cycle and need to be replaced.

2. Development of new independent REMC Systems with Interfaces to the existing

SCADA/EMS - Systems
This approach assumes that all REMCs require identical functionality and it would be easier to
introduce one new control system platform (based on state of the art standard products from reputed
SCADA vendors available in the market) and install as much instances of this control system type
as are needed. For the necessary information exchange with the existing SLDCs and RLDCs
international standardized telecommunication protocols shall be used. The advantage of this
approach is the independence from the existing SCADA/EMS - Systems and their different stages
in the evolution process. This will ensure more flexibility and less complexity.

The assessment of the existing SCADA/EMS systems has shown that the control systems installed in
the Indian Power Grid are of different type and make. Although they are based on the mainline standard
products of the different vendors, they have different software releases and different project specific
software packages installed. This means that software adaption would need to be project specific for
individual xLDCs. As described above this is very error prone. Furthermore the procurement could only
be limited to the original suppliers. Competitive procurement procedures will be difficult to implement.
On the contrary, single source procurement for each individual xLDC would be advisable.

In addition it has been shown that some of the existing SCADA/EMS Systems are at their performance
limits. The refresh cycles of the measurements are sometimes above 10 seconds.

These reasons are speaking against the integration of the needed functionality for the REMCs into the
existing control system. To keep the complexity of the overall system as low as possible, it is
recommended that each REMC is exactly assigned to one of the existing SCADA/EMS Systems (SLDC,
RLDC, or even NLDC).

50 | P a g e
 Proposed REMC System Conceptual Architecture

Figure above shows Conceptual Architecture of recommended REMC. REMC at control centre
comprises of following modules:
1. REMC SCADA
2. Forecasting tool (can be a 3rd party tool)
3. RE Scheduling Tool
4. Control Reserve Monitoring Tool
5. WAMS for RE Substations (optional)

REMC components in the field comprise of Data Interface units at the Grid interconnecting and RE
developer pooling substations (can be RTUs or Data interface units for integrating with existing RTUs).
REMC can integrate with SCADA systems of RE developers provided these support standard interface
protocols. Phasor Measurement Units (PMUs) can be provided at critical substations where remote
monitoring is required at each power frequency granularity level.

Further details on REMC are covered under a separate report on “Assessment of existing SCADA/EMS
Control Centres, Telecommunication Infrastructure and Conceptual Design of new REMCs”.

51 | P a g e
 5 Balancing Capability Enhancement

5.1 Methodology and Introduction

5.1.1 Introduction
This part of the project report deals with the challenges of balancing RE generation of RE-rich states in
India. The focus of this section is on hour(s)-or day(s)-ahead balancing of demand, RE and conventional
generation. Optimal load and generation balancing is done in order to avoid frequency deviation.
Balancing in terms of limiting frequency deviation in the short-term (seconds to minutes) will be dealt
with later on.
When physical delivery of power is concerned, better scheduling process and grid discipline is required
to ensure fewer mismatches. Therefore, proper balancing hours- and day(s)-ahead is a necessity for

4. Advanced Frequency Control Concept

proper integration of RE and for system operation in general. Balancing hour(s) and day(s)-ahead is
indirectly linked to frequency deviation. The distinction between the types of balancing is depicted in
Figure 23 balancing hour(s) - and day(s)-ahead and in short-term blend into each other in real system
Mode of operation
operation.
Figure 23: Focus of report and distinction between short-term (frequency control) and long-
term balancing (scheduling)

Focus of this report

14.000
Gujarat state load

9.000 Residual Load

(MW)

wind
solar
4.000
minihydro
Biomass
-1.000

„ Balancing (frequency control) „ Balancing (scheduling) hours- and

seconds and minutes ahead“ day(s)-ahead“

This report assesses the hours- and day(s)-ahead balancing capability in India and recommends
measures for improvement. The general assessment of available balancing capacity, actual practice
and enhancement options for the six states (Himachal Pradesh, Gujarat, Rajasthan, Andhra Pradesh,
Karnataka and Tamil Nadu) is presented in the next section of this report. This assessment is based on
experience of on-site investigations in India during which different stakeholders have been interviewed
and information has been collected. The SLDCs of the six states, the SRLDC in Bangalore, Powergrid,
the Ministry of New and Renewable Energy, the Ministry of Power, POSOCO and the NLDC were
visited. The information presented in this report also includes observations from the first series of
workshops conducted under the IGEN-GEC project.
A 3,more
Slide Sessiondetailed assessment of balancing capacity of one state shall be done; the results will be
5AO.8, 30.09.2013
presented in a separate report. For the purpose of the detailed assessment further analytical tools will
© Fraunhofer IWES
be used and a more in-depth methodology will be applied.

52 | P a g e
5.1.2 Methodology
This analysis is divided into the following steps:
1. Summary of the interview phase in India:
The results of the interviews with Indian stakeholders regarding balancing will be summarized
in this step. During the interview phase, both the perspectives of the states (integration
challenges, technical barriers, etc.) and that of the central level stakeholders were taken into
account. The most pressing challenges from each perspective are summarized within this
section.

2. Assessment of existing balancing capacity in the six states:

In this step, the generation mix of each state is analysed and a technical theoretical balancing
and ramping potential is derived based on installed capacity and technical parameters of plants.
The approach is based on the assumption that only the capacity between minimum and
maximum load of the plant is referred as balancing potential and not the complete capacity.
This is a rather conservative assumption as the total capacity can be used, but in the present
scenario, shut-down of power plants for balancing purposes is avoided if it is not clear that the
load of the next hours or days (weekend) can be taken care of with plants on bar. Furthermore,
the balancing potential by hydro power plants and pump hydro storage is assessed.

3. Enhancing Balancing Options

This step discusses various options to increase balancing capability. Possible measures to do
so are described in this section and include technical measures for regulating and balancing
practices. All possible options are finally categorized into short-term, medium-term and long-
term actions and are prioritized.

4. Costs of balancing options

In this step, the costs for new balancing technologies (storage options, retrofitting, and regional
balancing) are assessed based on literature review and qualitative discussions with
stakeholders. For the discussed storage options, costs of load shifting and investment costs
are compared. For retrofitting of plants a brief overview of possible cost and benefit ranges are
given.

53 | P a g e
 6 Enhancing Balancing Capacity

6.1.1 Wrap-up of interview phase in India

6.1.2 Introduction
Balancing the variable generation from RE is becoming challenging as new capacities are added.
Installed RE capacity in Indian states ranges between 639 – 8,075 MW including wind energy, PV,
biomass and mini-hydro. Except for biomass power plants, all of these RES are intermittent power
sources. The variable generation has to be integrated by the system operator by balancing the existing
flexibility in the system. Today capacity penetration of the states under analysis (Tamil Nadu, Andhra
Pradesh, Karnataka, Gujarat, Rajasthan and Himachal Pradesh) ranges between 18% – 56%
compared to 12% on national level. As most of the balancing is to be done by the state, the integration
challenges vary among the states according to their level of RE deployment which is depicted in Figure
24. The balancing capacity of fossil fuels and hydro power plants in relation to installed capacity of RE
is very different within the states. The challenges in balancing from a state and the central perspective
nstalled are described –
capacity Visited states
here.
Figure 24: Installed capacity and capacity penetration of RE in the analyzed states in India in
2014

35.000

30.000

25.000
RE (PV, Wind, Mini-Hydro)
20.000 Hydro
MW

Nuclear
15.000 Diesel

10.000 Gas
Coal
5.000

-
Tamil Karnataka Rajasthan Gujarat Himachal Andhra
Nadu Pradesh Pradesh

56% 38% 31% 18.5% 18.5% 9%

Capacit y penet rat ion

6.1.3 State perspective 2

aunhof er IWES There are numerous challenges that the states have with respect to integration of RE. The following
major key integration challenges have been identified:

Limited ability to back-down generation:

Description of problem: In states with high penetration of RE (especially wind energy), the other
generation sources need to be backed down during the period of peak RE generation in order to
maintain load-generation balance. If the balance is not maintained, the state contributes to an over-
frequency in the grid. In Tamil Nadu this problem leads to the curtailment of wind generation to stabilize
frequency.
After discussions with the stakeholders at the State Load Dispatch Centres (SLDC) the following
problems have been identified:
 Thermal power plants on bar can only reduce their generation to around 70% of their rated
capacity (where literature values range from 40% to 60%). Below this threshold oil support
is needed to stabilize the combustion process which is costly. Reducing generation below
this limit is therefore avoided.

54 | P a g e
  Retro-fitting to increase the reduction capability needs one year during which the plant is
not available for normal power production.
 Shut-down of power plants is often avoided due to balancing risks in the following hours. If
the load picks up and/or power from RE decreases, the respective thermal power plant may
not be available and on bar again quickly. Minimum time to bring back thermal generation
on-line is around 15 – 24 hours. In case of week-ends start-up may take up to 2 days.
 In some states central power plants are not used for balancing. This is due to the fact that
capacity charges apply even if the capacity is not used. States therefore tend to avoid these
costs without benefit in form of electricity generation. If allocated capacity is not being used;
it may be allocated to other beneficiaries. To avoid the risk of losing the entitlement to the
capacity which is usually needed to cover the electricity demand, balancing with central
units is not practiced.
 In some cases, due to contractual agreements (Power Purchase Agreements), power
plants can only be requested to change generation once a day. Further requests to change
the generation level are charged.3
 Power from central or thermal stations is usually cheaper than the feed-in tariff paid for RE
by the respective state utility. Therefore backing down of thermal plants might be avoided
due to economic reasons. However, stakeholders of SLDC stated that due to their must-
run status backing down of RE is usually the last option irrespective of the economic
implications.
 Backing down of thermal power plants threatens the fulfilment of Plant Load Factor (PLF)
targets of single power plants.

Low availability of hydro power for balancing:

Description of problem: The use of the installed hydro power capacity for balancing purposes is limited
due to four major reasons: the correlation between wind and hydro power availability during monsoon,
the multi-purpose use of water sources for power production and for irrigation, the fact that a large share
of units are run-of-river without any storage options and that pump hydro storage is not largely available
in States with high shares of RE.
 Wind energy in many of the Indian RE-rich states (like Tamil Nadu) is dominantly available
during monsoon season when hydro power is also available due to precipitation [CEA
2013a, PCGIL 2012]. In case water cannot be stored due to full reservoirs, hydro power
becomes must-run since spilling of reservoirs would imply a large economical loss. The
price of electricity from hydro power is cheaper in comparison to the feed-in tariff from wind
energy. In some states, the thermal units are put under maintenance during monsoon
season as sufficient power is available from hydro power and wind energy. This reduces
the problem of spilling hydro power reservoirs or curtailing wind energy. However, if the
total average daily generation during monsoon season is lower than the rated capacity,
there is no spilling of water needed. Instead the power plants can still be used for balancing
by varying their power output during the day. Whether this is usually the case for a
significant share of power plants during monsoon season needs to be investigated in more
detail.
 Multi-purpose use of most of the hydro power units is limiting the flexible dispatch of power
plants. This is true especially for water demand for irrigation purposes. Irrigation is the first
priority and the responsibility for reservoir level control and thus, possible plant dispatch is
usually taken by the state government. Stakeholders at SLDC level stated that they
effectively do not have the control over plants used for irrigation during long periods of the
year and they treat them as must-run according to the operating conditions of state
government. Irrigation demand should be especially low during the monsoon season due
to the high amount of precipitation. During this time the wind energy availability and the

3
Interview with staff at the SLDC in Andhra Pradesh, 25.02.2015.

55 | P a g e
 balancing demand is the highest. Therefore there should be less restriction due to irrigation
demand. In general, this topic needs more investigation to determine the balancing
potential from hydro power during high wind season.
 Run-of-the-river plants cannot effectively be used for balancing as they produce electricity
whenever water is available depending on level of river and rainfall. Thus, run-of-river plants
cannot effectively be used to mitigate integration challenges.
 Pump hydro storage is not fully available. On the one hand existing capacity is limited or
not operated in pumping mode. On the other hand pump storage is partly located in states
where pressure on integration of RE is lower than in other states. Due to lack of bilateral,
inter-state agreements the pump storage is not used effectively to integrate RE.

Low availability of thermal gas-fired power plants for balancing:

Description of problem: Gas-fired, thermal power plants are used to balance variable generation from
RE as their power is always dispatchable. However, the following problems limit the capacity for
balancing:
 Unavailability of gas: For a large share of gas power plants fuel availability is limited leading
to plant load factors (PLFs) of 5% – 25% are reality although they would be used more
often given sufficient fuel supply4.
 Costs of natural gas: Even if gas power plants would have a sufficient amount of fuel
available the dispatch of conventional power plants has to follow the economic merit-order.
Gas-fired power-stations are usually the most expensive to run.
 Due to these economic reasons they are only dispatched if they are necessary for
electricity generation and not for enhancing balancing capability5.

Uncertainty of RE supply:
Description of problem: Due to the absence of forecasting for RE, power availability is uncertain and
thus, production scheduling is challenging. Consequently thermal power plants are not optimally
scheduled. Power plants for example are not shut-down due to uncertainty of RE supply within the time
horizon of the next several hours and days. Assuming that high quality forecasting for RE is applied,
variability in RE generation and residual load coverage should be manageable in most states. A certain
degree of uncertainty of RE generation remains even with forecasting applied and may cause an impact
on proper scheduling practice (i.e. forecasting errors, earlier or later occurrence of ramps than
forecasted, extreme events, etc.). Moreover, sudden fluctuations of wind and solar may lead to
problems in balancing load and generation in the short-term as sudden load variation within the range
of less than 15 min cannot be taken into account for scheduling and generation dispatch.

Rate of change of power contribution from RE

Description of problem: Even if forecasting is applied and uncertainty of power contribution from RE is
decreased, the rate of change of power from wind energy needs to be balanced out which requires an
adequate reaction of conventional power plants (hydro + thermal). Stakeholders at SLDC level claimed
that ramping up and down is difficult. However, severe problems due to high ramp rates of RE have not
occurred so far. The changes in power production from RE are usually still lower than the variation of
the load so that there is sufficient experience to deal with the occurring ramps. In section “In this step,
the costs for new balancing technologies (storage options, retrofitting, and regional balancing) are
assessed based on literature review and qualitative discussions with stakeholders. For the discussed
storage options, costs of load shifting and investment costs are compared. For retrofitting of plants a
brief overview of possible cost and benefit ranges are given.

4
Interview with staff at SLDC Gujarat, 19.02.2015.
5
Stakeholder workshop, 22./23.04.2015, Delhi.

56 | P a g e
Enhancing Balancing Capacity it is also shown that the residual load (load less feed-in from renewables)
does not have significantly higher ramps than the load without feed-in from RE for the case of Gujarat.
However, in the future or in states with higher capacity penetration, a lack of flexible power plants for
balancing would increase the integration challenges occurring from ramp rates.

Lack of regional balancing:

Description of problem: Today balancing of RE is mainly done within states. Balancing potential (i.e.
increase or decrease of thermal power plant generation) available in other states is often not taken into
account for re-scheduling or improved dispatch in the first place. Regional balancing can be incentivized
given efficient market mechanisms to export power. A regulation which covers cost differences between
market prices of green energy and thermal power may be required additionally, in order to ensure that
RE generation is profitable. Currently these two framework conditions are not sufficiently implemented.
For wind power generation transmission charges apply based on installed capacity of the wind farm.
Due to low plant load factors the usage of transmission lines is more expensive than for conventional
units.6 This limits the incentive to export power. To some extent solar energy is exempted from paying
transmission charges. The lack of regional balancing is a problem which is seen as very critical by
almost all stakeholders at SLDC and NLDC level. 7

All challenges for balancing and integrating RE which have been identified and described are
summarised in Table 1.

Apart from depicting integration problems stakeholders at the SLDC level stated the following options
to increase balancing capacity or suggested the following changes in existing regulations:
 The threshold of allowed deviation limit within the Deviation Settlement Mechanism should be
raised. Today 150 MW or 12% of schedule (whichever is lower) is allowed as unscheduled
interchange. It was suggested that the system size (i.e. peak load) or the RE-share should
serve as an indicator to define new thresholds. The argument for this suggestion is that for
states with small electricity systems or given low RE penetration, minimization of unscheduled
interchange is easier to achieve if this suggestion is implemented.
 The increase of options for regional balancing and more efficient use of regional flexibility
options (hydro pump storage) were recommended.
 Establishment of market mechanisms which allow for exporting RE was recommended.
 It was suggested that natural gas supply in India should be increased in general and gas fuel
allocations should be higher for RE-rich states.
 Enhancing revenues of utilities and developers from Renewable Energy Certificates (REC) was
suggested. As there is a low demand, RECs are sold today at the Indian Energy Exchanges at
the floor price level of INR 1.5 for non-solar REC. Floor price for solar REC previously was INR
9.3 and now has reduced. However, vintage multiplier to older projects is available.
Furthermore, only around 10% of offered RECs are being bought. This implies that vendors
have difficulties to sell their RECs despite the existing market mechanism and the existence of
the floor price. Better compliance with Renewable Purchase Obligations (RPO) could increase
the demand for RECs and improve the situation.8
 Implementation of REMC and forecasting (partly demanded to be put on regional level) was
seen as a must.

6
Discussion during the Stakeholder workshop, 22./23.04.2015, Delhi; especially emphasized by representatives
of the wind energy industry.
7
Interviews at SLDCs in various states and the NLDC, discussion during stakeholder workshop, 22./23.04.2015,
Delhi.
8
Discussion during the Stakeholder workshop, 22./23.04.2015, Delhi.

57 | P a g e
  For future RE capacity addition plans it was recommended to evaluate generation, transmission
and distribution costs in order to reach at a realistic overall cost assessment for electricity
generation.

Table 1: Challenges for balancing and integrating RE in India – state perspective

Source: stakeholder interviews

Problems identified by Contributing factors
states level stakeholder
interviews
1. Limited ability to back-  technical limits of thermal units stated to be only 70%
down generation (additional fuel oil needed below)
 retro-fitting to increase limit needs one year during which
the plant is not available for normal power production
 shut-down and start-up (i.e. hot-start) of power plants is
often not being practiced
 central power plants are often not used for balancing
 conventional generation is less expensive than RE
generation
 threat to PLF targets of conventional units
2. Low availability of hydro  correlation of hydro and wind power availability
power for balancing  multi-purpose of hydro plants restricts flexible dispatch (i.e.
irrigation)
 high share of run-of-river plants in many RE-rich States
being not flexibly dispatchable
 low pump-storage capacity, many units out of work or
allocated in states with lower RE shares
3. Low availability of gas-  shortages in fuel availability
fired thermal power  high costs for natural gas
4. Uncertainty of RE supply  absence of forecasting for RE

5. Rate of change of power  fast changes in irradiation or wind speed

contribution from RE  low availability of flexible balancing plants
6. Lack of regional balancing  limited amount of market mechanism to export power
 high transmission fees for wind energy (solar energy is
exempted)
 no involvement of states with low shares of RE regarding
balancing and lacking balancing cooperation between all
states

58 | P a g e
6.1.4 Central perspective
The central perspective on balancing issues is very important. The national grid operator (POSOCO) is
responsible for maintaining load and generation balance in the national grid. However, due to the lack
of system reserves, frequency deviation depends mainly on the level of grid discipline of the states. The
occurrence of their load and generation imbalance is very frequent. The unscheduled interchange and
the variation of frequency deviation have therefore largely been reduced in the past years due to the
implementation of the Availability Based Tariff (now: Deviation Settlement Mechanism). This
mechanism incentivizes that the load and generation imbalance of a state does not exceed 12% or 150
MW of the state’s schedule for inter-state transmission.

The next aim of the national grid operator is to bring the grid frequency even closer to 50 Hz and
introduce primary and secondary frequency control reserves on state level. Today a regulation of CERC
is in place which requires the States to preserve 5% of their power plant capacity on bar as a reserve.
However, compliance and enforcement of this regulation is low. 9
POSOCO and stakeholder from NLDCs stated that the main reasons for imbalances between load and
generation occurring on state level are due to:
 Deviation of actual load from scheduled load (over-, under-drawal, line tripping)
 Deviation of actual generation from scheduled generation (outages, line tripping)

According to POSOCO, changes of generation in renewable power only play a minor role for
imbalances in the system. Correlation of change of demand or conventional generation with frequency
change is much higher compared to change in wind generation. Deviations above 150 MW occur also
throughout the complete year even when wind generation is low. In addition change of wind power and
conventional generation change are poorly correlated which indicates that balancing of RE is not done
actively and accurately with conventional units yet. Based on correlation analysis it was stated by
POSOCO that:
 For wind energy: “…Wind generation variability has negligible adverse effect on deviation from
the schedule …”
 For conventional generation: “…Conventional generation change affects deviation 2 ‐3 times
more than wind generation, though in high wind season, the two are comparable….”
 For load: “…Demand changes affects deviation 8‐9 times more than wind generation, which
drops to 3‐4 times in high wind season…”

POSOCO suggests that load forecasting and generation schedule accuracy should be improved within
the state. Forecasting for RE needs to be implemented in order to actively and accurately balance the
variability. Regulatory support should be given to incentivize flexibility of conventional generation and
compensate generators fully for balancing related costs (partial operation, start-up and stop costs).
Market based solutions for balancing should be further developed.

Regarding the ambitious capacity addition the Ministry of Power and the Ministry for New and
Renewable Energy in India are well aware of the importance to increase balancing capacity all over the
country and especially in RE-rich states. Regional balancing and eventually socialization of effort and/or
cost will be important in the future. Raising the 150 MW limit within the DSM is not considered as a
viable option by the central agencies to relieve the pressure of integrating RE efficiently on the states.

Central level stakeholders also acknowledged the important role of RE forecasting and regarded it as
next milestone to be achieved for proper RE integration.

The major problems identified during the stakeholder interviews on central level with respect to grid
operation and RE integration are summarized in Table 2.

9
Discussion during the Stakeholder workshop, 22./23.04.2015, Delhi; especially referred to by Mr. Sonnee,
POSOCO.

59 | P a g e
 Table 2: Problems in respect to grid operation and challenges of RE integration – central
perspective

Source: stakeholder interviews

Problems identified by Contributing factors
central level stakeholder
interviews
1. Poor generation schedule  Uncompensated outages of generation, line tripping
accuracy of states  Schedule inaccuracy of generators
2. Poor load schedule  Absence of high quality load forecast
accuracy of states  Line tripping, outage of (sub-)transmission equipment
 Difficulties of unmetered and unexpected load
3. Lack of forecasting for RE  Lack of high quality forecasting of RE power generation
 The current regulation requires RE operators for
commercial purposes to schedule their power
production on pooling station level; however schedules
are often not delivered or inaccurate and cannot be
used for system operation
 So far, there is no centralized forecast by either SLDCs,
RLDCs or the NLDC which can be used for system
operation
4. Lack of control reserve  Lack of available power capacity for provision of positive
(secondary and primary) control reserve
 Lack or no practice of regulation to apply control reserve
and lack of related market mechanisms; CERC regulation
says that states have to hold 5% of capacity on bar as
reserve is in place but not being enforced

60 | P a g e
6.1.5 Assessment of existing balancing capacity
The available balancing capacity of an electrical system depends on the actual situation and varies
according to prevailing system conditions. Important parameters considered for this assessment are
the amount of load, the RE penetration level, the available power plant capacity on bar and the available
power plants which can pick-up generation quickly. Availability of storage capacity of hydro power plants
is another important parameter which changes over time. The following analysis therefore can only
assess the flexibility of power plants and the theoretical potential for balancing independent from
specific situations. This is useful to compare the balancing capability between the states but cannot
serve as indicator for each and every situation. For more in-depth investigations, a dynamic assessment
of the balancing capability for a complete year is conducted for a single state in a later stage of the
project.

6.1.6 Flexibility of power plants – literature overview

For flexible generation the technical capabilities of power plants such as minimum load, rate of change
of generation (to follow the load gradient), start-up time and down-time (hot and cold start) as well as
minimum stand still times are important. These parameters influence the quantity of available balancing
power at any given moment of time. For Indian power plants such data is not at all or sparsely available
in public domain. In addition, power plants characteristics vary depending on the manufacturer and the
age of power plants. The relevant parameters might also vary from state to state. These numbers have
been taken from the literature as depicted in Table 3 for the thermal balancing plants (coal and lignite
fired steam turbines, natural gas fired combined cycle power plants and gas turbines) and hydro power
plants. For hydro power plants it is important that storage capacity is available in order to dispatch
power flexibly.
Table 3: Overview of flexibility parameters of power plants to be found in literature
(international practice today / state-of-the-art) 10

Parameter description Power plant type

Hard Combined cycle Gasturbine solo (natural
Lignite
coal (natural gas) gas)
gradient %PN/min 1,5 / 4 1 / 2,5 2/4 8 / 12
at load range of %PN 40 - 90 50 - 90 40 - 90 40 - 90
minimum load %PN 40 / 25 60 / 50 50 / 40 50 / 40
start-up time
hot-start h 3 / 2,5 6/4 1,5 / 1 <0.1
cold-start h 10 / 5 10 / 8 4/3 <0.1
minimum stand still
h 2-4 6 1-2 0
time
minimum operation
h 3-16 3-24 1-8 0
time

6.1.7 Turndown capability in India

One of the most challenging issues in India with regards to integration of RE today is backing-down
conventional generation. Power plants need to run on partial load or to be shut-down completely and
be disconnected from the grid. High minimum load or low turndown capability reduces the balancing
potential. Stakeholders at the SLDC level in various Indian states declared that the minimum load of
most thermal power plants is around 70%. Thus, turndown capability is only 30% which is much lower
than depicted in the literature (German and International Standard, see Table 3). For generating on
lower loads, fuel oil support is needed which is usually avoided due to higher costs [CEA 2013a].
Restriction to minimum load is also connected to economic implications and is not only a technical
challenge.

10
Source: VDE 2012

61 | P a g e
 Level of achievable minimum load by thermal plants is a fresh topic of debate among different Indian
stakeholders. In Maharashtra various technical parameters for all power plants within the state have
been analysed. The result showed that minimum load of power plants is around 70%.11 Enhancing the
flexibility parameters (i.e. by retro-fits) towards state-of-the-art values or beyond is recommended to
increase the balancing capability. Beside the technical capability of plants, lower minimum loads in India
compared to international standards could also originate from other factors such as lower quality of coal
(i.e. high ash content). Nonetheless, testing of flexibility parameters of power plants in all RE-rich states
could yield a reliable data set for technical plant parameters.
For the case of Gujarat detailed plant specific data is available and depicted in Figure 25.
Figure 25: Minimum load of power plants in Gujarat

Source: Fraunhofer IWES, data: Gujarat SLDC

80%
Ut ran-I
75%
GIPCL-I Wanakbori
70%
minimum (% -Pn)

Dhuvaran
65% GIPCL-II Sikka Essar GPEC Ghandiganar Ukai
TP-Aeco
60% EPGL
Panandhro
55%
GSEG Ut ran-II
50%
SLPP
45%

40% Akrimot a
0 200 400 600 800 1000 1200 1400
plant size (M W)

CC - Nat ural Gas Coal Lignit e

Most values for minimum load in Gujarat are below the value of 70%. For some units minimum load is
as low as 40%. The capacity weighted average is around 62%. Information about ramp rates, hot-start
and cold-start times, minimum stand still or operation times of power plants in India has not been found
© Fraunhof er IWES

in public domain. Some stakeholders at SLDC level state that bringing up power plants on bar again
might take up to around 48 hours.12 Shut-down of plants is therefore often avoided.13

6.1.8 Theoretical thermal balancing and ramping potential

The first step of assessing the balancing potential of the respective states is to estimate the theoretical
balancing potential of thermal generation units for each state. The assessment includes all central and
state power plants. Regarding central power plants the allocated state share has been considered for
any state wise assessment. The theoretical balancing potential is defined as the total installed thermal
capacity multiplied by one minus an assumed minimum load. This is 65% for coal power plants and
75% for Combined Cycle power plants. The theoretical balancing potential indicates the maximum
potential of reducing or increasing the actual generation (no shut-down or start-up of plants, all power
plants on bar). In
Figure 26 (left) this value is compared to the installed capacity of RE (wind, solar and mini-hydro).

11
Discussion during the Stakeholder workshop, 22./23.04.2015, Delhi; especially explanation of Karnataka SLDC
staff.
12
Interview with staff at the SLDC in Gujarat, 19.02.2015.
13
Also mentioned during interview with staff at the SLDC in Andrah Pradesh, 25.02.2015.

62 | P a g e
In respect to actual installed RE capacity

Figure 26: Total theoretical balancing potential for each state and comparison to installed RE
capacity

Source: Fraunhofer IWES, data: CEA 2015

10.000 3,00

Balancing Potential /
8.000 2,50
6.000
MW

RE capacity
2,00
4.000
2.000 1,50
- 1,00
0,50
0,00

Thermal Balancing Pot ent ial

Tot al Renew able Capacit y (M NRE)

In 5

Figure 26 (right) the ratio of balancing potential and the RE capacity is displayed. States with low ratios
© Fraunhof er IWES

have comparably low potential and thus face more difficulties in balancing RE generation if no additional
balancing power – i.e. from hydro – is available. Tamil Nadu with a ratio of below 0.5 has the lowest
balancing capacity of all analysed states. In this case, for each MW of RE capacity only 1 MW of thermal
balancing capacity is available (regular shut-down of power plants for maintenance etc. not yet
considered).
Due to the fact that additional balancing capacity can be activated by starting or shutting down power
plants, this value can only be regarded as an indicator for comparison of general thermal balancing
capability. If uncertainty in power dispatch from RE as well as uncertainty of load and scheduled
generation is reduced, power plants can be more effectively used for balancing by shutting down
completely and starting on demand.
In general, it has to be noticed that the regional balancing potential within the regions (Northern,
Southern and Western Region) and all over India is very large. This potential is shown in Figure 27 and
should be tapped in order to enhance integration of RE (for aspects of regional balancing see also
section 2.3).

Theoretical balancing
Figure 27: Theoretical potential
thermal balancing of thermal
potential plants
in RE rich-states compared to the potential
of regions and all India

60.000
Theoretical Balancing Potential (MW)

50.000

40.000

30.000

20.000

10.000

Thermal Balancing Pot ent ial w it hout cent ral unit s w it hout gas and cent ral unit s

63 | P a g e 3

© Fraunhof er IWES
The thermal ramping potential depends also on the thermal balancing potential. It is defined as the
potential to increase or decrease the generation within a certain time frame. It can only be assessed if
the plant specific ramp rates are known and it depends for any moment of time on the amount of
capacity on bar and the absolute potential to decrease or increase the generation. It is assumed that
ramp rates of Indian power plants are 2% and 1.5% of nominal capacity per minute for gas and coal
power plants respectively according to values on international practice based on literature review (see
Table 3). This means that a gas power plant is able to cycle from 100% of rated power to a technical
limit of 75% in 15 minutes and vice versa. Coal power plants accordingly need 23 minutes to cycle down
to 65% which is the technical limit assumed above for Indian conditions.

Figure 28 (left figure) demonstrates the theoretical ramping potential for a 15-min interval for all states.
This theoretical ramping potential is defined for the case that all thermal power plants are on bar and
are able to cycle between 70% and 100% of rated capacity. For Gujarat this potential is around 5,000
MW. For comparison, the demand for ramping in Gujarat derived from the maximum change of the
residual load in 2014 is around 380 MW per 15 min. The load and residual load (load minus feed-in
from RE) data for this analysis has been received for the year 2014 in hourly resolution from the SLDC
in Gujarat. The frequency distributions of load and residual load gradients show that the rate of change
is rarely higher than +800 or -800 MW per hour. In addition both frequency distributions are very similar.
The maximum positive gradient of the load is 1,154 MW per hour compared to 1,258 MW per hour in
the case of the residual load. The highest negative gradient of the load is -1,564 MW per hour compared
to -1,530 MW per hour in the case of the residual load. For the actual penetration level of RE in Gujarat,
this means that from the point of providing ramping power load coverage is a similar difficult task with
or without RE. But this is only valid provided that the RE production is forecasted with high quality.

Figure 28: Theoretical state wise ramping potential of all thermal power plants (left) and
ramping demand in Gujarat (right)

6000 160
Number of occurences

140
5000
M W pro 15-min

120
4000 100
3000 80
Coal 60
2000
40
Gas
1000 20
max change in RL
0 0
- -800 -600 -400 -200 0 200 400 600 800 1000
1000
Class (below X MW)

Load Residual Load

To which extent the power plants can activate their ramping potential and in which direction (ramp up,
ramp down) depends on their actual operating condition (actual power in percentage of rated capacity).
In practice the ramping potential is also influenced by unavailability of plants, fuel availability (especially
in regards to natural gas) or other factors limiting the ramping potential.

13
64 | P a g e
© Fraunhofer IWES
6.1.9 Theoretical hydro balancing potential
The hydro power balancing potential of the six RE-rich Indian states is depicted in
Existing Hydro Capacity with storage
Figure 29.

Figure 29: Installed capacity

 Karnataka has the large storage capacities and succesfully
Source: CEA, important reservoirs and associated hydroelectric stations Source
 Tamil Nadu has despite large hydro capacity only low storage reservoirs only imp
hydroele
complet
10000 3500

Storage duration (hours)

capacity (MW or MU)

3000
8000
2500
6000 2000
installed Capacity (MW)
4000 1500
1000 storage capacity (maximum energy
2000 at FRL, MU)
500
0 0 storage duration at FRL (at rated
power)

27/04/2015 Note: FRL = Full reservoir level Page 23

Only important reservoirs with associated hydroelectric power plants are included as these can be
effectively used for balancing. The storage capacity of all dams and the installed capacity for all
associated electricity generation have been aggregated for each state. It is visible that, despite similar
installed capacity storage, sizes of reservoirs are very different. Karnataka has by far the largest
reservoirs being able to theoretically dispatch power at rated capacity for 3000 hours at full reservoir
levels (FRL, number based on average for all plants given aggregated storage and plant capacity).
Especially Tamil Nadu has a very limited storage capacity in relation to its power capacity which even
theoretically would allow only for dispatching power at FRL for around 350 hours. The ability to dispatch
hydro power and use the plants for balancing highly depends on water inflow, storage level and irrigation
demand.

6.1.10 Conclusion
In all states except in Tamil Nadu theoretical balancing capacity is sufficient to integrate the current
amount of RE. Residual load following by conventional generation should currently be feasible due to
the ramping capabilities of conventional generation and hydro power. However, different shortfalls and
practical problems limit the efficient use of the existing balancing capacity. These are fuel supply
shortage, not using conventional power plants for balancing, low technical standards in terms of plant
parameters and especially lack of forecasting of RE and uncertainty in system operation.

However, given increasing penetration level of RE, integration will become more difficult. The Indian
government is planning for 45 GW of wind power and 37 GW of solar power within Himachal Pradesh,
Rajasthan, Gujarat, Andhra Pradesh, Karnataka and Tamil Nadu until 2022. Referring to the plans of
installing 100 GW of solar and 60 GW of wind power in India, these six states will be responsible to
provide 75% of total wind power and 37% of total solar power. Even if the conventional capacity
increases, balancing capability will decrease in relative terms within the single states. This is shown
indicatively in

Figure 30.

65 | P a g e
 Figure 30: Theoretical thermal balancing capability of RE-rich states today and up to 2022

Assumptions: 30% of conventional capacity addition in 2022 compared to level of today; tentative
state wise break-up of RE capacity addition until 2022 according to RE policy of Government of
India.

Source: Fraunhofer IWES, data: MNRE

Accordingly, measures have to be taken to foster RE integration. Some of the possible actions identified
are explained in the next section and are categorized into short-term, medium-term and long-term
actions.

Enhancing Balancing

6.1.11 Short-term solutions

Improving load forecasting:

Load forecasting is essential in order to maintain load and generation balance. Data analysis of actual
and scheduled load of Indian distribution companies suggest that a typical absolute error of day-ahead
load forecasting (yearly average of absolute difference of 15-min-intervals between scheduled and
actual load) is around 5.5% - 7% of schedule. For Germany and Europe an absolute error of around
2% is commonly stated in relevant literature. During the on-site investigations in India and during data
analysis it was noticed that SLDCs focus in respect to quality analysis of their load forecasting especially
on the “simple averages” of forecasting errors (average of negative and positive deviations) which lead
to an underestimation of the impact of the load forecast error. The “simple average error” is around
3.3% – 5.2%. Both error types for the data of three analysed DISCOMs are shown in Figure 31 in form
of yearly averages for each quarter of an hour of the day. It is clearly visible that the absolute average
error lies above the “simple average”.

66 | P a g e
Figure 31: Forecasting quality of three selected DISCOMs in 2014 14
12% 12%
“simple hourly average" of error

8% 8%

absolute average error

4% 4%
DISCOM 1
0% 0%
DISCOM 2
24 48 72 24 48 72
-4% -4% DISCOM 3

-8% -8%

-12% -12%
quarter-hour of the day quarter-hour of the day

The data also indicates that systematic errors are occurring on a regular base depending for example
on the hour of the day. Especially during peak time DISCOMs tend to underestimate the actual load
significantly (i.e. between quarter-hour 70 to quarter-hour 90 in the figure, left, which is from 17:30 p.m.
to 22:30 p.m.). The load forecast should therefore be improved and corrected by these and other
systematic errors.

In comparison to deviation of RE generation, differences between actual and scheduled load are quite 12
significant. For example: Given a typical load magnitude for Indian States of 10,000 MW and a forecast
© Fraunhof er IWES
error of 10%, over-drawl is 1,000 MW. On the contrary, given a typical wind generation of 2,000 MW in
high wind season and a wind forecasting error of 30% with respect to the actual generation, the impact
on imbalance is in the order of 600 MW.

Therefore proper load scheduling needs to be incentivized. The transmission company of Gujarat has
recently addressed the issue of over-drawl by DISCOMs and suggested an Automatic Demand
Management Scheme (Shah 2015) which is supposed to cut-off feeders from electricity supply if over-
drawl of DISCOMs is above a certain threshold. Although this idea will affect quality of supply and lead
to shortages of electricity, it puts the DISCOMS under pressure to fulfil their load forecasting
responsibility which is necessary. However, over-drawl could simply be avoided by providing a more
accurate load schedule.

Enhance system operation and plant flexibility by implementing RE forecasting:

A good RE forecast enables scheduling of power plants day-ahead and intra-day. It therefore has a key
role in balancing and decreases uncertainty in the power plant dispatch. Conventional power plants can
be shut-down completely in case of high feed-in from RE. Today, shut-down of plants is often avoided
due to uncertainty about load development and upcoming RE supply.

However, coal and combined cycle gas power plants are technically able to perform a hot-start within
1.5 to 3 hours’ time if they have not been shut-down for longer than 8 hours. A cold-start may be required
if a power plant is shut-down for a longer time period. For example, if coal power plants are off-line for
more than 48 hours, start-up time may reach 4 – 10 hours depending on plant type and technology (see
Table 4). These situations may occur e.g. due to lower demand during week-ends. The figures in Table
4 are literature values and are indicative only. Actual start-up and shut-down and the relation with stand
still times needs to be more deeply investigated under Indian conditions and should be assessed by
each state for all of its power plants.

14
left: simple yearly average per quarter-hour of the day, right: absolute yearly average error per quarter-hour of the day
(error: scheduled load – actual load)

67 | P a g e
 Table 4: Starting capabilities of power plants

Source: VDE
Power plant type
Hard coal Lignite Combined Gas turbine
cycle (natural single cycle
gas) (natural gas)
hot-start (stand still < 8 h 3 / 2,5 6/4 1.5 / 1 <0.1
hours)
cold-start (stand still < 48 h 10 / 5 10 / 8 4/3 <0.1
hours)

Normal practice / state-of-the-art

Shut-down of power plants thus enables the grid operator to decrease power not only to the common
technical limit of 70% but down to 0% being able to integrate higher penetration from RE. Generally it
can also be cost-efficient to reduce power output of a single plant instead of running several plants on
part-load efficiency. RE forecasting is a necessary condition to improve certainty in system operation
and power plant dispatch and should therefore be introduced.

In Germany forecasting is not done based on single generators or pooling stations. Many RE Power
plants without time-based metering in 15-min intervals are forecasted and scheduled by the TSOs for
the complete control zone (there are four control zones in Germany). For these plants the TSO is also
responsible to sell the power to the short-term market. Since 2012 RE power plants which have time-
based metering can be directly brought to these markets by any market participant. Since August 2014
this direct marketing is obligatory for new RE power plants larger than 500kW. From 2016 onwards this
threshold will be lowered to 100kW. The common habit for marketing is that many RE power plants are
pooled to one large portfolio in order to increase forecasting accuracy and decrease marketing costs
per energy unit. A forecast is done for all these plants within the portfolio by the direct marketing
participant.

Portfolios typically have large sizes ranging from one GW up to around 10 GW portfolios. The direct
marketing participant has to ensure that all generation is accounted for in so called balancing groups.
The rules for balancing groups enforce that schedules (thus forecasts in the case of RE) are submitted
by the direct marketing participants towards the balancing group manager and finally to the TSO.
Deviations from these schedules are penalized within the Deviation Settlement Mechanism in the
German system.

Improve balancing possibilities with central thermal power plants:

States can balance their RE generation also with central thermal power plants. Central stations are not
owned by the state utilities and are usually shared by different states (beneficiaries). Thus, every state
has only the entitlement to use a certain percentage of this capacity. To effectively use this share for
balancing, it needs to be assured that the allocation of capacity for a specific state is not withdrawn from
it, even if utilization of its share is low. If the available capacity is not needed at any moment of time by
the RE-rich state it should be allocated only temporally to other beneficiaries. This temporally re-
allocation can be organized by the RLDC.

If central power plants are used frequently for balancing and re-allocation of capacity is not possible,
the plant load factor of the plant decreases. Additionally, if the power plant runs in part load operation
the generation costs are higher compared to full load operation. These impacts should not lead to
negative economic consequences; instead balancing with central power plants should be encouraged.
The fixed capacity charges to be paid by the states for the usage of central plants could be socialized
for times of non-usage in order to incentivize balancing with central power plants. Today balancing with

68 | P a g e
central power plants is practiced in some of the states during the monsoon period. The allocation of
central power is then used during peak hours only and is varied between 0 - 100% of allocated capacity
according to the actual requirements of the system.15

New flexible conventional generation and definition of minimum flexibility

requirements
Due to the growing electricity demand in India, conventional capacity addition is necessary to enhance
quality of supply and meet peak load demand. RE generation does not contribute significantly to firm
power availability due to uncertainty of weather conditions. During the current and next five year plan
large conventional capacity addition is foreseen. In order to enhance balancing capacity, future power
plants should provide sufficient flexibility. It has to be studied in more depth, which composition of power
plants is technically and economically efficient and most suitable for the Indian case. The overall mix of
capacity addition will be a compromise of minimizing costs, fuel availability in the area and other aspects
of the overall supply strategy of the government. Considering the fact that power plants run for 40 – 50
years, it is crucial to already now assure that conventional generation built today is capable of
integrating RE production. This can be achieved by:

 Including a significant share of flexible generation capacity in addition plans

 Define a revised set of standard requirements for all new power plants regarding flexibility (low
minimum load, high cycling capability, resistance against numerous start and stops during the
lifetime, fast start-up and shut-down ability). These minimum flexibility requirements should at
least match state-of-the-art criteria of existing technology and should exceed existing technical
standards recommended by the CEA (i.e. CEA 2013)

Retrofitting of existing power plants:

Retrofitting is one option to increase the flexibility parameters of existing power plants. In India
increasing the flexibility of especially coal power plants will be relevant. Best practice examples show
that start-up and shut-down even twice a day is possible in order to run only during peak demand times.
Load following and running at minimum generation level between full load and around 25% of rated
capacity are possible with coal fired power plants. Test experience also shows that even cycling down
to 60 GW for a 480 GW coal power plant is possible with additional gas firing for up to 6 hours (Cochran
et al. 2013) although the plant was initially designed as base load power plant with only a few cold starts
per year. Until now, each coal unit of the plant has reached an average number of 1,760 starts during
its lifetime (523 cold-starts, 422 warm-starts and 814 hot-starts).

However, retrofitting needs an initial investment in technical improvements. Cycling and running coal
power plants at technical limits also increase the impact on equipment and thus, the costs of operation
and maintenance. Retrofit solutions from manufacturer usually address the following options:
 Lower turndown (minimum load)
 Faster ramping
 Faster and less expensive starts
 Keep emissions low despite increase in flexibility

The achievable reductions of retrofitting options according to (Venkataraman et al. 2013) are depicted
in Table 5

15
Stakeholder interview at Karnataka SLDC on 04.03.2015.

69 | P a g e
 Table 5: Improvement potential

Source: NREL 2013

GT CC Coal
Start-up or shut down time Improvement Up to 50% Up to 60% ~30-50%
Start-up fuel cost Improvement ~50-60% ~Up to 30% ~30%-50%
Ramp rate Improvement ~100% ~100% ~30%
Turn down improvement ~5-10% ~5-10% ~30%-50%
(minimum load)

Increase of gas availability or higher gas allocation to RE-rich states:

Domestic natural gas production is diminishing in India ranging from 131 million m³/day in 2009-2010
to 97 million m³/day in 2013-2014. This decrease is largely due to the reduction in offshore production
to the extent of 32%. The offshore share makes up for 74% to 81% of total domestic gas supply in the
past 5 years (seeFigure 32). The domestic resources account for two thirds of the overall gas supply in
India. One third is imported re-gassified Liquefied Natural Gas (R-LNG).
The lack of gas leads to severe shortage in the electricity sector where roughly one quarter of the fuel
is consumed. As gas power plants work with less CO2-emissions and are able to contribute very well
to balancing of renewable energies due to their high flexibility (especially start-up and down-time) it is
recommended to increase the gas availability for RE-rich states. On the one hand, this can be done by
increasing the general availability of natural gas. On the other hand it is possible to increase the gas
allocation of RE-rich states compared to other states. This suggestion was made by various
stakeholders at SLDC level and is also discussed in recent literature of Gujarat Energy Transmission
Corporation (Mehta/Nayak 2015).
Figure 32: Overview of natural gas sector in India16

Source: Ministry of Petroleum and Natural Gas

100
Ot hers Trend of natural gas production
90
250%
80
Production in MMSCMD

Pet rochemicals
70 200%
Ref ineries
60 Gujarat 150%
Domest ic & CNG… 50 Andrah Pradesh
100%
40 Tamil Nadu
Pow er
30 50%
Rajast han
Fert ilizers 20
Ot hers 0%
10
0,0 20,0 40,0 60,0
0 Tot al Of f shore
Use in MMSCMD
2014 (unt il
Dec)
Tot al R-LNG Domest ic

A drawback of importing gas is that it is more expensive than the use of domestic coal. Increased usage 9

of imported gas will increase prices of electricity generation. A compromise needs to be made here
© Fraunhof er IWES

taking into consideration the overall development of India. So far the Indian government has already
adopted strategies to improve the situation regarding fuel import, infrastructure and domestic supply.
These should be regularly updated and enforced.

6.1.12 Mid-term solutions

Usage of hydro power stations with storage for intra-day balancing during monsoon
period
In India many hydro power plants with storage cannot be used without restrictions for balancing as a
certain amount of water needs to be available for irrigation. The water level in the reservoir also may

16
sector-wise use and production in 2014 (in million metric standard cubic meter per day) and trend of production for different
states (onshore) and total offshore production

70 | P a g e
need to be maintained for a long time period in order to meet irrigation demand. Given high precipitation
intensity during monsoon season, plants partly become “must-run” to avoid spilling of basins.

However, power plants can still be used for balancing if the time of production is not completely
determined by irrigation department of governments. Today hydro plants with storage options through
basins are partly used to cover peak load within the states dispatching more power during peak times
than in average. Thus, they are dispatched very flexible and can also be used to balance fluctuations
of RE. Hydro power also has a very high inter-annual150.000variability and power production heavily depends
on precipitation during monsoon season. Exemplarily the reservoir level and inflow of the Mettur Dam
120.000

is depicted in figure 33 (left) for the year 2013. The reservoir

90.000 level increases drastically during monsoon
until maximum capacity. Before this point of time there is no power production, afterwards water stwhich
60.000
orage capacit y

st reservoir volume
cannot be stored needs to be turbined or spilled especially around 1 of August when inflow is peaking.
30.000 inf low
From 4th to 11th of August Mettur Tunnel was generating 24 hours on rated capacity being “must-run” t o st orage
0
(compare peak of power in right graph). However, after that week average daily production is always
Discharge

below the rated capacity and power during lighting peak is on the same level. This is indicating that the
plant is running 24 hours on the same level of production, but below rated power.
Figure 33: Storage capacity17

250
120.000
200
90.000 Power (MW)
M .cft

150
60.000
100
30.000
50
0
0

st orage capacit y reservoir volume inf low

Average daily pow er - met t ur t unnel
Pow er during light ing peak

However, resource-wise it would be possible to cycle the plant according to the power demand or 7

change of feed-in
© Fraunhof er IWES from RE. For irrigation purposes the same amount of water could be released during

the day to ensure water availability.

Therefore, it is recommended that each state transmission company develops a concept of using also
the irrigation based hydro power plants for balancing - at least for balancing within the day. This has to
be done in close coordination with the responsible department for irrigation at the state governments.

Increasing balancing area

At the moment balancing of renewables is left to a large extent to the capabilities of states producing
the energy from RE. This is ineffective as the complete available balancing capacity of a region cannot
be activated, if necessary. For example, in case of high feed-in from RE a single state has less power
capacity available to back-down. In Tamil Nadu it regularly happens that the system operator is not able
to back-down sufficient conventional generation which contributes to situations of over-frequencies and
eventually to curtailment of RE.18 The thermal balancing capability on the regional level is higher
compared to the states’ capabilities due to lower installed RE capacity in relation to the existing thermal
capacity. For the comparison of different thermal balancing capacities of regions and states see Figure
34.

17
reservoir volume and inflow of Mettur Reservoir in Tamil Nadu (left) and power production of Mettur Hydro Power Plants (tunnel)
during peak time of the day and daily average (right)
18
Stakeholder interview at Tamil Nadu SLDC on 16.02.2015.

71 | P a g e
Theoretical balancing potential of thermal plants
Figure
In 34: Thermal
respect Balancing
to actual Potential – Comparison
installed RE capacitybetween states, regions and India

3,00
Balancing Potential /

2,50 +31% / +5%

+133% / +92%
RE capacity

2,00
1,50
+79% / +281%
1,00
0,50
0,00

Thermal Balancing Pot ent ial w it hout gas unit s w it hout gas and cent ral unit s

x% / y% = Larger regional / nat ional pot ent ial compared t o st at e pot ent ial

The ratio of balancing potential and RE capacity is for example 2.09 for the complete Northern Region
5

compared to only 0.9 in Rajasthan itself. This means that the available thermal capacity for balancing
© Fraunhof er IWES

is 133% higher on regional level than on state level. In Gujarat this value is 31% compared to Western
Region and in Tamil Nadu 79% compared to Southern Region. The lower increase compared to
Rajasthan is due to the fact that in these regions more RE-rich states are located. Especially in the
Southern Region the overall thermal balancing potential relative to RE-capacity is lower than in other
regions.

Also the hydro power plants with water reservoirs can be used more effectively if balancing is done on
a regional level. The use of pump storage hydro power stations, which is very limited in India, could be
used for balancing over state borders.

Increasing the balancing area is also an important issue in the European interconnected power system.
Although dispatch of power plants and thus, balancing of RE generation is coordinated via the European
Power Exchange and bilateral cross-border trade, it is restricted to tie-line capacity. In order to balance
renewable energies more effectively European Countries plan to further increase the transmission
capacity between different countries. Compared to the EU, India already today has the advantage of
strong inter-state connections due to introducing central power production in the last decades. Gujarat
for example has nine 400 kV lines (4 times 2x400kV, one single line 400 kV) and one 2x200kV line
which make up for around 11 GW of transmission capacity given a peak load of 14 GW. In comparison,
Germany has a peak load of around 80 GW and plans for transmission capacity of 25 GW to
neighbouring countries in 2025.
Regional balancing can be enhanced by:
 creating a solution based on a complete new regulatory design
 by increasing possibilities for inter-state trading and gradually adjusting present regulation and
market designs

Suggestions for possible designs and amendments of regulation are to be outlined in working package
3 of this project.

Enhancing inter-state power exchange

Hand in hand with developments in market design changes for regional balancing, inter-state power
exchange needs to be enhanced and facilitated. In this regard the following measures should be taken:
 Reducing inter-state transmission fees for wind energy or decouple fees from installed capacity
of wind farms
 Strengthening bilateral trade and trade via the Power Exchanges
 Introducing new perspectives on inter-state power exchange beside tight inter-state flows
within the Deviation Settlement Mechanism (Batra 2015)

72 | P a g e
  Introduce new perspectives on inter-state Open Access power trade19
 Stabilizing the market and revenues from trading with Renewable Energy Certificates
 Enforcing the Renewable Energy Purchase Obligations to foster a liquid market for renewable
energies

Additionally more RE power plants could be connected directly to the inter-state grid as central power
plants. These plants then would fall under jurisdiction of central authorities and dispatching
responsibility would be under the rule of RLDC which would enable regional balancing.

Pump storage hydro power stations

Pumped hydro storage is needed for increasing the balancing capacity. Pumped storage hydro power
stations are the most flexible and fast reacting balancing option and furthermore most cost competitive
in a system with new balancing infrastructure and technology. Besides balancing RE, it can be used to
economically cover peak loads and enhance security of supply. The report “Large Scale Grid
integration” of the Central Electricity Authority additionally states the importance for system services
and security as follows: “…The other advantages of pumped storage development are availability of
large reactive capacity for regulation, availability of spinning reserve at almost no cost to the system
regulating frequency to meet sudden load changes in the network….”.
Today only 2.7% of the total pumped hydro storage potential is realized and working in pump mode
operation. This represents a cumulated capacity of 2,600 MW. Another 2,185 MW is installed but due
to various reasons not operating in pump mode (technical problems, tail pool dam still under
construction or not yet constructed). One new project of 500 MW is under way in Tamil Nadu and 2,100
MW under investigation in West Bengal and Maharashtra.

Figure 35: Potential and installed capacity of pump hydro storage in India

2.7% of
potential
30.000 realized
today
14.000
Ut t arakhand
East ern

12.000
Potential (M W)

10.000
West Bengal inst alled, but not w orking in
8.000 pump mode operat ion
6.000 Tamil Nadu
Sout hern

inst alled and working in

4.000 pump mode operat ion

2.000 Andrah Pradesh under const ruct ion

0
planned
Kerala

Orissa
M aharashtra

Karnat aka

West Bengal

Assam
M izoram
M anipur
Rajast han

Andhra Pradesh

Tamil Nadu
Bihar
Himachal Pradesh
Ut t ar Pradesh

M adhya Pradesh
Jammu & Kashmir

Gujarat

M aharashtra
West ern

survey & Invest igat ion

Gujarat

0 500 1.000 1.500 2.000

Nort hern West ern Sout hern East ern N. East ern
MW

Source: PCGIL 2012

10

The total remaining

© Fraunhof er IWES potential (assessed by CEA during 1978-87) is located in 63 different sites making
up for 96,500 MW which is a tremendous potential compared to the actual system size. It is
recommended for Indian central or state level stakeholders to:
 Draft an updated actual potential study reassuring feasibility and taking into account ecological
and social concerns.
 Identify concrete projects to be realized until 2022 and necessary financing schemes. Power
plants could be central level power plants and funded by the central level in order to enhance
balancing capacity in RE-rich regions.

19
Discussion during the Stakeholder workshop, 22./23.04.2015, Delhi.

73 | P a g e
 Demand Side Management and Demand Response
Demand Side Management (DSM) can be used to shift power consumption to times when availability
of electricity (from RE) is high. In contrary to load shedding, customers could voluntarily agree to shift
their load during the day. The incentive to participate in such DSM schemes can be set by the tariff
structure. Energy-intensive industry is most likely to be able to participate in demand side activity as
they already put emphasis on cheap procurement of power and consume power in relevant order (i.e.
aluminium industry, chemical products). In Germany, load management in cooling houses is already
done today as it can be easily integrated in normal operation of the cooling houses (i.e. lower cooling
load during night, higher cooling load during the day). In general, realization of demand side
management potential in India should be assessed systematically. One possible application for
Demand Side Management could for example be the provision of secured electricity to water pumps
which already today run under load sharing regimes. The supply could be controlled by the respective
SLDC and provided for a certain amount of hours per day.

Demand response is usually referred to as influencing consumer’s consumption patterns by introducing

time-of-the-day tariffs. This can be especially interesting in states where contribution from solar
generation will be high in the future as it is more reliably available during the same time of the day in
India than wind energy.

Increasing transmission capacity between regions

Increasing the transmission capacity is one of the most important measures to improve regional,
national or common international balancing activities. As the inter-state tie-lines in India are already
very well established, focus needs to be put on inter-regional capacity. Different activities and studies
have assessed this demand and large transmission capacity addition is already planned and under way
within the development of the Green Energy Corridors of India. The planned transmission capacity in
India until the year 2017 is depicted in Figure 36 [PGCIL 2012].
Figure 36: Increase of transmission capacity under the Green Energy Corridor Project Plans

Source: PCGIL 2012

74 | P a g e
6.1.13 Long-term solutions

Diversification of supply from wind energy

The geographical distribution of wind power can help to lower the integration effects of wind and solar
energy due to spatial smoothing effects. The wider the geographical area the larger the smoothing
effects are. Total power production is thus less volatile if RE-capacity is spread over a large area due
to different local weather conditions. Figure 37 demonstrates this smoothing effect for three different
locations in Germany. The left graph shows the three different locations on a map of Germany. The
right graph shows that the power production of the wind turbines (normalized to rated capacity) in
locations P1 and P2 are very closely correlated. P3 which is the more distant location shows lower
correlation. The normalised sum of all three locations shows a smoother power production than the
individual locations.

It should be studied if due to these smoothing effects, it is more cost effective to distribute wind energy
generation over a larger area especially because further capacity addition of RE is needed also after
the year 2022. For solar power production a broader spatial distribution is already planned as the yield
expectation is within a similar range all over India. For wind energy it has to be taken into account that
energy yield in other regions would be significantly lower.
Figure 37: Smoothing effects of wind energy supply from RE due to geographical
diversification

0.9 P1 (Pixel 50)

P2 (Pixel 115)
0.8 P3 (Pixel 150)
normalised sum
0.7
normalised power output

0.6

0.5

0.4

0.3

0.2

0.1

0
650 700 750 800 850
hour of the year

Source: Fraunhofer IWES

Deployment of further storage options

There are several further options to increase the balancing capability. However, given the actual RE
penetration these storage or sector-coupling options are of low relevance in India as integration of RE
can be managed by conventional power plants if regional balancing is enhanced. In the long-term they
can be more relevant. At which penetration level they should be deployed needs more investigation and
research. In other countries, some technologies are in the development stage (i.e. power-to-gas, use
of electric vehicle).

75 | P a g e
Some further storage options including sector-coupling options are:
 employment of battery storage
 use of electric vehicle for balancing (controlled loading schemes, grid injection of electricity from
battery during non-usage of vehicle)
 sector-coupling (heat and electricity)
 power-to-gas (generation of renewable hydrogen or methane, storage and re-usage in
electricity sector or in the transportation sector)

These options are only relevant on the large scale given very high energy shares of RE in the electricity
mix (for the German case estimates range from 50% – 100% RE share of total generation of electrical
energy depending on the technology [IWES/IAEW 2014, IWES 2014]). Especially power-to-gas is only
relevant if a gas grid is existing or gas storage is available and for energy scenarios with RE share of
80% or higher (i.e. due to large efficiency losses). As these technologies are quite expensive today,
cheaper options to increase the balancing capacity need to be used in the first place. However, from a
very long-term perspective these technologies might play an important role also in India and are
therefore described below. Especially batteries can be also used to increase power quality or deliver
system services (control reserve, voltage control).

Most of the storage options available, especially battery storage, are only able to bridge short-term
differences in supply and demand due to limited storage capacity. An overview of all storage options
and comparison in respect to storage duration and storage capacity is given in Figure 38. For seasonal
long-term storage the power-to-gas technology in combination with storage of gas in caverns or the gas
grid can be used. The system adequacy of the storage options needs to be elaborated and assessed
by analysing the characteristics of the imbalance between RE and demand – short- and long-term.

In India a lot of storage is already available through on-grid called inverters and today serves as security
against power cuts. Smart technical and commercial solutions need to be developed to make use of
these devices. Further storage options are thermal storages - like cold water used for chilling the next
days in central cooled houses. Peak electricity pricing would help to support such systems.

Figure 38: Overview of storage options and their typical storage capacity and possible cycle
duration

jhkkghjhj
1 year

Heat storage
1 mont h
1 w eek
jhkk
Deloading duration

1 day
jhkk
jhkk
jhkk
(cycle)

Pump hydro storage

1 Battery
Pressurized air
hour
storage
Condensators chemical
thermal
1 min mechanical
Rotational mass
storage Electro-chemical
Electro-magentic
electrical

1 sec

1 ms Windings jhkkghjhj

storage capacity
Source: Sterner/Stadler 2014
4

© Fraunhof er IWES

76 | P a g e
Table 6: Overview of storage options for scenarios with high shares of RE
Source: Fraunhofer IWES
Balancing Short Description Applications Relevant given RE-share in
option energy mix of*:
Storage option for a short-time Short-term storage, fast For energy balancing:
horizon (hours to a day), low balancing of RE, 60% - 80% RE share of total
energy density, different control reserve and grid generation
technologies available (lithium- stabilization (ancillary For ancillary services:
battery storage ion, natrium-sulfuric, lead services) Earlier if required, situation
battery), electrical storage and application dependent
Driver: market development
of electric vehicles, electrical
storage demand
Intelligent charging of electric Short-term storage, Controlled charging:
vehicles (EV) allows energy large amounts of units 50% - 80% RE
consumption if RE-supply is necessary to create Feed-in from storage:
high, feed-in from storage of EV relevant storage size, 70% ++ RE
to grid is possible; however it balancing of renewable
electric vehicle
requires a strategy and energies (use of RE- Driver: market development
technology for charging/ excess power and of electric vehicles, RE
discharging schemes supply in times of non- development (i.e. excess
(communication, charging availability of RE) power)
infrastructure)
Sector-coupling technology, no provision of energy for 50%++ RE
electrical storage option, but heating/ cooling and air use of excess RE power
conversion to other energy form, conditioning purposes ecologically preferable
electrical heating and cooling (households or (especially for electric
can be used to create a flexible industry); electrically heating/ cooling); use of
sector-coupling demand for electricity fitting to driven during times of conventional production
(heating or RE availability; technologies high availability of RE ecologically counter-
cooling & available: electric boilers, heat productive
electricity) pumps, cold/ ice store. Use of
power from conventional station Driver: RE development (i.e.
is ecological inefficient; excess excess power), CHP
RE power or energy mix with generation and district
high shares of RE ecologically heating
required
Technology converts electricity long-term (seasonal) 80%++ RE
to hydrogen (gas) via storage option in the only excess power required
electrolysis; refinement to long run given very high
methane and storage in gas RE shares Driver: RE development (i.e.
grids possible; due to the excess power)
different conversion processes
power-to-gas
efficiency is low (40-60%),
sector-coupling technology, no
electrical storage option but re-
electrification of gas by
conventional units possible (low
overall efficiency)
*Rough indication based on studies and future system analysis in Germany.

77 | P a g e
7 Cost analysis of balancing options
The comparison of costs of different balancing options is challenging. The costs depend on the
technologies used and largely on the penetration level of RE.

7.1.1 Regional balancing

Costs for regional balancing are basically costs for implementing the necessary regulations or
modifications to existing regulations. In fact, regional balancing can lead to a higher efficiency in power
plant dispatch. If the dispatch of more expensive power plants in one state can be avoided by
dispatching cheaper capacity in another state, positive economic effects from a global perspective may
occur. This is also the reason why bilateral agreements between different states are already made today
and short-term energy markets have been introduced enhancing the economic overall efficiency of the
power plant dispatch. In order to estimate the benefit of regional balancing, the complete energy system
in India would have to be modelled with and without regional balancing which is out of the scope of this
study. However, it is likely that regional balancing has negative costs for the Indian economy and is
therefore the cheapest balancing option available.

7.1.2 Retrofitting
Costs for retrofitting power plants very much depend on the individual set-up of the plant. In any case
hardware adjustment is needed, which is usually costly and may lead to interruption in power plant
operation. Parameters which are subject for improvement are especially the minimum load (turndown),
start-up and down time improvement and ramp rate improvement. Exact cost estimates are rare in the
technical literature and very much case-specific. Figure 39 gives a rough idea of what the costs to
achieve different parameter improvements regarding gas-fuelled power plants are.
Information has been taken from (NREL 2013), plant size has been estimated based on turbine
description in the report (B-, E-, F-frame turbines and exemplary models stated and additional literature:
Ginter/Bouvay 2006, Michalke et al. 2012). For decrease of minimum load (turndown improvement) an
estimated Rs 40 – 60 lac20 is needed to improve turndown capability by 5 – 10% which is roughly
between Rs 1.0 – 2.2 lac/ MW. Start-up time improvement of 50 – 60% may costs around Rs 150 – 500
lac for a plant between 50 – 300 MW. However, improving combined cycle start-up time for large size
plants of around 300 MW may reach up to Rs 1,250 lac compared to Rs 500 lac for single cycle plants.
Cost estimates vary between Rs 1.5 – 4.5 lac/ MW. Ramp rate improvement is mentioned in the same
publication only for large power plants (around 300 MW) and may reach Rs 250 – 400 lac which leads
to costs of Rs 1.2 – 1.7 lacs/ MW.

20
Cost data has been converted according to 1 US-Dollar = INR 63.29

78 | P a g e
 Figure 39: Cost estimates of retrofitting gas turbines in single cycle and combined cycle plants

Source: NREL 2013

1.400
1.200
1.000

Combined cycle (only)

lac Rs

800
600 < 50 M W unit
400 75 -150 M W unit

200 150 - 300 M W unit

-
M in M ax M in M ax M in M ax
Turndow n St art -up t ime Ramp Rat e
Improvement (5-10% ) improvement (50- Improvement (100% )
60% )

For coal-fired power plants the NREL report mentions around 30 different measures to improve plant
flexibility requiring hardware adjustment of boilers, coal mills, emission control system, balance of plants
system, turbines or chemistry related improvements. The typical costs vary largely between Rs 150 –
10,000 lac depending on power plant size (200 and 750 MW) and on necessary hardware adjustment
(Rs 0.6 – 13 lac/ MW). Improvements of plant parameters in terms of ramp rates, minimum load, start-
up and down-time range between 30% and 50% of (see
Table 7).

Table 7: Different retrofitting measures, estimated costs and benefits for coal-fired power
plants
15
Source: NREL 2013
© Fraunhof er IWES
Cost Range Typical benefit
(million $, small sub-critical range
Parameter Nr of [200MW] / large sub-critical [500 (improvement of
affected* measures** MW] / supercritical 750 [MW]) parameter)
Boiler Retrofits 8 0.3 – 3 / 0.5 – 5 / 1 – 7 30-50%
Coal Mill
Retrofits a), b), c) 5 0.5 – 10 / 1 – 12 / 1.5 – 16 ~30%
Emissions
Control Retrofits a), b), c) 2 0.5 – 2 / 1 – 3 / 1.8 – 4 50%
Balance of Plant
Retrofits a), b), c) 5 0.57 – 4 / 1.5 – 7.5 / 2.25 – 8 30-50%
Turbine Retrofits a), b), c) 6 0.25 – 1 / 0.75 – 2 / 1 – 4 30-100%
Chemistry-
related
Improvements a), b), c) 4 0.3 – 1.5 / 0.5 – 3 / 3 – 4 50-100%
* a) ramp rate, b) Minimum load (turndown), c) start-up and shut-down ** Measures are outlined in the annex.

7.1.3 Storage options

From a long-term perspective storage options become a viable option to increase the integration of
RE. Due to the (currently) high costs of storage it is recommended to increase storage options after
considering less costly balancing options (like adaptation of conventional power plants, regional
balancing and Demand Side Management).

79 | P a g e
 The most competitive storage technology is pumped hydro storage. The tremendous potential in India
has been outlined in the previous section. Implementing new storage capacity should therefore focus
on the set-up of new pumped hydro storage power plants. Other storage options such as electrical
battery storage are today roughly 3 to 8 times more expensive due to very high investment costs (Rs
1,300 to 3,300 lac/ MW [IWES/IAEW 2014]).

Figure 40: Cost of electric load shifting for different storage options

Source: IWES/IAEW 2014

80
elect ricit y shif t ing cost s (in rs/kWh)

70

60

50

40

30

2020
2030
2050
2012
2020
2030
2050

20

10

0
PbS

NaS

AA-CAES
elect . Vehicle

H2-Kaverne

CH4-Gasnet z
(2)
PSW
Lit hium-Ion

Redox-Flow

large

CH4 in gas
H2 in gas
hydroPSW
Pumped

Pumped
hydro

grid

grid

short -t erm st orage short -t erm st orage long-t erm

st orage
decent ralized st orage cent ral large scale st orage

Cost of shifting electrical energy with this technology is estimated to range between Rs 9 – 11 per kWh
15
(assumptions: 8% interest rate, 35 years of lifetime, investment costs between Rs 780 – 950 lac/ MW,
© Fraunhof er IWES
7.5 hours of storage capacity, 3% of investment p.a. for operation and maintenance and a usage of
1,000 hours per year). The costs however depend very much on local site conditions, technical
parameters of the plant and financing details. All storage options including different battery types
(Lithium-Ion, lead-acid [PbS], Natrium- Sulphur [NaS], Redox-Flow), electric vehicles, Compressed Air
Energy Storage (CAES) and long-term storage options in form of power-to-gas in combination with
cavern storage or gas pipeline storage are compared in Figure 40.

7.1.4 Incremental Cost of Transmission

Integrating large amounts of RE and associated balancing mechanisms have an incremental cost of
balancing associated with them. This cost can be calculated by the following methodology. A grid
Simulation will have to be conducted. Based on the simulation transmission loss for the current
operating condition of the grid will have to be calculated. This exercise will have to be repeated with
more RE feed-in. A Comparison of losses between the two simulations will have to be conducted. This
would give us the incremental costs of transmission. Actual electricity costs of the overall energy mix
will have to be taken into account. Grid enforcement and ancillary services will have to be factored in
separately.

80 | P a g e
7.1.5 Partial load operation:
One of the key enablers of balancing will be the partial load operation of conventional power plants.
These plants would effectively function as the spinning reserves which can be called into action at short
notices. An analysis of the overall efficiency curve of the power plant will have to be done. This analysis
will need to factor in startup costs and also downtime costs. A commercial mechanism will have to be
put in place to support the maintenance of these capacity reserves.

81 | P a g e
8 Summary and Recommendations
For six Indian states where high penetration of renewables is expected or even already present the
capacity for balancing fluctuations of RE is assessed. The assessment includes existing and planned
renewable energies (RE) and balancing capacities from conventional power plants and hydro power
plants. An outlook on the use of storage technologies for balancing is given.

The electricity systems of all Indian states are interconnected to one single power grid. The grid size is
comparable to the European interconnected system – interconnections between different states are
better established than many grid connections between European countries. This offers a great
potential to integrate a high share of RE in the power system. Balancing in the sense of day- or hours-
ahead scheduling has a vital role within this integration task.

8.1.1 Main outcomes and recommendations

All over India the balancing potential is sufficient to handle todays and even higher shares of
RE generation. The crucial question is how to utilize the regional or national potential of balancing and
how to distribute the effort for RE integration within all states.

Organizing a burden sharing for the balancing task will become more and more urgent in order
to support a cost effective way of RE integration: This is valid not only for balancing of electricity
demand and supply, but also for RE electricity production. Burden sharing in terms of costs will be a
component of a successful strategy which realises the ambitious capacity addition targets set-up by the
Indian government. Efficient market mechanisms (i.e. for exporting power and selling power between
states) need to be found and existing regulation needs to be adjusted. A proper refinancing scheme for
RE will support these developments. The spatial enlargement of the balancing area and the
enhancement of inter-state power exchange of RE is most important to harmonize balancing potential
in non-RE rich states with the variable generation from RE in different regions in India.

Increasing the balancing capabilities of the Indian states from a technical perspective is of high
priority. In order to prepare the Indian power system for additional RE generation a variety of measures
are proposed which have been outlined summarized and evaluated on a qualitative basis.

For an efficient balancing the implementation of high quality forecasting of RE as well as load
is vital. Both will significantly reduce the uncertainty in system operation and is prerequisite for an
optimized scheduling and dispatch of conventional power plants.

The setup of new flexible power plants is of high importance from the technical infrastructure point
of view. Flexible power plant solutions add costs on future capacity addition. However, in relation to the
overall investment in new generation capacity these expenditures are rather small. In this context, a set
of very high flexibilities standards should be obligatory for new capacity addition. It should be enforced
that the flexibility can be effectively used in real operation.

Retrofitting of plants regarding technical flexibility is important also. Especially to handle

situations of high feed-in from RE which today contribute to over-frequencies in the grid or lead to
curtailment of RE. In addition, retrofitting would be beneficial for fast and flexible residual load following.
The increase of gas fuel availability would also enhance the balancing capabilities as already existing
generation capacity can be made fully operational if sufficient fuel is available.

Flexible hydro capacity with storage capability has to be further developed. The high seasonal
correlation of hydro power availability and power from RE is both chance and challenge. It is
recommended to study the use of hydro power plants for intra-day balancing of load and supply in each
state and assess its potential taking into account restrictions from multi-purpose use (i.e. irrigation).

82 | P a g e
A special focus should also be put on developing further pumped hydro storage plants. There
is a tremendous potential for pumped hydro storage plants in India that should be utilized in large scale.

A large spatial distribution of RE supply is recommended. This will minimize the balancing effort,
because geographical distribution of RE in combination with the large Indian power grid offers the
potential to smooth RE fluctuations significantly. This topic will be examined within this project in more
detail. In combination with the large Indian power grid it offers the potential to minimize RE fluctuations
in the first place.

Table 8: Measures to increase balancing capability in RE-rich states in India and qualitative
evaluation of priorities, costs and impact

Measures Priority Costs Potential Impact

(1 = highest, 3 = on Balancing
lowest)
Short-term solutions
Improve load forecasting 1 very low High
Implement RE forecasting 1 Low Very High
Improve balancing possibilities with central
1 no costs Medium
power plants
Retro-fit existing power plants 2 high High
Increase gas availability by import of LNG
2 high High
and higher domestic production
Increase of gas allocation to RE-rich states 1 very low Medium
Setup of new flexible conventional generation 1 high High
Definition of minimum flexibility requirements
1 very low High
for new conventional power plants
Medium-term solutions
Usage of hydro power stations with storage
1 Low Medium
for intra-day balancing
Increase the balancing area 1 negative costs Very High
Enhance inter-state power exchange 1 Low Very High
Setup of additional pumped hydro storage
1 high High
power plants
Demand Side Management 3 Low Medium
Increase transmission capacity of regions
2 high High
and to other countries
(very) Long-term solutions
Regional diversification of supply from wind
3 high Medium
energy
Battery storage 3 very high Low
Use of electric vehicle for balancing 3 high Low
Sector-Coupling (heat/electricity) 3 high Low
Power-to-gas 3 Very high Low
Source: Fraunhofer IWES

83 | P a g e
9 Overall Strategy Roadmap & Recommendations
The recent increase as well as the future projections in variable RE power generation introduces
considerable challenges in the electricity system’s management and operation. Due to the significant
variability of wind and solar resources wind and solar power forecasting, appropriate balancing actions,
as well as an effective control infrastructure are becoming mandatory.

This analysis provides a reliable inventory of the current electricity sector and its potential to meet the
needs for an increased RE integration. Based on this analysis, recommendations for the implementation
of forecasting techniques and balancing actions and for the establishment of an effective control
infrastructure are given.

9.1 Summary of the strategy and the recommendations

For the establishment of forecast systems as integral ports of the Renewable Energy Management
Centers the following recommendations are made
• Forecasting should be concentrated on the regional level. Regional forecasting results in significantly
lower uncertainties due to spatial smoothing effects.
• There is no need for single site forecasting–except for economic reasons given by the market
mechanisms.
• Forecasting by both, the RE power producer and the concerned RLDC, as is proposed in the recent
‘Framework for Forecasting, Scheduling & Imbalance Handling for Wind & Solar Generating Stations
at Inter-State Level ’is not recommended.
• The RE forecast system should at least provide the following functionalities:
- Wind and solar power forecasts on state (i.e., SLDC) level
- Forecast horizons of up to two days
- Temporal resolution of the forecasts 15 minutes
- Updates on a intra-day time scale
- Option for forecasts in the time scale of up to six hours
- Ramp forecasting (time of occurrence, duration, magnitude, ramp rate)
- Detailed information on forecast uncertainty
- Capability to make use of on-line measurement data
- Continuous forecast evaluation according to community-accepted accuracy measures.

• RE forecasting needs to be accompanied by a load forecasting scheme of at least the same

accuracy. This should take into account the regional differences in load structure.
• Standardised processes for forecast evaluation should be implemented. An Evaluation Handbook
should be considered as an option.
• The establishment of an educational program with respect to basic meteorological concepts, post-
processing techniques, probabilistic methods, forecast evaluation, and load forecasting is strongly
recommended. This includes training activities by external forecast providers.
• It is recommended to include the India Meteorological Department (IMD) in future wind and solar
power forecasting activities. Appropriate resources should be provided.
The recommendations regarding REMC framework and functional architecture are described below.
• REMCs should be established at State, Region and Nationals LDC levels. These will support
SCADA of RE generation injection at ICT and Pooling substations. The REMC will obtain forecasted
RE generation data for the area of responsibility from the external forecasting system and make it
available to REMC scheduling team in a RE Scheduling tool. REMC team can manually override the
forecast data before submitting this to the existing general scheduling tool. In future, a dedicated

84 | P a g e
 module to facilitate realtime monitoring of control reserves in area of responsibility as well as
neighbouring regions can be introduced (control reserve monitoring tool). Basic functional
capabilities of REMCs at all levels – SLDC, RLDC and NLDC should be same. This will give freedom
to configure the REMC as per prevailing regulations. REMC system will be the single point window
for RE developers to help them view and submit data pertaining to their plants (at pooling substation)
level.REMC system will also be the base reference and repository for commercial settlements.
• REMC implementation should be project managed through a single nodal agency.
• Dedicated staff for REMC should be appointed and trained immediately on forecasting, despatching,
monitoring and balancing of RE generation. XLDC staff and REMC staff should be rotated across
RE and conventional xLDC roles to help cross pollinate experential learnings.
• Until the new REMC systems are established, REMC staff can perform their duties by getting
dedicated additional operator and engineering consoles from existing xLDC systems.
• REMC at NLDC level is recommended to provide overall governance on RE generation with respect
to safe operation of the grid. REMC at RLDC and SLDC levels are recommended to address optimal
despatch of Renewables and conventional generation based on prevailing regulations and market
models.
• A dedicated SCADA and communications team needs to be setup at all xLDC levels.
• RE Developer system should not be integrated into the grid until they provide facility to acquire
SCADA data at 2-4 second refresh rate from the pooling station over a nationally standardised
interface.
• Communication Infrastructure for all ICTs and in future for Pooling substations should be upgraded
to support refresh rates of 2-4 seconds for SCADA data.
• Roles and responsibilities of Power Procurement Committees, Renewable Energy Development
Authorities, RE Developers and other new actors should be defined at policy and regulatory levels.
For the balancing domain the study concluded with two major general outcomes: (i) In light of the
sufficiently available balancing capacity it is essential to share the burden between all Indian states both
physical and economically (e.g. by adapting balancing areas, stimulate inter-state exchange, adequate
market mechanisms). (ii) In addition, a set of measures is recommended which contribute to the
effective increase of balancing capabilities:
• State-of-the-art forecast systems for RE production and load are necessary for efficient balancing
by reducing the uncertainty in system operation and providing the basis for an optimized scheduling
and dispatch of conventional power plants.
• New flexible power plants should be set up with high priority. A set of very high flexibility standards
should be mandatory for the addition of new capacity.
• Retrofitting of plants aiming at increasing their technical flexibility is important. This is essential for
handling situations of high RE feed-in and for a fast and flexible residual load following.
• Flexible hydro capacity with storage capability needs to be further developed. The high seasonal
correlation of hydro power availability and RE power generation is beneficial for intra-day balancing
of load and supply in each state. Restrictions from multi-purpose use need to be assessed.
• Developing further pumped hydro storage plants is a further focus area.Its huge potential in India
should be utilised in large scale.
• A large geographical distribution of RE generation is highly recommended. This minimizes the
balancing needs due to a strong smoothing of fluctuations in RE power generation.

85 | P a g e
Annexures

Annexure 1 - Yearly load and generation pattern

86 | P a g e
References

CEA 2013a: Large Scale Grid Integration of Renewable Energy Sources – Way Forward, Central
Electricity Authority, 2013.

CEA 2013: Standard technical features of BTG system for supercritical 660/800 MW thermal units,
Government of India, Central Electricity Authority, New Delhi, July 2013.

Ginter/Bouvay 2006: Uprate Options for the MS 7001 Heavy Duty Gas Turbine, GE Energy, Atlanta,
2006.

Michalke et al. 2012: Powerful Products for the Enhanced Flexibility of Gas Turbines, Siemens AG,
Cologne.

NREL 2013: Cost-Benefit Analysis of Flexibility Retrofits for Coal and Gas-Fueled Power Plants,
National Renewable Energy Laboratory, 2013.

Shah 2015: Automatic Demand Management Scheme in Gujarat Power System, Gujarat Energy
Transmission Corporation Limited, 5th International Exhibition & Conference: New Technologies T & D,
Renewable Energy Integration, Smart Grid, Energy Efficiency and Communication, New Delhi 2015.

Cochran et al. 2013: Flexible Coal – Evolution from Baseload to Peaking Plant, National Renewable
Energy Laboratory / Intertek, 2013.

Venkataraman et al. 2013: Cost-Benefit Analysis of Flexibility Retrofits for Coal and Gas-Fueled Power
Plants, National Renewable Energy Laboratory, GE Energy, Intertek, 2013.

Mehta/Nayak 2015: Improvement Needed in RRF Mechanism for Grid Management – How long we
can postpone RRF Mechanism, Gujarat Energy Transmission Corporation Limited, 5 th International
Exhibition & Conference: New Technologies T & D, Renewable Energy Integration, Smart Grid, Energy
Efficiency and Communication, New Delhi 2015.

PGCIL 2012: Transmission Plan for Envisaged Renewable Capacity, Vol.1, Power Grid Corporation of
India Ltd., 2012.

Batra 2015: Integration of Electricity from Renewable Energy Sources in the Grid, National Reliability
Council for Electricity, 5th International Exhibition & Conference: New Technologies T & D, Renewable
Energy Integration, Smart Grid, Energy Efficiency and Communication, New Delhi 2015.

VDE 2012: Erneuerbare Energie braucht flexible Kraftwerke - Szenarien bis 2020, Organization of
electrical engineering, electronics and information technology (Verband der Elektrotechnik, Elektronik
und Informationstechnik, VDE), 2012.

IWES/IAEW 2014: Roadmap Speicher, Bestimmung des Speicherbedarfs in Deutschland im

europäischen Kontext und Ableitung von technisch-ökonomischen sowie rechtlichen
Handlungsempfehlungen für die Speicherförderung, Kurzzusammenfassung, Fraunhofer Insitute for
Wind Energy and Energy System Technology (IWES), 2014.

IWES 2014: Kombikraftwerk 2 – Abschlussbericht, Fraunhofer Insitute for Wind Energy and Energy
System Technology (IWES), 2014.

87 | P a g e
Sterner/Stadler 2014: Energiespeicher, Bedarf - Technologien - Integration, Springer-Verlag Berlin
Heidelberg, 2014.

88 | P a g e
Imprint
The findings and conclusions expressed in this document do not

necessarily represent the views of the GIZ or BMZ.

The information provided is without warranty of any kind.

Published by
Deutsche Gesellschaft für

Internationale Zusammenarbeit (GIZ) GmbH

Indo – German Energy Programme – Green Energy Corridors

Registered offices: Bonn and Eschborn, Germany

B-5/2, Safdarjung Enclave

New Delhi 110 029 India

T: +91 11 49495353

E: [email protected]

I: www.giz.de

Authors

Christoph Richts (Fraunhofer IWES)

Dr. Philipp Strauß (Fraunhofer IWES)

Dr. Detlev Heinemann (University of Oldenburg)

Editors

Ernst and Young LLP

New Delhi, May 2015

This project/programme’ assisted by the German Government, is being

carried out by Ernst and Young LLP on behalf of the Deutsche
Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH.

89 | P a g e