2019 Innovations in Power and Advanced Computing Technologies (i-PACT)

ANN Based Solar Power Forecasting in a Smart

Microgrid System for Power Flow Management
Divya R, Gopika N P, Manjula G Nair
Department of Electrical and Electronics Engineering
Amrita Vishwa Vidyapeetham
Amritapuri, India
[email protected], [email protected], [email protected]

Abstract—Under the current energy scenario, the growth of of the MPPT controller is to achieve maximum available power
the existing grid is purely depending on the microgrids. Solar from the photovoltaic panels irrespective of the atmospheric
PV and Wind energy system are the two prominent microgrid parameters like solar irradiance and temperature. Such con-
sources that are widely and wisely accepted. The existence of
the non-linear loads in the power system is indelible, but it is trollers grant the desired amount of power, thereby curtailing
very necessary to avoid the power quality issues caused by them. the cost and enhancing the system reliability. Artificial Neural
These issues become worse during microgrid integration. Now Network (ANN), Hill Climbing method, Perturb and Observe
Power system utilities are promoting the microgrid integration method (P&O), Fractional Open Circuit Voltage algorithm
from the prosumers. The excess electricity will be purchased (FOCV), Incremental Conductance method (Inc- Cond) and
by the utility with a feed-in tariff which is acceptable by both
the prosumer and the utility. So the microgrid integration with Particle Swarm Optimization (PSO) are the widely used MPPT
the grid should be done economically. Since the microgrids are methods. Similarly, the Wind energy system also has many
variable power sources, they are hardly predictable in the nearby atmospheric dependencies. The potential wind energy resource
future. One of the major hindrances for proper grid integration in an area is severely influenced by various factors like air
is that solar power is unpredictable in nature. Power forecasting density, the speed of the wind and blade radius [2]. These
aids in economic integration, along with intelligent power flow
management in the integrated grid system. In this paper, a Fuzzy effects make a hindrance for a bulk integration with the main
Logic Controller(FLC) based intelligent power flow management grid. If the power forecasting is properly implemented in an
is implemented for a smart microgrid system with two different interconnected system, then by using different types of MPPT
microgrids. An Artificial Neural Network (ANN) based solar techniques, the full and optimal utilization of both solar PV
Power forecasting is also done using MATLAB/Simulink. Active and wind energy systems are possible. Different approaches
Power sharing, Reactive Power compensation, and Harmonic
Current eliminations are realized using a knowledge-based al- were proposed to forecast the power of these RES [13].
gorithm. The quality of power was depleted due to the harmonic
Index Terms—Fuzzy logic Controller, IcosΦ, Solar PV, Wind polluting nature by the current power electronic devices. Non-
Energy, Artificial Neural Network linear loads such as power electronic converters, DC motor
drives, rectifiers, battery chargers, controllers for variable
I. I NTRODUCTION speed motor drives, electronic power supplies, electronic bal-
The technical environment of the current power system lasts are presently used in a great extent [5]. The power quality
has changed drastically due to the integration of Renewable related problems mainly arise due to the harmonic effects
Energy Sources (RES). Massive interfacing of these variable produced by non-linear loads. They constitute the primary
power sources leads to the evolution of smart grid. However, source of harmonic distortion. The harmonic current formed
the fact is that the operators still have to face the great by these nonlinear loads is injected back into the power
challenge of integrating this kind of sources into the power system through the point of common coupling. The deceived
system. Many solutions were proposed to dilute or to address supply current distorts the supply voltage profile. This further
this technical issues [1]. affects the microgrid that is integrated with the main grid
The significant or main RES include mainly solar PV and [14]. So a proper active shunt filter can be used to eliminate
Wind energy systems. The major advantages of these are, they the harmonic issues. [6] discuss the smart park unit which
contribute pollution-free energy and causes no greenhouse gas can provide active, reactive power compensation. A solar PV
to be emitted, virtually no maintenance is required as solar system in the grid-connected mode as a shunt active filter with
panels last over 30 years in the case of solar PV, can be a suitable control algorithm is discussed in [7]. Powerflow
installed virtually anywhere, can integrate batteries for reliable management is an essential factor in the field of the smart
operation. The major disadvantage of solar PV is that the grid. Many intelligent controllers are proposed for better power
accurate knowledge of the solar cell parameters depends on flow management in the smart grid system. This paper deals
irradiance level and temperature. The Maximum Power Point with the intelligent controller based power flow management
Tracking (MPPT) [3] [4] method accurately tracks the maxi- in a smart microgrid system, where the solar power forecasting
mum available power from a solar PV. The primary objective aids power flow management.

978-1-5386-8190-9/19/$31.00 ©2019 IEEE 1

 2019 Innovations in Power and Advanced Computing Technologies (i-PACT)

II. S YSTEM M ODELLING

The block diagram shown in fig.1 depicts the grid
interfaced, two micro grid system with intelligent controllers.
The microgrids are solar PV and wind energy system. These
are connected to the utility grid via an interfacing inverter. The
integrated power flow management system is implemented
using a fuzzy logic controller (FLC). The FLC inputs are the
forecasted power of the microgrid and the load demand, and
the FLC output is the tariff value for the microgrid suggested
by the microgrid owners in response to the power availability
form the microgrid. The power forecasting controller is Fig. 2. FIS editor window
realized by an Artificial Neural Network (ANN) controller.
The tariff value from the two microgrids are checked by the
tariff checking block and the grid will select that microgrid Power (HP) as shown in fig.3, and load demand are Low Load
with low tariff value. (LL), Medium Load (ML) and High Load (HL) as shown in
fig.4. Fig.5 shows the membership functions for the output
variable tariff value as suggested by each microgrid. They are
high tariff value and low tariff value.

Fig. 3. Graphical view of input membership function forecasted power

Fig. 1. Basic Block Diagram of the grid integrated system

III. P OWERFLOW M ANAGEMENT C ONTROLLER

The power flow management is essential in a smart grid
system. Microgrids refers to the production of power near the
consumption place. These microgrids have an important role
in the reduction of power loss, and in maintaining the power
flow management in the interconnected grid system. In this
proposed system, Powerflow management is done by using
FLC controller. Fuzzy systems (FS) are based on fuzzy set Fig. 4. Graphical view of input membership function load demand
theory and associated techniques founded by Lotfi A Zadeh.
The rules defining the FLC operation are in the form of The output from FLC controller is the tariff value suggested
linguistic variables represented by fuzzy sets. An interface by the microgrid owners according to the availability of the
mechanism helps to achieve the FLC output [8]. A Sugeno power from the microgrid. The prosumer will suggest a low
model based FLC approach is adopted. The two inputs of tariff value if and only if the power from the microgrid is
the FLC are the forecasted power of microgrid and the load sufficiently large, which shows the willingness of the micro-
demand which are defined in different ranges. Fig.2 depicts grid to supply the grid. The Grid will have a comparator, that
the FIS editor window for the controller. The forecasted power compares the tariff suggested by both the microgrids and will
from the ANN controller and the load values are converted into have a selection among them.
linguistic variables by the fuzzy controller. Based on the set of Fig.6, Fig.7, Fig.8 depicts the decision making capability of
rules defined by the user the output tariff value is found. The the FLC power flow management controller. Fig.7 shows the
membership functions for one of the input variable forecasted low tariff value corresponding to high microgrid power and
power are Low Power (LP), Medium Power (MP) and High high load condition. In the same way, irrespective of the load

2
 2019 Innovations in Power and Advanced Computing Technologies (i-PACT)

IV. S OLAR P OWER F ORECASTING C ONTROLLER

Renewable energy resources are variable power sources,
which have the ability to support the grid. But the major
impediment is that they are hardly predictable in the nearby
future and the power output depends upon the atmospheric
parameters [12]. Solar PV power is mainly depending on the
environmental conditions like the atmospheric temperature and
solar irradiance. The solar PV power forecasting can be estab-
lished by using different knowledge-based algorithms. In the
past few years, many practices and techniques were developed.
Fig. 5. Graphical view of output membership function tariff value for the Some of the best methods include regressive methods which
micro grid
work well in data-rich and poor environmental conditions,
the stochastic methods and ANN modeling, which provides
enhanced non-linear data prediction [9] [10]. Here a solar
power forecasting algorithm is performed by using an ANN
controller. The ANN controller requires historical weather data
like the atmospheric temperature, solar irradiance, and the
corresponding solar power generation [11].





P = Pref I/Iref 1 + b T − Tref (1)

The equation, shown is used to verify the historical data

needed for the ANN controller training, where the solar output
Fig. 6. FLC rule viewer for low MG power and low load condition
power is a function of atmospheric temperature and solar
irradiance [11]. About 2061 sample power values correspond-
ing to the different irradiance and temperature conditions are
obtained and is used for ANN training.

Fig. 7. FLC rule viewer for high MG power and Low load condition

Fig. 9. ANN network with 20 hidden layers

Neural Network Fitting Tool (NN tool command) is the

method adopted to train the required network. The MATLAB
Fig. 8. FLC rule viewer for medium MG power and high load condition environment is used for framing the ANN model for solar
power forecasting. The ANN model has an input layer, hidden
layers, and an output layer. The hidden layers include 20
condition the microgrid tariff value for the microgrid will be nodes. The NN tool is trained in such a way that 60% samples
low with the high availability of the microgrid power. For a towards training, remaining 20% samples for validation and
medium availability of the microgrid power, the FLC will have testing. The neural network size can be adjusted according
a check for the load condition. If the load value is high then to the designer by adding or deleting the hidden layers. fig.9
only the tariff generated will be low. depicts the ANN controller with 20 hidden layers.

3
 2019 Innovations in Power and Advanced Computing Technologies (i-PACT)

The training automatically ceases when generalization halts in three modes based on a gain value k. This gain value is
correcting, and the accuracy of the training can be analyzed by calculated by considering different system conditions. Some
going through the regression plot and error histogram. From of the parameters include battery SoC level, availability of
fig.10, it is evident that the predicted solar power is unbiased RES and source conditions. Based on the mode selected, the
since the residual plots have both negative and positive plots. control algorithm generates the compensation current which
The R-value depicted in fig.11 indicates that the predicted is exactly in phase opposition to the harmonic content in the
value is closer to the actual target values. system [6].
VI. M ODELLING OF P ROPOSED I NTELLIGENT S YSTEM
A 3Φ AC grid with 415 V, 50 Hz is considered as the
primary source of energy. Two microgrids are integrated
with the main grid. A solar PV module sun perfect solar
CRM115S125M-48 module is considered as the first mi-
crogrid, a PMSG based wind energy system is the second
microgrid. A non-linear load of 15 kW is considered for the
analysis, which is introduced in a step by step manner. The
specifications for the solar PV used is shown in table 1.

TABLE I
PV SYSTEM DESCRIPTION

Description Rating
Maximum power(Pmp) 115 W
Maximum current(Imp) 5A
Maximum voltage(Vmp) 23 V
Fig. 10. Error histogram plot Short circuit current(Isc) 5.49 A
Temperature(T) 25 o C
Open circuit voltage(Voc) 28.5 V
Number of parallel strings 2
Number of series strings 5

Both the microgrids, solar PV and the Wind system are

designed to provide a maximum power of 1150 W and 1000
W respectively at STC condition.

TABLE II
B OOST CONVERTER DESCRIPTION

Description Rating
Inductor 2.13 mH
Capacitor 100 uF
Vin 115 V
Vout 650 V

The boost converters (with specification given in Table II)

Fig. 11. Regression plot
are used to link the microgrids with the DC link having a
voltage equals to 650 V. PI controller is used for the closed
loop operation of boost converter for both the microgrids. The
V. K NOWLEDGE -BASED C ONTROLLER DC link voltage is calculated based on the given equation.
The IcosΦ controller is one of the most efficient controllers Where Vll is the system line to line voltage, m=1.
used for achieving effective harmonic and reactive power √
2 × 2 × Vll
compensation [5]. The aforementioned controller is used to Vdc = √ (2)
balance the nonlinear load-based system. The case will be 3×m
worse for the renewable resources interfaced grid system. The Modified IcosΦ is used as the active shunt filtering algo-
grid tie inverter itself can be modified to act as an active shunt rithm in the system. The power flow management FLC is
filter. Modified IcosΦ controller is having an advantage that the controlled by the forecasted power obtained from the ANN
generated reference current will be automatically adjusted for controller and different load conditions. The FLC for both the
any variations in the source current, grid tariff and the SoC of microgrids provides a suitable tariff value, which is a function
the integrated renewable energy resources [12]. The modified of power availability from the respected microgrids and load
IcosΦ algorithm for active shunt filtering is mainly defined status as explained in the previous power flow management

4
 2019 Innovations in Power and Advanced Computing Technologies (i-PACT)

section. And based on the suggested tariff value by the mi-

crogrid, the main grid will have a selection among them. The
active power sharing, reactive power compensation, harmonic
elimination is analyzed by using MATLAB/Simulink.

VII. A NALYSIS AND V ERIFICATION OF THE S YSTEM

The change in load conditions in the integrated system is

met by the main grid and the two supporting microgrids,
namely solar PV and wind energy system. The overall system
works intelligently based on two controllers which are FLC
and ANN power forecasting controller. The ANN power
forecasting controller output for the solar PV is shown in
fig.12. The figure shows the forecasted power for different
irradiance and temperature conditions. This forecasted power
value can be verified with the actual solar power value by Fig. 13. Plot for tariff value and control signal for micro grids
using the equation 1. This forecasted power is one of the
input variables for the FLC controller. The second input for the
FLC controller is the forecasted load values. FLC controller A. Case 1: Under Unequal Tariff condition
output is the corresponding tariff value of each microgrid. From fig.13, until 0.3 sec the solar PV is having low tariff
Fig.13 shows the FLC controller output for the given solar value in comparison with the wind system. Hence the solar
forecasted power and different load conditions. Indirectly the PV is to be integrated with the main grid which is depicted
tariff value is based on the availability of the microgrid power, in the fig.14. From 0.3 sec to 0.5 sec wind energy system is
which the prosumer can find out with the help of the ANN having the low tariff, hence it is sharing power with the main
controller. So that the microgrid owners or the prosumers can grid. The compensation current produced in different periods
suggest a feed-in tariff value to the main electricity providers and the MG generated current are depicted in fig.15.
depending on that tariff value, the tariff checking block will
have a selection among the two microgrids. Fig.13 also depicts
the control signal for the grid integration of the microgrid
based on the tariff value.

Fig. 14. Plot for active power sharing between micro grids and main grid
under different tariff conditions

B. Case 2: Under Equal Tariff condition

Fig.16 depicts the active power sharing between the micro-
Fig. 12. Plot for ANN power forecasting controller grids and the main grid under equal tariff condition. In case, if
the power availability of both the MGs is equal, then both the
Till 0.3 sec, the tariff of the solar PV is low compared to that MGs will be having an equal tariff. Hence both the systems
of the wind energy system. Hence the corresponding control will be integrated with the main grid. Hence in both cases, the
signal will be high for the solar PV integration unit. The active active power sharing is achieved.
power sharing depends on the suggested tariff values is shown THD after compensation by using modified IcosΦ controller
in fig.14, fig.16. The two possible tariff conditions are: is shown in fig.17.

5
 2019 Innovations in Power and Advanced Computing Technologies (i-PACT)

controller outputs are verified, the overall system under dif-

ferent tariff conditions was analyzed in MATLAB/Simulink.
The power forecasting can be used to mitigate some of the
challenges that arise from the uncertainty in the resource and
also in the overall power flow management of the smart grid.
The solar power forecasting is witnessing growing attention
from the research community.
R EFERENCES
[1] N. Phuangpornpitak and S. Tia, ”Opportunities and challenges of inte-
grating renewable energy in smart grid system,” 10th Eco-Energy and
Materials Science and Engineering Symposium Energy Procedia, vol.
34, pp. 282 290, 2013.
[2] V. Yaramasu, B. Wu, P. C. Sen, S. Kouro and M. Narimani, ”High-
power wind energy conversion systems: State-of-the-art and emerging
technologies,” in Proceedings of the IEEE, vol. 103, no. 5, pp. 740-788,
May 2015.
Fig. 15. Compensation current and MG generated current [3] K. R. Bharath and E. Suresh, ”Design and implementation of improved
fractional open circuit voltage based maximum power point tracking
algorithm for photovoltaic applications,” International Journal of Re-
newable Energy Research- IJRER, vol. 7, no. 3, 2017.
[4] A. A. Anu and R. Divya, ”Multiple input DC-DC converters for solar
cell power supply system and its maximum power point tracker,”
2013 International Conference on Energy Efficient Technologies for
Sustainability, Nagercoil, pp. 287-290, 2013.
[5] K. Elissa, G. Bhuvaneswari and M. G. Nair, ”Design, Simulation, and
Analog Circuit Implementation of a Three-Phase Shunt Active Filter
Using the I cos Φ Algorithm,” in IEEE Transactions on Power Delivery,
vol. 23, no. 2, pp. 1222-1235, April 2008.
[6] V. Raveendran and M. G. Nair, ”SmartPark as Shunt Active Filter using
modified ICos controller,” 2014 POWER AND ENERGY SYSTEMS:
TOWARDS SUSTAINABLE ENERGY, Bangalore, pp. 1-6, 2014.
[7] L. S. Stanly, Divya R and M. G. Nair, ”Grid connected solar photovoltaic
system with Shunt Active Filtering capability under transient load con-
ditions,” 2015 International Conference on Technological Advancements
in Power and Energy (TAP Energy), Kollam, pp. 345-350, 2015.
[8] B. Sivaneasan, C. Yu, and K. Goh, ”Solar forecasting using ANN with
Fig. 16. Plot for power sharing under equal tariff condition fuzzy logic pre-processing,” Energy Procedia, vol. 143, pp. 727 732,
2017.
[9] Y. K. Semero, J. Zhang and D. Zheng, ”PV power forecasting using an
integrated GA-PSO-ANFIS approach and Gaussian process regression
based feature selection strategy,” in CSEE Journal of Power and Energy
Systems, vol. 4, no. 2, pp. 210-218, June 2018.
[10] R. Srivastava, A. N. Tiwari and V. K. Giri, ”Forecasting of Solar
Radiation in India Using Various ANN Models,” 2018 5th IEEE Uttar
Pradesh Section International Conference on Electrical, Electronics and
Computer Engineering (UPCON), Gorakhpur, pp. 1-6, 2018.
[11] C. M. Whitaker, T. U. Townsend, H. J. Wenger, A. Iliceto, G. Chimento
and F. Paletta, ”Effects of irradiance and other factors on PV temperature
coefficients,” The Conference Record of the Twenty-Second IEEE
Photovoltaic Specialists Conference - 1991, Las Vegas, NV, USA, vol.1,
pp. 608-613, 1991.
[12] D. R. Nair, Devi S, M. G. Nair and K. Ilango, ”Tariff based fuzzy
logic controller for active power sharing between microgrid to grid
with improved power quality,” 2016 International Conference on Energy
Efficient Technologies for Sustainability (ICEETS), Nagercoil, pp. 406-
409, 2016.
Fig. 17. THD value after compensation [13] J. Liu, W. Fang, X. Zhang and C. Yang, ”An Improved Photovoltaic
Power Forecasting Model With the Assistance of Aerosol Index Data,”
in IEEE Transactions on Sustainable Energy, vol. 6, no. 2, pp. 434-442,
April 2015.
VIII. C ONCLUSION [14] X. Liang, ”Emerging Power Quality Challenges Due to Integration
of Renewable Energy Sources,” in IEEE Transactions on Industry
This paper has proposed a two micro grid integrated smart Applications, vol. 53, no. 2, pp. 855-866, March-April 2017.
grid system with power flow management based on the
availability of power from each microgrid. An ANN-based
solar power forecasting unit is implemented which can assist
the prosumer to suggest a feed-in tariff value based on the
microgrid power availability. The power flow management
using FLC controller and ANN based solar power forecasting