Control and Management of Stand-

alone Microgrid Using Artificial

Neural Network
A report submitted for partial fulfillment of degree of

BACHELOR OF TECHNOLOGY IN ELECTRICAL ENGINEERING

By

Kangkan Jyoti Kalita (20-1-3-015) & Abhijit Doley (20-1-3-021)

Under the supervision of

Dr. Saheli Ray
Assistant Professor

Electrical Engineering Department

National Institute of Technology Silchar
May 2024

1|Page
 DECLARATION
This is to certify that the project work entitled “CONTROL AND MANAGEMENT
OF STAND-ALONE MICROGRID USING ARTIFICIAL NEURAL NETWORK” is a work
carried out by Kangkan Jyoti Kalita (2013015) & Abhijit Doley (2013021) for the award of
Bachelor of Technology, Silchar. It is certified that the project report of their own work carried
out during the period from January 2024 to May 2024 (8th semester) under my supervision and
this report has not been submitted elsewhere for any degree/diploma.

Signature of Supervisor

Dr. Saheli Ray

Assistant Professor
Electrical Engineering Department
National Institute of Technology Silchar
Date: --/05/2024

2|Page
 ACKNOWLEDGMENT

We would not have been able to complete this project without the creative freedom that
our supervising teacher, Dr. Saheli Ray has provided us with. She never put unnecessary
pressure on us and never rushed us in any manner. She gave us enough time and support to
come up with this project and guided us throughout, and to its completion.

We all hope that we are able to produce some good results and maybe design a small
model of the real- life implementation of our project. We all hope to make her proud.

We would also like to thank all the faculties of the department to provide us with all the
support that were needed during this project

The constant support of the institute in inculcating innovation and research among the
students has also been the key of motivation to the students and hence needs to be
acknowledged.

3|Page
 .

Table of Contents
Declaration................................................................................................................................ 2
Acknowledgement .................................................................................................................... 3
Table of Contents ..................................................................................................................... 4
List of Figures ........................................................................................................................... 5
List of Tables ............................................................................................................................ 6
List of Abbreviations ............................................................................................................... 6
Abstract ..................................................................................................................................... 7
Chapter 1: Introduction 8-11
1.1 Introduction ................................................................................................................... 8
1.2 Literature review ........................................................................................................... 9
1.3 Motivation ................................................................................................................... 10
1.4 Objectives ..................................................................................................................... 11
Chapter 2: System Studies 12-16
2.1 Photovoltaic Cell ............................................................................................................ 12
2.2 DC-DC Converter .......................................................................................................... 14
2.3 Maximum Power Point Tracking ................................................................................... 16
Chapter 3: Solution Approach 17-20
3.1 System Identification using Artificial Neural Network ................................................ 17
3.2 Model Predictive Control ................................................................................................ 19
Chapter 4: Simulation Result and Discussion 21-27
Chapter 5: Conclusion 28

4|Page
 List of Figures
1. Fig 1.1. Block Diagram of Stand-alone Microgrid
2. Fig 2.1. Photovoltaic Effect in p-n diode
3. Fig 2.2. Equivalent Circuit Diagram of PV cell
4. Fig 2.3. I-V characteristics of a PV cell
5. Fig 2.4. PV array Block diagram in SIMULINK
6. Fig 2.5. Configuration of DC-DC Buck-Boost Unidirectional Converter in one cycle
7. Fig 2.6. Design of DC-DC unidirectional Buck-Boost Converter
8. Fig 2.7. (a) Connection diagram of PV array to load with DC-DC converter, (b) I-V and power
characteristics with load line passing through mpp.
9. Fig 2.8. Flowchart of P&O based MPPT controller
10. Fig 2.9 Duty cycle control of a PV cell around the MPP.
11. Fig 3.1 Simulink Plant used for System Identification
12. Fig 3.2 input dataset – Irr, temp, dutyCycle; output dataset – power for System Identification
generated in MATLAB
13. Fig 3.3 Block diagram of NARX model
14. Fig 3.4 Block diagram of MPC Control Techniques
15. Fig 3.5 MPC control action for tracking reference r
16. Fig 4.1 Linear ARX Model response without any custom regressors
17. Fig 4.2 NARX Model response with wavelet as nonlinear function
18. Fig 4.3 NARX Models response with different no. of neural network
19. Fig 4.4 NARX Model response with ‘relu’ activation in both the hidden layer, neuron size (a)
[10 10], (b) [10 15].
20. Fig 4.5 MATLAB generated dataset of Irradiance and temperature
21. Fig 4.6 Practical dataset of Irradiance and temperature
22. Fig 4.7 P&O based plant V-I output for MATLAB generated dataset
23. Fig 4.8 P&O based plant V-I output for practical dataset
24. Fig 4.9 P&O based plant output power for (a) MATLAB generated dataset, (b) practical dataset
25. Fig 4.10 MPC controlled Plant Output for dataset 1
26. Fig 4.11 MPC controlled Plant Output for dataset 2

5|Page
 List of Tables
1. Table 2.1 Specification of solar module.
2. Table 2.2 Specification of converter.
3. Table 3.1 Sampling Period of Input/output dataset of Model
4. Table 3.2 Linear and Polynomial Regressors set in NARX Model
5. Table 3.3 Custom Regressors set in NARX Model
6. Table 4.1 P&O based plant output analysis for both dataset (all unit are in watt)
7. Table 4.2 MPC based plant output analysis for both dataset

List of Abbreviations

1. MPC – Model Predictive Control

2. MPP – Maximum Power Point

3. MPPT – Maximum Power Point Tracking

4. CCM – Continuous Condition Mode

5. Voc – Open Circuit Voltage

6. Isc – Short Circuit Current

7. PV – Photovoltaic

8. Imp – Current at maximum power

9. Vmp – Voltage at maximum power

10. P&O – Perturb and Observation

11. Vpv – PV voltage

12. Ipv – PV current

13. Pout – Actual power output

14. PWM – Pulse Width Modulation

6|Page
 Abstract
With the advancement of technology and increase in population, consumption of energy
increased to a very high value. So to keep an eye on the limitation of conventional energy
resources, people are shifting to different renewable energy sources - solar, wind, hydro energy
etc. Microgrid integrates distributed energy sources, battery storage for stable power regulation
which make it independent from conventional grid.
In this project, we have presented a stand-alone microgrid with a Solar PV. The main objective
of the project is to develop a faster, accurate and adaptive dynamic controller for extracting
maximum power from solar module for varying irradiations and temperatures. We have used
Deep Learning System Tool box in SIMULINK to model the PV system dynamics and used
Model Predictive Control algorithm for MPPT.

7|Page
CHAPTER 1

1.1 Introduction
Microgrid uses distributed energy resources like solar, wind energy which makes it more
efficient and economical as it reduces cost of installing long power transmission infrastructure
thereby reducing power losses. Microgrid has the flexibility of operating in stand-alone mode,
making the microgrid energy independent from conventional grid. Based on the types of
distribution network and loads connected, microgrid can be classified as DC microgrid, AC
microgrid and hybrid AC/DC microgrid.

As the Stand-alone microgrid has its own sources of generating electricity and storage system
it is used where power transmission and distribution from major centralized energy sources are
too far and costly to operate. We have chosen microgrid in place of conventional grid because
of the following reasons – firstly, conventional grid rely on large centralized power plants to
generate electricity such as coal, natural gas etc. but microgrid incorporated with distributed
energy resources like solar panels, wind turbines ,combined heat & power systems and energy
storage to electricity locally , secondly conventional grid has only one directional power flow
from centralized power plants to consumers but microgrid can operate in both grid connected
and islanded mode allowing two way of power flow between grid & local sources. In grid
connected mode, it can be synchronized with the main grid for the import and export of
electricity during peak, off-peak load or excess power generation from distributed sources and
maintain grid stability. Also when the main grid experiences a fault, the microgrid facilitate to
the Islanding capability as it manages the isolation of microgrid from the main grid and
maintain power supply.

Also for incorporating our microgrid simulation we have chosen Model Predictive Controller
(MPC) with Artificial Neural Network Model, in place of other controller such as PID
controller, Fuzzy logic controller etc. because microgrid is highly nonlinear , unpredictable and
complex dynamic systems , so if we used PID then tuning of these PID gains to control the
error in in system which is not feasible , control using fuzzy logic is very slow and requires
creation of complicated rules but ANN have high self-learning and nonlinear mapping
capability with high generalization capabilities . So using neural network to identify system
dynamics we can trained the data gaining information from the system being modelled then
more fast output response we can produce .So control of microgrid using MPC with neural
network identified model will give better system performance.

8|Page
1.2 Literature review
Xuesong Zhou, Tie Guo, Youjie Ma [1] –This paper gives the basic concept of microgrid, then
describes about microgrid research in the current domestic and international scale. It discuss
about the key problems and development prospects associated with microgrid technology.
Zia Ullah, S Wang, J Lai, M Azam, F Badshah, G Wu, M R Elkadeem [2]- This paper discuss
about the implementation of various intelligent control techniques for efficient management of
energy in hybrid microgrid system which integrates renewable energy sources with battery
storage. It compares different control methods such as Artificial neural Networks , Fuzzy logic,
Sliding Mode Control and Maximum Power Point Tracking to optimize power and voltage
balance in the microgrid. Key Finding include the superior performance of ANN and PID
controllers in voltage power stability compared to FL and SMC controllers.
Y Boujoudar, M Azeroual, H E Moussaoui, T Lamhamdi[3]- The paper presents an intelligent
management of energy strategy for stand-alone DC Microgrid system with an Artificial Neural
Network Controller (ANNC). The focus is on controlling the DC/DC Bidirectional Converter
that connects a Li-ion battery with the DC Bus of the stand-alone power system. The results
demonstrate the robustness and satisfactory performance of the ANNC in managing energy in
stand-alone systems.
Samer Said, Ahmed Massoud, M Benammer, Shehab Ahmed[4]- The paper presents a
MATLAB/SIMULINK-based model for photovoltaic arrays using the SimPowerSystems
Toolbox. It focuses on simulating and monitoring the performance of PV arrays by considering
the factors such as solar insolation, temperature and shading effects. It gives study of dynamic
performance and characteristics of PV array in maximum power point tracking in different
partial shading patterns.
Daniel C Zacharek, Filip Sundqvist[7]- The paper having contribution of design of
bidirectional buck-boost converter to handle battery application. It focuses on the design of a
bidirectional DC/DC battery management system for electrical yachts like power transfer
mechanisms between different voltage batteries, software requirements for controlling the
system.

Max Schwnezer, Muzaffer Ay, Thomas Berges, Dirk Abel [17] – This paper gives us a detailed
review on Model Predictive Control (MPC) in the context of automation for technical systems.
It covers the historical evolution, theoretical foundation, practical consideration and application
of MPC. It addresses the computational burden associated with MPC design and
implementation. The paper discusses key methods in MPC development and points towards
future trends in the field.

Oliver Nelles [18] –The book gives comprehensive guide to understanding and applying
various models and methods for nonlinear system identification. This book covers topics such
as optimization techniques, model validation and model structures etc. for nonlinear system
identification.

9|Page
Akira Ohata, Katsuhisa Furuta and Hiroaki Nita [19] – This paper presented a method to
identify a nonlinear system using input-output data and represent it with a nonlinear ARX
model with parameters that depend on the input and output. The approach integrates linear
ARX model to identify the nonlinear model and expresses the parameters approximating their
dependence on the input and output.

Yuya HASUIKE, Masaki IZUTSU, and Shosiro HATAKEYAMA [20] – The paper proposes
a variable order model for representing nonlinear systems that have different orders in different
ranges. The variable order model combines the NARX model with an error model. The model
parameters are identified using the method of least squares. The paper compares the variable
order model with the NARX model and shows that the variable order model represents the
dynamics of the system more accurately. The paper concludes that the variable order model is
effective for representing the nonlinear systems and suggests future applications in online
identification and adaptive control systems.

Feroz Shaik, Syam S Lingala, Punnaiah Veeraboina [21] – The paper discusses the impact of
various parameters on the performance of Solar PV Power Plants, considering environmental
factors like solar insolation, temperature, humidity and dust accumulation. Different models
for parameter estimation of PV cells are explored, including single diode as well as two diode
models. The study also covers methods for reducing the effects of environmental factors on PV
productivity, such natural cleaning, water cleaning and manual cleaning.

Morcos M, Sertac B, Robert S. Balog, Haitham Abu Rub [22] – This paper discuss the
application of Model predictive Control (MPC) for PV maximum Power Point Tracking
(MPPT) using a fixed step technique combined with a boost converter. The proposed MPC-
MPPT method is more efficient and robust compared to traditional MPPT strategies.

Yingying Zhao, Aimin An, Yifan Xu, Q Wang and Minmin Wang [23] – This presents the a
new MPC method for grid-connected PV power generation systems. The study proposes an
MPC-based MPPT method for PV modules and a Finite Control Set Model Predictive Current
Control strategy for the inverter.

Rohan S. Kulkarni, Dhananjay B. Talange [24] – This paper uses a black box modeling
approach to models solar PV modules based on solar irradiance and module temperature data.
The identification experiments demonstrated that a black-box model can effectively capture
the dynamics of a solar PV array system with a small amount of data.

1.3 Motivation
The project sheds light on the multi-variate factors for designing our PV system as per our
given irradiation and temperature .We have collected our own data from NASA Power
Technology for hourly simulation of the system and will be trained by using Artificial Neural
network to identify the PV system dynamics with its response from plant input and output.

10 | P a g e
The project will solved the problem of accuracy, oscillation as long as we can trained the data
from input then more and more fast response output can be generated.

In the end, we will also demonstrate the comparison between the conventional P&O based
MPPT with MPC controlled neural network Model output.

1.4 Objective
Our objective is to maintain the peak power operation for given irradiation and temperature
from PV system ,we need to keep tracking the maximum power operating point .This can be
achieved by adjust the load line so that it passes through maximum power operating point of
Power-Voltage curve of PV system using dc-dc unidirectional converter .

With increase in the size and complexity of microgrid, the traditional way of achieving
maximum power point tracking may not be efficient and take longer time, so we will design
a MPC controller with neural network based identified model to harness the maximum
power from varying irradiance and temperatures.

Fig 1.1. Block Diagram of Stand-alone Microgrid

11 | P a g e
Chapter 2

System Studies
2.1 Photovoltaic Cell
A photovoltaic cell is simply a semiconductor p-n diode which works based on the principle of
the photovoltaic effect. The conversion of light into voltage and electric current is called the
photovoltaic effect. When a light beam falls on the junction of p-n diode, it creates an electron-
hole pair which is swept across the junction by electric field near space charge region, thereby
developing potential across the junction. In this way, minority e- from P-side will come to the
N-side (leaving behind their positively charged partner hole) and minority holes from P-side
will come to the N-side (leaving behind their negatively charged partner e-). PV cells are
commonly made up of mono-crystalline silicon and poly-crystalline silicon.

- -
EC
EC
Light
Light
- -
EC EF EC
EF
EV EV EF
EF + +

EV EV
+ + Space Charge
P-type Space Charge P-type N-type
N-type Region
Region
Fig 2.1. Photovoltaic Effect in p-n diode
The characteristics of PV cells can be shown
using the equivalent circuit shown in Fig
2.2. The equivalent circuit is composed of a
current source, diode, series and parallel
resistor representing the practical current
and voltage source. Applying KVL in the
circuit we get,

Fig 2.2. Equivalent Circuit Diagram of PV cell

𝑉𝑑 𝑉𝑑
𝐼 = 𝐼𝑃 , 𝐼𝑂 (𝑒 𝑛𝑉𝑇 , 1) , , 𝑅𝑠 ≪ 𝑅𝑠ℎ , , , , , ,(1)
𝑅𝑠ℎ

12 | P a g e
Where IP is current generated by light, I0 is diode saturation current, q (= 1.6 × 10-19 C) is
charge of one electron, VT is thermal voltage constant, Vd is the diode voltage, n is the ideal
factor, RSH and RS represent the series and shunt resistance.

As long as the radiation increases current generated by PV cell increases. From eqn (1). Short
circuit current can be given as:
𝑉𝑑 𝑉𝑑
𝐼𝑆𝐶 = 𝐼𝑃 , 𝐼0 (𝑒 𝑛𝑉𝑇 , 1) ,
𝑅𝑠ℎ
𝑉𝑑 𝑉𝑑
𝑉𝑑 = 𝐼𝑆𝐶 . 𝑅𝑠 ≈ 0, , →0
𝑅𝑠ℎ 𝑉𝑇
𝐼𝑆𝐶 = 𝐼𝑃 ∝ 𝑠𝑜𝑙𝑎𝑟 𝑝𝑜𝑤𝑒𝑟 (𝑖𝑛𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛) , , , , , , , (2)

Thus short current is directly proportional to solar irradiation or insolation. And Open circuit
voltage ca be give as:
𝑉𝑂𝐶 𝑉𝑂𝐶
𝑉𝑂𝐶 = 𝐼𝑃 , 𝐼0 (𝑒 𝑛𝑉𝑇 , 1) , , 𝑅𝑠ℎ ≫ 𝑉𝑂𝐶
𝑅𝑠ℎ
𝐼𝑃 + 𝐼0
𝑉𝑂𝐶 = 𝑛𝑉𝑇 ln ( ) , , , , , , , ,(3)
𝐼0

Thus from the above equation (2) we can conclude that open circuit voltage is logarithmically
proportional to irradiation.

I
Increasing solar radiation
ISC

VOC V

Fig 2.3. I-V characteristics of a PV cell

13 | P a g e
The specifications of the solar module used in this project is shown in table 2.1.

S.No. Parameter Value

1 Parallel string 1
2 Series-connected modules per string 1
3 Maximum Power (W) 219.978
4 Open circuit voltage Voc (V) 36.72
5 Short-circuit current Isc (A) 7.98
6 Voltage at maximum power point Vmp 30.3
(V)
7 Current at maximum power point Imp (A) 7.26
8 Temperature coefficient of Voc (%/deg.C) -0.35 Fig 2.4. PV array Block diagram in
9 Temperature coefficient of Isc (%/deg.C) 0.05 SIMULINK

Table 2.1 Specification of solar module.

2.2 DC-DC CONVERTER

For operation and control of electric systems we need different voltage levels or even
sometimes requires conversion from AC to DC as per power requirement. These conversion
and control over voltage can be achieved using different power converters. We will be requiring
dc-dc buck-boost unidirectional and bidirectional converters for maximum power point
tracking of solar cells and for controlling charging and discharging of battery.

2.2.1 DC-DC Buck-Boost Unidirectional Converter:

(a) For 0<t< DT, switch S is ON,
Diode D is OFF

𝐼𝐶 = 𝐼0
𝑉𝐿 = 𝑉𝑃𝑉

(a)
(b) For 0<t<DT, switch S is OFF,
Diode D is ON
𝐼𝐶 = 𝐼0 , 𝐼𝐿
𝑉𝐿 = ,𝑉0

Applying volt-sec balance,

∫ 𝑉𝐿 . 𝑑𝑡 = 0
(b)
Fig 2.5. Configuration of DC-DC Buck-
𝑉𝑃𝑉 . [𝐷𝑇\ + (,𝑉0 ). [(1 , 𝐷)𝑇\ = 0 Boost Unidirectional Converter in one
cycle
14 | P a g e
𝑉𝑃𝑉 . 𝐷 = 𝑉0 (1 , 𝐷)
𝐷
𝑉0 = 𝑉
1 , 𝐷 𝑃𝑉

0 ≉ 𝐷 ≉ 0.5, 𝑉0 ≉ 𝑉𝑃𝑉 (𝑏𝑢𝑐𝑘 𝑜𝑟 𝑠𝑡𝑒𝑝 𝑑𝑜𝑤𝑛 𝑚𝑜𝑑𝑒)

0.5 ≉ 𝐷 ≉ 1, 𝑉0 ≊ 𝑉𝑃𝑉 (𝑏𝑜𝑜𝑠𝑡 𝑜𝑟 𝑠𝑡𝑒𝑝 𝑢𝑝 𝑚𝑜𝑑𝑒)

Design of Converter: The specification of converter is shown in following table 2.2

Description Parameter
Input voltage range VI(min) = 480V, VIM =
545.4V; VI=512.7V
Output voltage VO=2.5kV
Forward voltage drop of VF = 0.7V
the rectifier diode
Input voltage ripple VI(PP) = 5% of VIM = 27.27V
Output voltage ripple VO(PP) = 0.5% of VO = 125 V
Equivalent Series ESRCI = ESRCO = 10 mΩ
Resistor (ESR)
Switching frequency of f(SW) = 25 kHz Fig 2.6. Design of DC-DC
the internal switch S1 unidirectional Buck-Boost Converter
Inductor Current Ripple 15 A

Table 2.2 Specification of converter.

Maximum Duty Cycle D = 0.943
Inductance L = 1.28 mH
Minimum Input capacitance C1 = 0.8 mF
Minimum Output capacitance C2 = 0.3 uF

2.3 Maximum Power Point Tracking

As the solar radiation and ambient temperature changes the maximum power generated by solar
cell also varies with it. So to extract the maximum power from PV panel a controller is used
that implement the concept of maximum power point tracking. PV produces maximum power
at a voltage less than Voc. The Connection diagram of PV array with load and PV
characteristics is shown in the following figures.

15 | P a g e
 PV Array DC-DC R0
Converter

RT
(a)

1,𝐷 2
𝑅𝑇 = 𝑅0 ( ) Where, RT is impedance seen from PV array, Pm is the
𝐷 (b)
peak power, ISC is the short circuit current, VOC is the
open circuit voltage

Fig 2.7. (a) Connection diagram of PV array to load with DC-DC converter,
(b) I-V and power characteristics with load line passing through mpp.

From the above figure it is clear that to extract the maximum power we have to operate the PV
array at peak power point. Now, since the load connected to the PV array is variable or which
depends upon the consumers. Also, the I-V and power characteristics change with solar
irradiation and temperature. So we have to design a controller which adjust load line such that
it always passes through the peak power point. This can be achieved by using DC-DC buck-
boost converter with an effective duty cycle control algorithm. One of the simplest and most
effective algorithms is the Perturb and Observation (P&O) method. In this method we cause a
small perturbation in the voltage and changes in power is continuously monitored. If the
operating point of PV lies on left side of the MPP increment in voltage causes increase in
power. On the other hand if operating point lies on right side of MPP, power drops with increase
in voltage. The following figure 2.8, 2.9 illustrate this technique.

Fig 2.9 Duty cycle control of a PV

Fig 2.8. Flowchart of P&O based MPPT controller cell around the MPP.
16 | P a g e
Chapter 3
Solution Approach
To implement the MPPT with the model predictive controller, first we have to design a model
that represent the PV system. We have used the Nonlinear ARX model with artificial neural
network as output function to model the PV system. Later this NARX model has been used
with Model Predictive Controller to track the MPP of the PV system. In the following section
we have discussed about the System Identification and Model Predictive Control strategy in
details.

3.1 System Identification using Artificial Neural Network

System Identification is a methodology to develop the mathematical models of dynamics
systems using measured input-output signals of the system. The process of system
Identification starts with generating the input and output signal from the system. Then select a
model structure among the neural state space model, NARX or Hammerstein-Wiener Model
along with neural network and fuzzy logic model etc. Then apply the estimation method to
develop the model using the collected data. And finally evaluate the estimated model with
different data set.

We have used the following PV system shown in Fig 3.1, to generate data for modeling. So
inputs for model are irradiance, temperature and duty cycle. The meteorological data,
irradiance and temperature are collected from NASA Power website. We have presented these
data in step into the PV system. And a random generator to generate duty cycle for DC-DC
converter. The output power collected from the PV system is being used as a response for
modeling the system.

Fig 3.1 Simulink Plant used for System Identification

17 | P a g e
The following fig 3.2 shows the dataset generated in MATLAB using above Solar PV system.

data Sampling
period (s)
Duty cycle 0.024
Output 0.024
power (watt)
Irradiance 4
(watt/m2)
Temperature 4
(deg C)
Table 3.1 Sampling
Period of Input/output
dataset of Model
Fig 3.2 input dataset – Irr, temp, dutyCycle; output dataset – power
for System Identification generated in MATLAB

We have used the Nonlinear Auto-Regressive with eXogenous inputs (NARX) model to
identify the nonlinear behavior of the PV system. Because this model can identify the complex
nonlinear behavior using nonlinear functions, such as wavelet, sigmoid or neural networks in
addition to the Offset and Linear Function as mapping function in the model. The structure of
NARX model is shown in the following fig 3.2

Fig 3.3 Block diagram of NARX model

A nonlinear ARX model consists of model regressors and an output functions. Regressors is
set of delayed input and output which is used as variable in output mapping functions. It
provides flexibility to add physical knowledge of system in the modeling as a custom
regressors in addition to linear and polynomial regressors. The output function in NARX
model provides linear and Non-linear mapping function along some dc offset. In our model,
we have used Multi-layer feedforward neural network as nonlinear mapping function. The
regressors being used in the modeling are shown in following table 3.2
Linear Regressors Polynomial Regressors
Irr Irr(t-1), Irr(t-2) Irr(t-1)^2, Irr(t-2)^2
temp temp(t-1), temp(t-2) temp(t-1)^2, temp(t-2)^2
Duty cycle dutyCycle(t-1), dutyCycle(t-2) dutyCycle(t-1)^2, dutyCycle(t-2)^2

18 | P a g e
 Output PowerOutput(t-1), PowerOutput(t-2) PowerOutput(t-1)^2, PowerOutput(t-2)^2
power
Mixed PowerOutput(t-1)*PowerOutput(t-2), Irr(t-1)*Irr(t-2), temp(t-1)*temp(t-2),
Regressors dutyCycle(t-1)*dutyCycle(t-2)

Table 3.2 Linear and Polynomial Regressors set in NARX Model

Along with these regressors, to improve the fitting of our model we have added physical
knowledge of the PV system as custom regressors. Practically, the peak output voltage of the
PV panel depends on temperature coefficient (KV) of open circuit voltage (VOC), logarithm of
normalized irradiation and peak voltage at STC (Vmp) [table 2.1]. The peak output current
depends on current coefficient (KI) of short circuit current (ISC), irradiation and peak current
of PV panel at STC (Imp) [table 2.1]. Since the peak output power is product of Vmp and Imp so
all these parameters will also affects the power produced by PV panel. The following table
3.3 shows the custom regressors being used in modeling of PV system.
kI*(Irr(t-1)/Gs)
(2*kV*kI)*(Irr(t-1)/Gs).*(temp(t-1)-Ts),
Custom kI*(Irr(t-1)/Gs).*log(Irr(t-1)/Gs),
Regressors kI*((temp(t-1)-Ts)/Gs),
(kV*kI/Gs)*(temp(t-1)-Ts).^2
(kI/Gs)*(temp(t-1)-Ts).*(1+log(Irr(t-1)/Gs))
Table 3.3 Custom Regressors set in NARX Model

We have used MATLAB System Identification Toolbox for developing the PV plant model.

3.2 Model Predictive Control

The model developed through System Identification process is used internally by MPC to track
the reference maximum power of PV panel. MPC predict the future behavior of the plant using
identified model over a finite time horizon and compute the optimal control input, while
satisfying the given system constraints and minimizes errors between reference and actual plant
output. To be more precise, the duty cycle of DC-DC converter will be calculated by solving
at each sampling instant a finite horizon open-loop optimal control problem; the first part of
the resulting optimal duty cycle signal is applied to the plant until the next sampling instant, at
which the horizon is shifted and the whole procedure is repeated again. The following the
shows the block diagram of MPC with controlled plant.

Fig 3.4 Block diagram of MPC Control Techniques

19 | P a g e
We have used the Model Predictive Control Toolbox in MATLAB to design the MPC
Controller for PV system. For choosing proper values for designing MPC we must need to
justify about the parameters which are affecting the performance of the controller and the
complexity of computation for MPC at each step of time that solves problem of online
optimization . The following figure 3.5 shows the MPC control action tracking reference r.

Fig 3.5 MPC control action for tracking reference r

Design Parameters include – controller sampling time, prediction horizon, control horizon,
constraints and weights. If the controller sampling time is chosen too big it won’t be able to
react to disturbance fast enough and if it is too small the controller can react faster to
disturbances and causes excessive computational load. We have to choose sampling time to be
0.024 sec for open loop response so that within the rise time 10 samples can be fitted . Here
in the above figure 3.5, k is the present time step. At each time step k, MPC predict the future
output of PV plant (here control variable y will be the Power output), for prediction horizon
N2. The prediction horizon is chosen such a way that it covers the significant dynamics of the
system without increasing unexpected computational complexity. We have chosen the
prediction horizon to be 10 time samples considering satisfactory plant response to step input.
The number of control action leading to the future plant output to follow the reference is Nu.
Increasing the control horizon improves the prediction but cost of complexity increases. We
have chosen prediction horizon value to 4 time sample considering its step response. Lastly the
design parameters constraints and weights of input and output of the plant is chosen not to
violate the physical constraint. Duty cycle range is between 0 and 1 and the maximum output
power is constrained between 0 and1000 watt considering power of a single solar module/panel
can’t exceed 1000 watt. Since the tracking the reference power is the main objective, so the
weights of the controlled plant output is set to 1.

20 | P a g e
Chapter 4
Simulation Result and Discussion
We have used System Identification tool box in SIMULINK/MATLAB to model the PV
system dynamics using Nonlinear ARX model. We have used the previous data sets collected
from Simulink PV system – irradiation, temperature and duty cycle as input and power as
output of the plant. The specification and design of PV module and Converter used in Simulink
Plant is discussed in previous section 2.1, 2.2. We first start the identification by using linear
ARX model. The following figure 4.1shows the simple ARX model response and validation
results without any custom regressors. The regressors being used are linear, polynomial and
mixed discussed in section 3.1. The Fit to Estimation and validation is very low, only 40-45%
and Mean Square Error is 215.9.

Fig 4.1 Linear ARX Model response without any custom regressors

The following figure 4.2 shows the model response when we have added wavelet function in
NARX model. It has been observed that use of wavelet function as a nonlinear mapping
function in NARX model has not given any improvement in fitting as well as validation of
model. It is still around 45%.

Fig 4.2 NARX Model response with wavelet as nonlinear function

21 | P a g e
To improve the fitting we added custom regressors to the model. The following figure 4.3,
shows the fitting results of different model where we have used different combination artificial
neural network as nonlinear mapping functions. The fitting got significant improvement to
nearly 75-80%. The nlarxModel2 (shown in the figure 4.3) gives highest accuracy to actual
output. The size of Multi-Layer Network used in nlarxModel2 is [3, 15, 15, 1]. Activation
function being used are: [“relu”, “sigmoid”]. We have used fmincon (sqp) nonlinear optimizer
to estimate the model. The max iteration of optimization is 50.

Fig 4.3 NARX Models response with different no. of neural network

We have more improved the model response by hit & trial method using different sizes of
neuron and activation functions. The following figure 4.4 (a), (b) shows the model response
when we have used “relu” activation function in both the hidden layer. The size of hidden layer
is [10 10]. The fitting got improved upto 90% and MSE reduces to 5.4 (in nlarxModel9).

(a)

(b)

Fig 4.4 NARX Model response with ‘relu’ activation in both the hidden layer,
neuron size (a) [10 10], (b) [10 15].

22 | P a g e
We have used this model with MPC controller for MPP tracking of solar module. The following
figure 4.5 shows MATLAB function generated irradiance and temperature plot, where
irradiance bound is 200-1200 watt/m2 and temperature is varied between 15-35 oC.

Fig 4.5 MATLAB generated dataset of Irradiance and temperature

The fig 4.6 shows the practical hourly irradiance and temperature plot of NIT Silchar Campus,
(varied in step of 4sec) collected from the NASA Power site. We have used both these data set
to evaluate the performance of the model.

Fig 4.6 Practical dataset of Irradiance and temperature

23 | P a g e
The following figure 4.7, 4.8 shows current and voltage output of PV system based on P&O
algorithm. It is observed that voltage and current output is not smooth, it oscillate back and
forth continuously.

Fig 4.7 P&O based plant V-I output for MATLAB generated dataset

Fig 4.8 P&O based plant V-I output for practical dataset

24 | P a g e
The following fig 4.9 shows the power output of PV system based on P&O algorithm for
MATLAB generated data and practical meteorological data consecutively.

(a)

(b)

Fig 4.9 P&O based plant output power for (a) MATLAB generated dataset, (b) practical dataset

We have P&O Result

observed that the dataSet 1
Irr Temp
Peak Trough ref Peak-ref (peak-ref)^2 trough-ref (trough-ref)^2
(watt/m2) (deg C)
instantaneous 200 22 43.1 33 40.3 2.8 7.84 -7.3 53.29
400 24 87.8 24.2 70.7 17.1 292.41 -46.5 2162.25
power tracked 1200 32 255.1 140.8 225 30.1 906.01 -84.2 7089.64
MATLAB 800 27 41 35.5 40.7 0.3 0.09 -5.2 27.04
through P&O GENERATED DATA 600 23 174.7 41.2 137.3 37.4 1398.76 -96.1 9235.21
algorithm 300 19 66.8 31.4 54.5 12.3 151.29 -23.1 533.61
Total SUM: 100 2756.4 -262.4 19101.04
oscillates about std deviation: 23.4793526 61.80783122

the reference dataSet 2

Irr Temp
Peak Trough ref Peak-ref (peak-ref)^2 trough-ref (trough-ref)^2
(watt/m2) (deg C)
maximum 0 11.43 0 0 0 0 0 0 0
0 10.49 68.4 12.6 37.1 31.3 979.69 -24.5 600.25
power. Table 304.28 16.61 68.4 31.4 37.1 31.3 979.69 -5.7 32.49
PRATICAL
4.1, shows peak METEROLOGICAL 695.6 23.32 154.3 27.1 115.8 38.5 1482.25 -88.7 7867.69
99.06 20.54 21 20 20.6 0.4 0.16 -0.6 0.36
DATA
and trough 0 15.24 0 0 0 0 0 0 0
Total SUM: 101.5 3441.79 -119.5 8500.79
power limit for std deviation: 26.2365775 41.23297224

each irradiation
Table 4.1 P&O based plant output analysis for both dataset (all unit are in watt)
and temperature
values of each data set. For e.g., irradiation and temperature value of 400 watt and 24oC, the
instantaneous power oscillates between 87.8 – 24.2 watt about the reference value 70.7. It is
observed that the standard deviation of peak and trough instantaneous power is 23.48 watt and

25 | P a g e
61.80 watt for MATLAB-generated data. For Practical meteorological data, standard deviation
of peak and trough instantaneous power from reference is 26.36 watt and 41.23 watt.
We have used the Model Predictive Control tool box for developing MPC controller for the PV
system. The prediction and control horizon of MPC is 10 and 4 sample time step consecutively.
The period of each sample is 0.024 second. The following fig 4.10, 4.11 shows the power
output of PV plant controlled by MPC controller for both the dataset.

We have observed that, MPC controlled system output doesn’t contain any oscillation while
tracking the reference power. There is a little overshoot in system output due to controller input
into the converter. Now the system can track the reference power more accurately as compared
to the P&O-based system, as it is seen from the above plot that, system output is much closed
to the reference maximum power.

Fig 4.10 MPC controlled Plant Output for dataset 1

Fig 4.11 MPC controlled Plant Output for dataset 2

The table 4.2 shows the response of above MPC controlled PV plot. It is seen from the table
that system response becomes faster with a small delay of 0.04 sec and average settling time
0.34 sec for MATLAB generated dataset. There is a little overshoot or undershoot in the
26 | P a g e
response. For e.g. irradiance and temperature value of 400 watt and 24 oC, its peak overshoot
is 8.06% with settling time 0.16 sec and the steady state error is 1.1 watt. The delay of the
system is 0.04 sec. For MATLAB generated dataset, the steady state average error is around
1.166 watt and average steady state percentage error is 1.20%.
For practical meteorological dataset, the average settling time is 0.50 sec, average percentage
error is 1.55% and average overshoot and undershoot is 14.39% and 7.12% respectively.

MPC Result
Percentage
Steady State Overshoot Settling Steady state
dataSet 1 Irr (watt/m2) Temp (deg C) Peak Trough ref Undershoot Delay time (s) Steady state
value (%) time(s) error
error(%)
200 22 40.9 6.8 40.7 40.3 1.49 0 0.04 0.8 0.4 0.99
400 24 76.4 N/A 71.8 70.7 8.06 N/A 0.04 0.16 1.1 1.56
1200 32 237.9 N/A 228.3 225 5.73 N/A 0.04 0.56 3.3 1.47
MATLAB
800 27 N/A N/A 136.4 137.3 N/A N/A 0.04 0.2 0.9 0.66
GENERATED
600 23 N/A 37.9 41 40.7 N/A 2.8 0.04 0.12 0.3 0.74
DATA
300 19 N/A N/A 53.5 54.5 N/A N/A 0.04 0.2 1 1.83
avg: 5.09 1.40 0.04 0.34 1.17 1.21

Percentage
Steady State Overshoot Settling Steady state
dataSet 2 Irr (watt/m2) Temp (deg C) Peak Trough ref Undershoot Delay time (s) Steady state
value (%) time(s) error
error(%)
0 11.43 28.3 -20.2 0 0 N/A N/A 0.04 0.24 0 0.00
0 10.49 0.1 -1 0 0 N/A N/A 0.04 0.32 0 0.00
304.28 16.61 43.7 33.8 35.7 37.1 17.79 8.89 0.04 0.16 1.4 3.77
PRATICAL
695.6 23.32 127.2 N/A 120.5 115.8 9.84 N/A 0.04 0.24 4.7 4.06
METEROLOGI
99.06 20.54 23.8 19.5 20.9 20.6 15.53 5.34 0.04 0.48 0.3 1.46
CAL DATA
0 15.24 34.9 -116 0 0 N/A N/A 0.04 1.56 0 0.00
avg: 14.39 7.12 0.04 0.50 1.07 1.55

Table 4.2 MPC based plant output analysis for both dataset

27 | P a g e
Chapter 5
Conclusion
In this project, we develop a NARX model that represent the PV system dynamics which is
used by MPC controller to track the maximum power for varying irradiance and temperature.
We have used System Identification tool box and Model Predictive Control tool box in
MATLAB. The system Identification result has been improved from 45% to 90%. The model
has been tested for MATLAB generated Irradiance and temperature data as well as Practical
meteorological dataset collected from NASA Power sit. We have observed that voltage and
current output of P&O based PV system contain lots of oscillation. The instantaneous power
oscillates around the reference maximum power. The standard deviation of output power is 26-
60 watt. On the other hand MPC controlled plant output is faster and more accurately track the
reference power. The steady state error is only 1-2%. The plant output gets settled near
reference power (almost 98% of maximum power) within 0.3-0.5 sec with average delay of 0.5
second. Thus MPC controlled PV system improved the speed and accuracy the Maximum
Power Point Tracking and removed unnecessary oscillation around the referenced Maximum
Power.

28 | P a g e
References

1. Xuesong Zhou, Tie Guo, Youjie Ma. An Overview on Microgrid Technology. IEEE Int
Conf on Mechatronics and Automation(2015): 76-81
2. Oussama, Elhoussian, Y Amirat, E S Gooya. Optimal Sizing and Energy Management
of Microgrids with Vehicle-to-Grid Technology: A critical Review and Future Trends.
MDPI Energies 2021,14,4166: 1-35
3. S. Monesha, S. Ganesh Kumar, M. Rivera. Microgrid Energy Management and
Control: Technical Review. IEEE ints Conf on Automatica(2016): 1-6
4. Z Ullah, S Wang, J Lai, M Azam, F Badshah, G Wu, M R Elkadeem. Implementation
of various control methods for the efficient energy management in hybrid microgrid
system. Ain Shams Engineering Journal(2022): 1-16
5. Y Boujoudar, M Azeroual, H E Moussaoui, T Lamhamdi. Intelligent Controller based
energy management for stand-alone power system using artificial neural network. Int
Trans on Electrical Energy Systems(2020):1-12
6. Samer Said, Ahmed Massoud, M Benammer, Shehab Ahmed. A matlab/simulink based
photovoltaic array model employing simpowersystems toolbox. Journal of Power and
Energy Engineering(2012): 1965-1975
7. Zhixin Miao, Ling Xu, Lingling Fan, V R. Disfani. An SOC-Based Battery
Management system for Microgrids. IEEE trans on Smart Grid, Vol 5, 2014: 966-973
8. Hanane Hemi, N K M’Sirdi, A Naamane. A new proposed shepherd Model of a Li-ion
Open circuit Battery Based on Data Fitting. IMAACA 2019: 1-9
9. Simon Ellefors. Implementation of semi-empirical, electrochemistry-based Li-ion
battery model for discharge characterization. Bachelor of Science Thesis in Electrical
Eng., Linkoping University, 2018: 1-27
10. Oliver Tremblay, Louis-A. Dessaint. Experimental Validation of a Battery Dynamic
Model for EV Applications. World Electric Vehicle Journal Vol.3 (2009): 0289-0298
11. Daniel C Zacharek, Filip Sundqvist. Design of Bidirectional DC/DC battery
Management system for Electrical Yacht. Bachelor of Science Thesis in Electrical Eng.,
Linkoping University, 2018: 12-26
12. Alexander Walberg. Design and construction of bidirectional DC/DC converter.
Uppsala Universitet (2019):4-13
13. S Salman, Xin AI, Z WU. Design of a P-&-O algorithm-based MPPT charge controller
for a stand-alone 200w PV system. Protection and Control of Modern Power systems
3,25(2018): 2-7
14. Narendra KS, Mukhopadhyay S. Adaptive control using neural networks and
approximate models. IEEE Trans Neural Network. 1997;8(3): 475-485
15. Mohammed A Alkhafaji, Yunus Uzun. Modeling and Simulation of A Photovoltaic
Cell Module Controlled with Nonlinear Autoregressive Moving Average-L2
Controller. Int Conf on Electrical, Communication, and Computer Eng. (2020): 1-5

29 | P a g e
16. Adel Akbarimajd, Behrooz Sobhani, H Shayeghi. Neural Network Based NARMA
Controller for a DG in Islanding Mode. Int Conf on Intelligent Systems Design and
Applications.(2011): 1-4
17. Max Schwnezer, Muzaffer Ay, Thomas Berges, Dirk Abel. Review on model predictive
control: an engineering perspective. Int Journal of Advanced Manufacturing
Technology (2021)117:1327–1349.
18. Oliver Nelles. Nonlinear System Identification From Classical Approaches to Neural
Networks, Fuzzy Models, and Gaussian Processes (Second Edition)
19. Akira Ohata, Katsuhisa Furuta and Hiroaki Nita. Identification of Nonlinear ARX
Model with Input and Output Dependent Coefficients. IEEE Int Conf on Control
Applications Munich, Germany, Oct 4-6, 2006.
20. Yuya HASUIKE, Masaki IZUTSU, and Shosiro HATAKEYAMA. Identification of
Nonlinear ARX Model with Input and Output Dependent Coefficients. 39th Annual
Conf of the IEEE Industrial Electronics Society: 2013
21. Feroz Shaik, Syam S Lingala, Punnaiah Veeraboina. Effect of various parameters on
the performance of solar PV power plant: a review and the experimental study.
Sustainable Energy Research, Springer Open Access.
22. Morcos M, Sertac B, Robert S. Balog, Haitham Abu Rub. Model Predictive Control for
PV Maximum Power Point Tracking of Single-Phase subMultilevel Inverter. 2016
IEEE Power and Energy Conf 2016.
23. Yingying Zhao, Aimin An, Yifan Xu, Q Wang and Minmin Wang. Model predictive
control of grid-connected PV power generation system considering optimal MPPT
control of PV modules. Prot Control Mod Power Syst (2021) 6:32
24. Rohan S. Kulkarni, Dhananjay B. Talange. Modeling of Solar Photovoltaic module
using system identification. 2017 7th Int Conf on Power Syst: 1-3

30 | P a g e