Computation of an Effective Hybrid DFA-SVM Approach Aimed at

Adaptive PV Power Management

A. R. DANILA SHIRLY1, M. V. SUGANYADEVI2, R. RAMYA3,

I ARUL DOSS ADAIKALAM4, P. MUTHUKUMAR5

1Department of Electrical and Electronics Engineering,

Loyola-ICAM College of Engineering and Technology,

Chennai,

INDIA

2Department of Electrical and Electronics Engineering,

Saranathan College of Engineering,

Trichy,

INDIA

3Department of Electrical and Electronics Engineering,

SRM Institute of Science and Technology,

Chennai,

INDIA

4Department of Electrical and Electronics Engineering,

Chennai Institute of Technology,

Chennai,

INDIA

5Department of Electrical and Electronics Engineering,

Saveetha School of Engineering,

Chennai,

INDIA

Abstract: - Predominantly focussed in environmental conditions that are dynamic in nature the energy

harnessed from the photovoltaic systems has to be maintained at high efficiency for which maximum power has

to be extracted so a novel hybrid DFA-SVM control has been implemented using SEPIC converter. There are

many algorithms to perform this function mentioned but in order to track the power at a faster rate and to avoid

oscillations at the settling peak point this new methodology has been implemented. In this paper the novel

algorithm used to track the peak power is Dragon Fly Algorithm-Support Vector Machines (SVMs). The

algorithm is a combination of optimization and machine learning technique, so that this new methodology can

incorporate both instantaneous and steady state features. The benefits of both the optimization and supervised

learning technique are used to track most efficiently the maximum power with less oscillations. The DFA-SVM

technique is implemented in the controller of the DC-DC converter used to regulate the supply voltage

generated by the PV. The suggested MPPT's performance is demonstrated under demanding experimental

conditions including temperature and solar irradiation fluctuations across the panel. To further illustrate the

superiority of the suggested approach, its performance is contrasted with that of the P&O method, which is

commonly employed in MPPT during difficult exams.

Key-Words: - Dragon Fly Algorithm, Support Vector Machine, Maximum Power Point Tracking, Photo

Voltaic, Single-Ended Primary Inductance Converter (SEPIC), DC-DC Converter, Perturb &

Observe MPPT Algorithm.

Received: May 9, 2023. Revised: May 11, 2024. Accepted: June 18, 2024. Published: July 30, 2024.

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

276

Volume 19, 2024

1 Introduction

Generating power with photovoltaics is a type of

energy that is renewable in nature with several

benefits. Its inherent attributes that make it easy to

incorporate into domestic micro-grids set it apart

from other solar energy sources. Although there

have been tremendous advancements in PV systems,

including lower costs, increased cell efficiency, and

improved building structural integration, one major

drawback remains their low energy conversion

efficiency, [1]. Conversely, the natural

environment—which includes temperature and solar

irradiation—affects how much energy photovoltaic

devices generate. Therefore, in order for SEPIC

converters to extract the maximum quantity of

energy produced by PV modules, a control linked to

maximum power point tracking should offer a

suitable control to the switch of the converter used

to regulate the PV power. There have been

numerous studies on MPPT approaches; the most

well-known ones include the incremental

conductance algorithm (INC), extremum seeking

control methods (ESC), and perturb and observe

(P&O) techniques. These techniques typically use

the PV module's immediate resultant either current

or voltage in order to produce signals for control,

tracking the MPP using a duty ratio, reference

voltage, or reference current, [2]. The minimal

computing cost and straightforward structure of the

P&O approach are its advantages. The surveys give

the idea algorithms rely on calculating the MPP

using data that includes past temperature or

radiation readings. Among these are machine

learning-based systems and optimisation techniques.

To boost the PV system's efficiency, the

following conditions must be satisfied: an integrated

and configured MPPT algorithm and an appropriate

DC-DC converter.

• Quick tracking reaction.

• There are no fluctuations near the steady-state

response, or MPP.

• Reaction time to temperature variations and sun

radiation.

• Easy structure that requires less calculation.

A number of studies pertaining to support vector

machines (SVMs) and MPPT algorithms have been

published. Machine learning techniques like Support

Vector Machines (SVM) are utilized for linear

regression and problem classification. ANN was

trained using a vast quantity of training data

produced by the SVM classifier. It is difficult to

implement the suggested method on budget-friendly

processors since it uses a typical support vector

regression to estimate radiation concentrations via

Solar and necessitates input features that are not

consistently and individually generated, [3]. Using

the traditional approach, the sized perturbation step

was calculated for two distinct sites utilizing a solar

irradiance assessment method created using SVM.

However, the impacts of partial shading are not

taken into account because for a particular location,

the impacted dimension of steps selection is

achieved offline and updated either monthly or

seasonally. This paper presents a n ew DFA-SVM-

based MPPT method. The unknown transfer

function of a PV module's multivariable nonlinear

P-V characteristics is estimated using the DFA-

SVM. This issue is referred to as r egression

estimation for numerous variables in machine

learning. The capacity of determining variables

using fewer parameters, provide reliable forecasts

even in the presence of noise and nonlinearities in

the framework, and build each regressor by taking

into account all of the inputs and outputs combined

are the primary advantages of utilizing the DFA-

SVM based approach

2 Proposed System

This work proposes a novel approach that removes

irregularities about the MPP in a constant state when

computing a PV module's MPP using numerous

inputs to a single-output control. For training the

SVM , the Dragon Fly Algorithm provides a

heuristic technique that is significantly more

successful. The SVM may not be trained as

successfully because it only relies on the algorithm's

capacity to monitor MPP. However, when DFA-

SVM is employed for MPP monitoring, it enables a

substantially faster rate of convergence. Therefore,

the DFA technique is included to avoid all of the

aforementioned shortcomings, [4]. It is possible to

incorporate the suggested MPPT algorithm with the

DC-DC SEPIC converter's double loop control in a

low-cost commercial DSC. Through simulations,

the suggested method's performance has been

confirmed, demonstrating good repeatability and

precision. The MPPT control schematic for a

photovoltaic module uses a SEPIC converter to

charge the load shown in Figure 1. The module uses

voltage and current to perform conversion of solar

photons into electrical energy, charging the load.

The P-V & I-V curve for the panel is shown for

various irradiance & surrounding temperatures. The

SEPIC converter architecture is selected for use as a

DC-DC converter in this work. Models are included

for various irradiation conditions shown in Figure 2.

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

277

Volume 19, 2024

Fig. 1: An overview of the proposed DFA-SVM

MPPT scheme

Fig. 2: IV and PV Characteristics of the PV panel

3 DC-DC Converter

A SEPIC (Single Ended Primary Inductor

Converter) is the type of DC-DC converter. It is

possible to effectively convert a DC voltage to a

different DC voltage level using a switching power

supply, or SEPIC. This specific type of buck-boost

converter may adjust the input voltage to produce

the desired output voltage by stepping it up or down.

An inductor, a capacitor, and two switches—

typically MOSFETs—connected in a certain

arrangement make up the SEPIC converter

topology, as seen in Figure 3. The inductor and the

output capacitor are both charged via a diode when

the switch linked to the input voltage is switched on.

The regulated output expected from converter is

entirely governed by the controller which controls

the signal set to the MOSFET. The regulated output

can be boosted or reduced by the SEPIC converter

by controlling the pulse signal given. Moreover

SEPIC gives a perfect isolation even without using

an isolation transformer for the source and load.

which is an additional benefit for using this

converter, [5]. As a result this converter design is s

desirable one, as it delivers the required output in

numerous conditions. The list of components used

along with its specifications for the design of SEPIC

converter is shown in Table 1.

When the MOSFET is kept in ON condition, the

first inductor receives power from the PV supply.

The other inductor absorbs energy from C1 or the

linking capacitor, while the output capacitor remains

to supply the load. When the MOSFET is kept

activated, the first capacitor charges the second

inductor, and the supply charges the input inductor.

During this period, the load capacitor receives no

energy.

Fig. 3: SEPIC Converter Topology

The energy contained in the inductor is

transferred to when the power switch is switched

off. The diode facilitates the transmission of energy

from the stored energy, providing the load with

energy, [6]. At this point, the load is also linked to

the second inductor. Because it experiences a

current pulse during the off-time, the output

capacitor is by nature noisier than a buck converter.

The duty cycle L & C in the circuit are the main

factors that determine the extent to which the SEPIC

converters raise or lower the voltage.

Table 1.Converter Components Values

Parameter

Values

Inductor L1

25 

Capacitor C1

10 

Resistor

1000 Ω

Duty Cycle

61.54 %

Inductor L2

380 m

Capacitor C2

47 

4 MPPT Algorithm Techniques

The aim for maximising the efficiency and power

extracted from PV, the maximum power point

tracking regulator is used to track voltage, current to

maximise power output extracted from the panel,

[7].

4.1 Perturb and Observe Algorithm

The most predominant tracking method for PV

panel that tracks MPP is the perturb and observe

algorithm, [8]. This technique finds the peak power

where the maximum power output can be achieved

by changing the panel's voltage and current.

The technique relies on frequent monitoring and

calculation of the power. The panel parameters are

regularly increased or decreased by the tracker. If

one change leads to an increase in the output, the

next one is generated in the similar direction, [9].

The operation is repeated while varying the duty

cycle of the SEPIC until the maximum power point

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

278

Volume 19, 2024

is attained. The perturb and observe algorithm's

fundamental steps are as follows, with the flow

chart displayed in Figure 4:

1. Check the PV panel's voltage and current.

2. Gradually raise the voltage and gauge the panel's

power output.

3. If the power output has increased, measure the

power output at each step while keeping the

voltage increased in the same direction.

4. Reverse the direction of the voltage adjustment

and repeat step 3 i f the power output has

dropped.

5. Take note of the electrical voltage and current

values once the power point's highest value is

attained and use them as the panel's set point.

6. To guarantee that the panel is always running at

its maximum power point, repeat the procedure

on a regular basis.

The limitations of this traditional method

include its inability to follow MPP during

significant changes in the insolation level and its on-

going oscillation around MPP. But the advantages

of this approach are its low cost, easy

implementation, and straightforward structure.

Fig. 4: Perturb & Observe MPPT Algorithm

4.2 The Proposed DFA-SVM MPPT

Technique

This paper presents a heuristic technique for training

the SVM using the Dragon Fly Algorithm, which is

significantly more successful. Machine Learning

Techniques always yields results which is far better

than the traditional methods [10], [11], [12], [13],

[14]. The SVM alone may not be trained as

successfully because it only relies on the SVM and

the algorithm's capacity to monitor MPP. However,

when DFA & SVM are employed for MPP

monitoring, it enables a substantially faster rate of

convergence.

Therefore, the DFA technique is included to

avoid all of the aforementioned shortcomings. One

limitation of the SVM is that training data that is not

sufficiently exact and optimized can lead to

improper training of the SVM. Different

environmental conditions could lead to misbehavior

from the structure and system malfunctions, which

would make the system, operate inefficiently. DFA

was chosen in this work because it monitors data

faster, is simpler, and has a greater rate of

convergence than other metaheuristic techniques.

Furthermore, the SVM is included since it is

considerably simpler to build and can track the

MPP, preventing oscillations and variations in

power during the first stage, and has a better grasp

of the system with the right training. In order to

collect the dataset with the voltage, current, and

duty cycle parameter that correlate to the system

performance with regard to various scenarios, the

DFA stimulates the PV system under a variety of

climatic situations. The way SVM operates is by

identifying the hyperplane with the greatest margin

that divides the two classes of data points. Stated

differently, it s eeks to identify the optimal line or

plane that divides the two classes in order to

maximize the distance between the line/plane and

the nearest data points for each class. The kernel

trick is a technique that allows the SVM algorithm

to handle non-linearly separable data. The input data

is mapped via a kernel trick on a multi-dimensional

sector which can be segregated using hyperplane.

As a result, the data points can be divided using

a non-linear border using SVM. Here, swarm

intelligence-based DFA is used to tackle nonlinear

problems.

Within a swarm, there are two different kinds of

movement: dynamic and static. A static swarm is

made up of a few DF searching for food in a small

region. They can only move in little leaps that

resemble the exploitation of search space. The

acronym of the equations from (1) to (7) is given in

Table 2.

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

279

Volume 19, 2024

Fig. 5: Flow Diagram of the Proposed MPPT

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Table 2.Equation Symbols

Parameter

Acronym

Separation factor of ith DF

Separation factor

Alignment factor

Alignment of ith DF

Cohesiveness factor

Cohesion of ith DF

Food factor

Food attraction

Enemy factor

∆X

Enemy location

Inertial weight

Step size of DF

Thus by implementing DFA the generated

dataset from the optimization algorithm is fed into

the SVM to train the system and modify the

structure's parameters through DFA, which

culminates in the creation of a precise and effective

control strategy, as illustrated on t he following

Figure 5.

5 Outcome and Discussions

The tool MATLAB Simulink is utilised for system

assessment and analysis. The system is tested,

maintained, and contrasted with the use of P&O

MPPT and the hybrid algorithm in order to analyse

and evaluate its performance in a variety of working

scenarios. The Hybrid Dragon Fly and SVM

algorithm, which is interfaced with a SEPIC for

normalizing solar supply and output the same to the

load, allows the PV system to extract MPP under a

variety of climatic conditions.

Fig. 6: Output Voltage and Current measurement

using P&O MPPT Technique

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

280

Volume 19, 2024

Fig. 7: Output Voltage and Current measurement

using DFA-SVM MPPT Technique

The Figure 6 and Figure 7 depicts the output

when P&O and DFA-SVM MPPT methodology are

used respectively. The SEPIC alongwith DFA-SVM

tracker has several added advantages when

compared with the old controllers. It is more

accurate, efficient, and capable of adapting to

change in environmental conditions.

The design and implementation of the system is

complex when compared with traditional

controllers. The use of machine learning algorithms

requires much more computational resources, which

could increase the overall cost of the system, [15].

Overall, the SEPIC converter with hybrid MPPT

controller shows great potential in improving the

performance of solar panel systems. The following

inferences were extracted while analyzing the

system which is shown in Table 3 (Appendix) and

Table 4.

Thus by the implementation of DFA-SVM based

system the following benefits are achieved:

• Maximizing the output of solar PV arrays in a

wide range of environmental conditions

• Faster tracking when compared with other

traditional strategies as hybrid approach has been

proven to settle faster in MPP

• Faster output tracking which shows that the

system can respond more quickly to

environmental conditions and load requirements.

Table 4 Comparison of DFA-SVM and P&O

• A combination of AI based algorithms (such as

SVM) and optimization algorithms (such as

DFA) allows the method to combine the benefits

of both techniques to create a more complete and

effective MPPT strategy

• Enhanced system effectiveness by faster

response time providing tangible benefits for end

users

• Higher yields due to improved reliability which

effects in lower costs over time

The hardware implementation using the hybrid

MPPT was done which is depicted in Figure 8.

Fig. 8: Hardware implementation

6 Conclusion

This paper uses a novel hybrid Dragon Fly

Algorithm-SVM MPPT technique to maximize

output in a range of environmental circumstances.

After a comparison between the hybrid

Conditions

Voltage

Ripple

Power

Efficiency

Settling time

With P&O MPPT

3.11

81.35%

1.0

With DFA-SVM

MPPT

3.0189

89.81%

0.2

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

281

Volume 19, 2024

methodology and the traditional strategy, it was

found that the hybrid approach settled in the MPP

and tracked the output at a faster rate. In order to

enhance the solar PV array's performance, this

method recognizes the benefits of both artificial

intelligence-based algorithms and optimization

algorithms. The DFA-SVM-based MPPT controller

enhances system effectiveness and benefits end

users, with faster response times compared to

traditional P&O algorithms. This work will

significantly impact the development of more

reliable and efficient solar systems for renewable

energy applications, enhancing overall system

efficiency. The future work that can be done is

to refine the hybrid algorithm's performance

metrics (like convergence speed, efficiency,

etc.,) and include more parameters to increase

the robustness of the system over a wide range

of environmental conditions.

References:

[1] Ridha, Hussein Mohammed, Hashim Hizam,

Noorulden Basil, Seyedali Mirjalili,

Mohammad Lutfi Othman, Mohammad

Effendy Ya'acob, and Masoud Ahmadipour.

"Multi-objective and multi-criteria decision

making for Technoeconomic optimum design

of hybrid standalone renewable energy

system." Renewable Energy (2024), Vol. 223,

Issue 12, pp.1-15,

https://doi.org/10.1016/j.renene.2024.120041.

[2] Belghiti, Hamid, Khalid Kandoussi,

Abdelkhalek Chellakhi, Youssef Mchaouar,

Rabie El Otmani, and El Mostafa Sadek.

"Performance optimization of photovoltaic

system under real climatic conditions using a

novel MPPT approach." Energy Sources, Part

A: Recovery, Utilization, and Environmental

Effects, 46, no. 1 (2024), pp.2474-2492.

[3] I G Venkatesh, Chintala, and Sukhi Yesuraj.

"Efficient and Reliable Fast Charging Station

for Electric Vehicles: Integrating PV System

and Optimized Control." Electric Power

Components and Systems (2024), pp.1-21,

https://doi.org/10.1080/15325008.2024.23202

75.

[4] Bhardwaj, Sanjay, Da-Hye Kim, and Dong-

Seong Kim. "Dragonfly Interaction Algorithm

for Optimization of Queuing Delay in

Industrial Wireless Networks." Journal of

Bionic Engineering (2024), 1-39,

Vol. 21, pp.447-485, (2024).

[5] Saxena, Nishant, Rahul Kumar, Yarrapragada

KSS Rao, Dilbag Singh Mondloe, Nishikant

Kishor Dhapekar, Abhishek Sharma, and Anil

Singh Yadav. "Hybrid KNN-SVM machine

learning approach for solar power

forecasting." Environmental Challenges Vol.

14, (2024), 1-20 100838,

https://doi.org/10.1016/j.envc.2024.100838.

[6] Danila Shirly, A.R., Roshini, R., Priyanka, E.,

Sindhuja, M., Steffy Jones, A. (2022). Design

and Implementation of Photovoltaic Powered

SEPIC DC-DC Converter Using Particle

Swarm Optimization (PSO) Technique. In:

Ranganathan, G., Bestak, R., Palanisamy, R.,

Rocha, Á. (eds) Pervasive Computing and

Social Networking. Lecture Notes in

Networks and Systems, Vol. 317, Springer,

pp.478-483 Singapore,

https://doi.org/10.1007/978-981-16-5640-

8_26.

[7] Salah Anis Krim, Fateh Krim, Hamza

Afghoul, Feriel Abdelmalek, "An Improved

Perturb and Observe MPPT for Photovoltaic

Systems using Fuzzy Step Size," WSEAS

Transactions on Power Systems, vol. 19, pp.

105-114, 2024,

https://doi.org/10.37394/232016.2024.19.13.

[8] Zhu, Honglu, Tingting Jiang, Hai Zhou, Yahui

Sun, Wenwen Ma, and Xi Zhang. "A novel

soft sensing method using intelligent

modeling method for solar irradiance and

temperature in distributed PV power

plant." Energy Science & Engineering (2024),

12(3), pp.1041-1058,

https://doi.org/10.1002/ese3.1679.

[9] Bonthagorla, Praveen Kumar, and Suresh

Mikkili. "Hardware implementation of a novel

hybrid MPPT technique for fast tracking of

GMPP in solar PV system under

PSCs." Circuit World, Vol. 50, Issue 1 (2024),

pp.132-148.

[10] Tercha, Wassila, Sid Ahmed Tadjer, Fathia

Chekired, and Laurent Canale. "Machine

Learning-Based Forecasting of Temperature

and Solar Irradiance for Photovoltaic

Systems." Energies, Vol.. 17, Issue 5 (2024),:

1124,

DOI:https://doi.org/10.3390/en17051124.

[11] Hichri, Amal, Mansour Hajji, Majdi

Mansouri, Hazem Nounou, and Kais

Bouzrara. "Supervised machine learning-

based salp swarm algorithm for fault

diagnosis of photovoltaic systems." Journal of

Engineering and Applied Science Vol. 71,

article number 12, (2024), 1-19.

[12] Abdelaal, Ashraf K., Amira IA Alhamahmy,

Hossam El Deen Attia, and Attia A. El-

WSEAS TRANSACTIONS on POWER SYSTEMS

DOI: 10.37394/232016.2024.19.25

A. R. Danila Shirly, M. V. Suganyadevi,

R. Ramya, I Arul Doss Adaikalam, P. Muthukumar

E-ISSN: 2224-350X

282

Volume 19, 2024

Fergany. "Maximizing solar radiations of PV

panels using artificial gorilla troops reinforced

by experimental investigations." Scientific

Reports 14, no. 1 (2024): 3562.

[13] Refaat, Ahmed, Qays Adnan Ali, Mohamed

Mohamed Elsakka, Yasser Elhenawy,

Thokozani Majozi, Nikolay V. Korovkin, and

Medhat Hegazy Elfar. "Extraction of

maximum power from PV system based on

horse herd optimization MPPT technique

under various weather conditions." Renewable

Energy, Vol. 220, Issue1, 1-22. (2024),

https://doi.org/10.1016/j.renene.2023.119718.

[14] Mohamed I. Abu El-Sebah, Aly M. Eissa,

Mohamed Fawzy El-Khatib, "Investigation of

Artificial Intelligence Algorithms for MPPT

of Solar Photovoltaic System," WSEAS

Transactions on Systems and Control, vol. 18,

pp. 561-568, 2023,

https://doi.org/10.37394/23203.2023.18.58.

[15] Pramanick, Dipankar, and Jitendra Kumar.

"Performance and degradation assessment of

two different solar PV cell technologies in the

remote region of eastern India." e-Prime-

Advances in Electrical Engineering,

Electronics and Energy, Vol. 7, (2024),

sp:100432, 1-22,

https://doi.org/10.1016/j.prime.2024.100432.

APPENDIX

Table 3. Comparison of DFA-SVM and P&O

parameters