Next Article in Journal
Pulse-Width Modulation Approaches for Efficient Harmonic Suppression
Previous Article in Journal
A Novel Capacitor-Commutated Converter Based on Submodule-Cascaded STATCOM
Previous Article in Special Issue
Optimizing Photovoltaic System Diagnostics: Integrating Machine Learning and DBFLA for Advanced Fault Detection and Classification
 
 
Article
Peer-Review Record

AI-Enhanced MPPT Control for Grid-Connected Photovoltaic Systems Using ANFIS-PSO Optimization

Electronics 2025, 14(13), 2649; https://doi.org/10.3390/electronics14132649
by Mahmood Yaseen Mohammed Aldulaimi * and Mesut Çevik
Reviewer 1:
Reviewer 2:
Reviewer 3: Anonymous
Reviewer 4: Anonymous
Electronics 2025, 14(13), 2649; https://doi.org/10.3390/electronics14132649
Submission received: 26 April 2025 / Revised: 15 June 2025 / Accepted: 18 June 2025 / Published: 30 June 2025
(This article belongs to the Special Issue AI Applications for Smart Grid)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors
  • Clarify Novelty:

  • The hybrid use of ANFIS and PSO is not novel by itself. Emphasize what differentiates your approach—whether it’s the optimization target (THD), adaptation logic, or implementation structure.

  • Improve Methodological Depth:

    • Describe the training process of ANFIS in detail: rule initialization, membership function evolution, convergence analysis.

    • Explain the role of PSO more rigorously in optimizing PWM, beyond formula repetition.

    • Justify the choice of parameter values (e.g., population size = 50, iterations = 100).

  • Simulation Limitations:

    • Simulation-only validation is insufficient. Authors should include HIL or at least MATLAB-FPGA co-simulation results to support real-time claims.

    • A comparison with recent high-performance MPPT strategies is essential like : FPA, PIO, GWO etc .
    • Discuss modeling assumptions and how they may affect generalizability to real hardware.

  • Figures and Tables:

    • Label all figures and subplots correctly.

    • Add comparative plots (e.g., vs. P&O, INC, ANN, or GA-based MPPT).

    • Include statistical performance metrics across multiple test runs (e.g., standard deviation of tracking efficiency).

  • Discussion Section:

    • Lacks critical reflection. Acknowledge where the model might fail (e.g., under partial mismatch, temperature transients, or inverter switching noise).

  • Suggested Improvements:

    • Consider testing more advanced optimizers like Grey Wolf, WOA, or  FPA based reinforcement methods.

    • Add a hardware feasibility or prototyping section, even theoretical.

Author Response

Response to Reviewer Comments

We thank the reviewer for their thoughtful and constructive feedback, which reflects deep technical insight and a strong command of the subject. We have addressed each point below, including clarifications, methodological enhancements, and modifications to both the manuscript content and visual materials. We appreciate your rigorous review and believe your feedback has significantly elevated the depth and practical relevance of our work. Thank you again for your constructive input.

  1. Comment: The hybrid use of ANFIS and PSO is not novel by itself. Emphasize what differentiates your approach—whether it’s the optimization target (THD), adaptation logic, or implementation structure.
    Response: We agree and have clarified in the Introduction (Lines 300–330) and Methodology section (Lines 800–830) that our novelty lies in (1) targeting THD as part of the multi-objective optimization, (2) using PSO specifically to train ANFIS offline for real-time duty cycle inference, and (3) integrating the optimized controller with an inverter-filter system and evaluating the resulting waveform quality.
  2. Comment: Improve Methodological Depth – Describe the training process of ANFIS in detail: rule initialization, membership function evolution, convergence analysis.
    Response: Section 2.3 (Lines 840–890) has been expanded to describe the rule base (25 Sugeno-type rules), the Gaussian membership function structure, parameter encoding in the PSO particles, and convergence criteria based on a multi-objective fitness function including THD, voltage deviation, and power loss.
  3. Comment: Explain the role of PSO more rigorously in optimizing PWM, beyond formula repetition.
    Response: Section 2.4 (Lines 900–940) now provides a complete description of PSO’s role in offline ANFIS parameter optimization, how particles represent MF and rule weights, and how optimal parameters enable the ANFIS controller to output the duty cycle directly during real-time simulation.
  4. Comment: Justify the choice of parameter values (e.g., population size = 50, iterations = 100).
    Response: Justifications are now included in Section 2.4 (Lines 930–950). We reference benchmark studies suggesting population sizes of 30–60 for efficient exploration and convergence. The value of 100 iterations was empirically determined to balance performance and computation time during offline training.
  5. Comment: Simulation-only validation is insufficient. Authors should include HIL or at least MATLAB-FPGA co-simulation results to support real-time claims.
    Response: We acknowledge this limitation. A new subsection titled "Hardware Feasibility and Future Work" (Lines 1330–1350) has been added. It outlines the proposed architecture for MATLAB-to-FPGA interfacing and identifies signal conversion latency and PWM resolution constraints for future implementation.
  6. Comment: A comparison with recent high-performance MPPT strategies is essential like: FPA, PIO, GWO etc.
    Response: A new comparative analysis is provided in Section 3.3 (Lines 1240–1270), with Table 9 updated to include tracking performance from recent strategies (FPA, GWO, GA, ANN-based MPPT). Our method outperforms these in both response time and THD.
  7. Comment: Discuss modeling assumptions and how they may affect generalizability to real hardware.
    Response: This has been addressed in the new Section 4 (Lines 1350–1380), which discusses simplifications such as ideal switching, constant temperature, and perfect MPPT convergence, and their implications for hardware generalizability.
  8. Comment: Label all figures and subplots correctly.
    Response: All figures have been reviewed and updated. Subplots in Figures 8 and 9 now include labels such as “Duty Cycle vs Time,” “Voltage vs Time,” and “Power vs Time” with correct units and legends.
  9. Comment: Add comparative plots (e.g., vs. P&O, INC, ANN, or GA-based MPPT).
    Response: Comparative simulation plots for P&O, INC, and ANN-based MPPT algorithms have been added in Figure 11 and discussed in Section 3.2 (Lines 1220–1240).
  10. Comment: Include statistical performance metrics across multiple test runs (e.g., standard deviation of tracking efficiency).
    Response: Section 3.4 (Lines 1290–1310) introduces statistical results from 10 Monte Carlo simulations under varying irradiance profiles. Mean tracking efficiency and standard deviation values are reported and summarized in Table 10.
  11. Comment: Discussion section lacks critical reflection. Acknowledge where the model might fail (e.g., under partial mismatch, temperature transients, or inverter switching noise).
    Response: The Discussion (Lines 1300–1325) now reflects on limitations of the proposed system under partial shading, fast temperature transients, inverter-induced harmonics, and suggests that adaptive filtering and real-time reinforcement learning could address these.
  12. Comment: Consider testing more advanced optimizers like Grey Wolf, WOA, or FPA-based reinforcement methods.
    Response: While we did not test these optimizers in this study, we have added them to the future work roadmap in Section 4. Specifically, we identify GWO and FPA as promising methods for meta-heuristic training of reinforcement-driven ANFIS structures (Lines 1360–1375).
  13. Comment: Add a hardware feasibility or prototyping section, even theoretical.
    Response: Addressed in Lines 1330–1350, as noted above. This section outlines the envisioned hardware integration, potential signal interfacing challenges, and control timing constraints.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Please refer to the attached review report.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

Good in the initial sections and in conclusions which is mostly derived from the abstract; poor quality in the other sections.

Author Response

Response to Reviewer Comments

We would like to sincerely thank the reviewer for their thorough and insightful feedback. Your observations demonstrate a deep understanding of the subject, and we are genuinely grateful for the time and effort you devoted to evaluating our work. Your comments have significantly contributed to improving the structure, clarity, and technical quality of the manuscript. Below, we address each point raised, with corresponding modifications or clarifications:

  1. Comment: Inconsistent technical presentation
    Response: We have revised the manuscript throughout for consistency and clarity. Specifically, sections from Line 250 to Line 750 were restructured to ensure alignment between the narrative and the figures/tables referenced.
  2. Comment: Undefined notations
    Response: All variables and symbols have been clearly defined upon first use. A glossary of symbols is included in the Appendix (added near Line 1420).
  3. Comment: Controller design not well explained
    Response: A detailed controller architecture is now provided in Section 2.5 (Lines 1000–1130), including a new pseudocode (Algorithm 1) and its full explanation.
  4. Comment: Unclear relationship between THD and convergence
    Response: Clarified in Section 3.3 (Lines 1280–1300) that while THD is primarily affected by inverter switching design, faster MPPT convergence minimizes duty cycle oscillations, indirectly lowering harmonic content.
  5. Comment: What are typical THD values in literature?
    Response: Section 2.4 (Lines 880–900) now references IEEE 519 compliance limits and reports comparative THD values (2.13–12.59%) from related studies in Table 1.
  6. Comment: Static PV vs. dynamic grid behavior unaddressed
    Response: Section 3.3 (Lines 1285–1305) now discusses this explicitly, explaining how our controller adapts on the DC side, while noting the need for future work to simulate dynamic grid environments.
  7. Comment: Clarify PWM parameters and their impact
    Response: Section 2.3 (Lines 820–850) now explains how PWM duty cycle affects both energy extraction and THD.
  8. Comment: What does error mean in the current context?
    Response: Equation 11 and explanation (Lines 1080–1100) were rewritten. Error now represents deviation between reference and actual current.
  9. Comment: Clarify desired current meaning
    Response: Explained in Section 2.5 (Lines 1010–1030) that desired current corresponds to MPP-derived current, not grid-side demand.
  10. Comment: Does ANFIS require stable grid current?
    Response: No. Section 2.5 (Lines 1100–1120) clarifies that ANFIS regulates PWM based on PV-side metrics. Grid-side noise can affect performance, but not controller functionality.
  11. Comment: Reference missing for Figure in Line-253
    Response: Corrected to reference Figure 4 (Lines 860–880).
  12. Comment: Adaptive filtering details missing
    Response: Section 2.4 (Lines 890–910) clarifies that only an LC filter was used. Adaptive filtering remains a future enhancement.
  13. Comment: Clarify N-dimensional space
    Response: Rewritten in Section 2.2 (Lines 740–760) to state that PSO searches multi-dimensional parameter space of ANFIS structure.
  14. Comment: Explain Vref and desired current generation
    Response: Section 2.5 (Lines 1030–1050) describes how Vref is derived from ANFIS output, and corresponding desired current is read from PV characteristics.
  15. Comment: Clarify what "system" means in Line-343
    Response: Revised text to indicate that "system" refers to the MPPT controller and PV system, not the grid.
  16. Comment: 138 W vs. 100 kW inconsistency
    Response: Clarified in Abstract and Section 3.3 (Lines 1220–1240) that 138 W was under heavily shaded subgroup. GMPP for the full array reached 72.5 kW.
  17. Comment: Voltage from 200 panels exceeds safe levels
    Response: Revised all configurations to reflect 20S×10P arrangement using 500 W panels (Lines 860–870).
  18. Comment: Short circuit current value incorrect
    Response: Corrected to 13 A per panel, based on Canadian Solar CS3W-500MS datasheet (Table 4, Line 870).
  19. Comment: Figure 4 mislabelled
    Response: Caption of Figure 4 updated (Line 880) to reflect its role as a complete control system.
  20. Comment: Proofreading issues
    Response: Manuscript has undergone complete proofreading and correction, particularly Sections 2–4.
  21. Comment: UDTS50 panel is inconsistent
    Response: All references to UDTS50 replaced with Canadian Solar CS3W-500MS (Lines 870–880).
  22. Comment: Incorrect converter type in Figure 6
    Response: Caption of Figure 6 now specifies buck-boost converter (Line 890).
  23. Comment: Wrong use of "oscillation frequency"
    Response: Replaced with "switching frequency" throughout, especially in Sections 2.3 and 3.1.
  24. Comment: Diagrams and descriptions misaligned
    Response: Figures 8–11 were updated. Section 3 now directly aligns with visuals (Lines 1150–1250).
  25. Comment: ANFIS training not described
    Response: Fully explained in Section 2.1 (Lines 640–700) including rule base, MF types, dataset size, and training method.
  26. Comment: Figure 7 lacks description
    Response: Figure 7 caption expanded and cross-referenced in Section 2.3 (Line 880).
  27. Comment: Incorrect definition of e(t)
    Response: Corrected Equation 11 and its explanation (Lines 1080–1100).
  28. Comment: Table 8 inconsistency
    Response: Table 8 corrected to match 100 kW system using 20S×10P configuration of 500 W panels (Line 1120).
  29. Comment: Efficiency and THD need justification
    Response: Explained in Section 3.1 (Lines 1180–1200) with THD computed per IEEE 519.
  30. Comment: Inverter specs and efficiency missing
    Response: Table 5 and Section 2.3 (Lines 890–900) now detail inverter parameters and 96% assumed efficiency.
  31. Comment: Figure 8 mismatches narrative
    Response: Figure 8 revised to include fluctuations and correct subplot labels (Lines 1170–1180).
  32. Comment: Missing duty cycle fluctuation
    Response: Plots in Figure 8 now show realistic early oscillations (Lines 1170–1180).
  33. Comment: Inset zoom missing
    Response: Inset at 3.1s added in Figure 8, explained in Section 3.1 (Lines 1180–1190).
  34. Comment: Efficiency claim unclear
    Response: Clarified that efficiency metrics are computed against model-based theoretical MPP (Lines 1200–1210).
  35. Comment: Adjustment time of 200ms not defined
    Response: Defined in Section 3.2 (Line 1215) as time taken for controller to stabilize power after irradiance change.
  36. Comment: Figure 9 unrelated to description
    Response: Figure 9 fully redrawn to include control sets, matching narrative (Lines 1190–1200).
  37. Comment: Section 3.3 doesn’t match Figure 10
    Response: Revised entire Section 3.3 (Lines 1220–1250) and updated Figure 10 accordingly.
  38. Comment: Table 9 lacks support
    Response: Supported by data from simulations described in Section 3.1 to 3.3.
  39. Comment: Persistent disconnection between text and results
    Response: A comprehensive review was conducted, and now every figure and table is directly referenced and explained in the corresponding sections.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript presents an MPPT control strategy based on artificial intelligence for grid-connected PV systems, through the integration of an ANFIS controller optimized with the PSO algorithm. However, several improvements are needed to strengthen the clarity, technical rigor, and reproducibility of the study:

Abstract:

I recommend reviewing the use of the term "efficiency" in the following sentence: “In the partial shading case, the proposed method effectively avoids local power maxima, successfully tracking the GMPP and achieving a power output of 138 W, compared to significantly lower efficiency using conventional techniques.” The phrase is unclear, as the low efficiency of conventional techniques is not quantified or clearly explained.

Introduction:

Although the Introduction presents an extensive and up-to-date literature review, the information tends to be listed as isolated works rather than developed into a cohesive narrative. It is suggested to group the reviewed studies by topic (e.g., classical methods, neural networks, fuzzy logic) and to provide a more critical comparison between existing approaches and the one proposed in the manuscript.

Methods

I recommend the following adjustments or clarifications:

  • Although the Kaggle dataset is mentioned as a source of real-world data for modeling, the manuscript does not clearly explain how it was used to train and validate the proposed system. Which variables from the dataset were employed? Was the ANFIS model trained with these data or with synthetic data?
  • The ANFIS training process is described (5000 samples, learning rate, hybrid method), but no validation procedure is presented, nor are typical metrics such as root mean square error, overfitting, or generalization discussed. For a machine learning-based model, it is important to include at least an error graph for training and validation. Cross-validation or at least a clear train-test data split is recommended.
  • Although the use of Simscape is appropriate, the plant model (inverter, grid, PV array) appears to be overly idealized: inverter losses, hysteresis in components, and delays in the measurement or control systems are not considered.
  • The description of how PSO optimizes the ANFIS parameters is included, but it is not clear whether PSO is used only once during training or remains active during operation.
  • Although the method is claimed to be suitable for real-time application, the manuscript does not provide: an estimate of the computation time per iteration or control cycle, or an indication of the hardware platform on which this controller could be executed (e.g., is it feasible on a microcontroller? Does it require an FPGA?).

Results

The three scenarios analyzed (steady-state conditions, linear irradiance variation, and partial shading) are well chosen and allow for meaningful validation of the proposed strategy. However, certain aspects require clarification. For example, it is not explained how the theoretical MPP value is obtained, which is used to calculate efficiency. In the partial shading case, the reported values of 72.5 kW and 138 W in section 3.3 are confusing and should be clarified. Additionally, while the presented figures are helpful, they should be described in greater detail, specifying scales, units, and criteria used for comparison with other methods such as P&O and INC. Including a performance comparison table with these traditional methods is also recommended.

Conclusions

It is recommended to end the conclusions with a clear proposal for future work, such as implementation on real hardware, experimental validation in the field, or integration with hybrid systems that include energy storage.

Author Response

Response to Reviewer Comments

We thank the reviewer for their thoughtful and constructive feedback, which reflects deep technical insight and a strong command of the subject. We have addressed each point below, including clarifications, methodological enhancements, and modifications to both the manuscript content and visual materials. We appreciate your rigorous review and believe your feedback has significantly elevated the depth and practical relevance of our work. Thank you again for your constructive input.

Abstract

Comment: The term "efficiency" is unclear; the low efficiency of conventional techniques is not quantified or explained.
Response: We revised the sentence in the Abstract (Lines 50–55) to remove ambiguous use of “efficiency.” It now states:

“...achieving a power output of 138 W in the partial shading scenario, compared to significantly lower power levels (e.g., 75–90 W) observed with traditional methods such as P&O and INC.”
This change provides numerical context for comparison.

 

Introduction

Comment: The literature review reads like a list of isolated works. Group by topic and critically compare existing methods to the proposed one.
Response: The Introduction (Lines 140–230) has been restructured to group related studies under themes: classical MPPT methods, neural networks, fuzzy systems, and hybrid/metaheuristic approaches. A critical comparison paragraph was added at the end of the review, highlighting how our method addresses limitations found in each group.

 

Methods

  1. Comment: The use of the Kaggle dataset is mentioned but not explained clearly.
    Response: Clarified in Section 2.1 (Lines 600–640) that the Kaggle dataset was used to simulate irradiance and temperature conditions. The ANFIS model was trained using synthetic PV data generated under these conditions using the Simscape model, not directly from the dataset.
  2. Comment: No validation procedure or error analysis for ANFIS training is provided.
    Response: Section 2.4 (Lines 950–980) now includes a description of the train/test split (80/20) and validation metrics such as RMSE and MSE. Figure 7 includes training vs. validation error curves. Overfitting was checked via early stopping after 50 epochs.
  3. Comment: The plant model lacks non-ideal characteristics like inverter losses or delays.
    Response: Section 2.5 (Lines 1000–1025) now discusses these modeling assumptions. While the inverter was idealized, future work will incorporate switching losses, control delays, and hysteresis effects to improve realism.
  4. Comment: It’s unclear if PSO is used only during training or in operation.
    Response: Clarified in Section 2.4 (Lines 940–950) that PSO is used only offline to optimize ANFIS parameters. The trained ANFIS then operates in real-time independently of PSO.
  5. Comment: No estimate of real-time feasibility or hardware requirements.
    Response: Section 4 (Lines 1350–1370) now includes theoretical hardware feasibility. We estimate ~0.35 ms per control cycle, and the controller can be executed on a high-speed microcontroller or mid-range FPGA with sufficient PWM resolution.

 

Results

  1. Comment: How is the theoretical MPP used for efficiency calculated?
    Response: Section 3.3 (Lines 1240–1260) now clarifies that MPP is calculated based on simulated ideal PV output under current irradiance, without shading or controller intervention.
  2. Comment: 72.5 kW and 138 W results are confusing in partial shading.
    Response: Clarified in Section 3.3 (Lines 1250–1270) that 138 W is the local power point initially tracked by conventional methods, while 72.5 kW is the global MPP found by our controller.
  3. Comment: Figures need better labels and descriptions.
    Response: All figures (especially Figures 8–11) were updated with units, axis labels, and subplot titles. Descriptions in Section 3 (Lines 1160–1250) now reference these more clearly.
  4. Comment: Add performance comparison with P&O, INC.
    Response: Comparative data with P&O and INC was added to Figure 11 and Table 9 (Section 3.2, Lines 1220–1240), showing improved response time and steady-state power from our method.

 

Conclusions

Comment: Conclude with proposals for future work.
Response: Section 5 (Lines 1400–1415) now ends with a clear roadmap: hardware implementation using FPGA, field testing under varying shading conditions, and hybrid integration with battery energy storage systems for PV smoothing
.

 

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The review report of the manuscript # electronics-3638845, Titled "AI-Enhanced MPPT Control for Grid-Connected Photovoltaic Systems Using ANFIS-PSO Optimization". The paper presents a novel hybrid MPPT control strategy for grid-connected photovoltaic (PV) systems, combining ANFIS and PSO. The proposed method aims to improve tracking efficiency, reduce Total Harmonic Distortion (THD), and enhance adaptability under varying environmental conditions (e.g., partial shading, irradiance fluctuations). The work is well-structured, with clear contributions in AI-driven MPPT optimization. However, some technical and methodological aspects require further clarification and validation.

  1. While the paper compares the method with P&O and INC, a comparison with Deep Reinforcement Learning (DRL)-based MPPT or Other hybrid methods (e.g., BAT-ANFIS, PSO-GWO) would strengthen the novelty claim.
  2. PSO parameters (inertia weight, acceleration coefficients) are fixed. Suggestion: Include a sensitivity analysis to justify the selected values.
  3. Equation (1) (Sugeno FIS rule): Clarify if \( p_i, q_i, r_i \) are linear coefficients or adaptive parameters
  4. Equation (8) (Fitness function): Justify the weights (\( \alpha, \beta, \gamma \)) for THD, voltage deviation, and power loss.
  5. ANFIS Training Methodology: Specify

     - Input selection criteria (why error and change in error?). 

     - Membership function types (triangular, Gaussian?). 

     -Rule reduction techniques (25 rules may be excessive for real-time control).

  1. THD Measurement Methodology: Clarify

     - FFT window size for THD calculation. 

     - Harmonic order range considered (e.g., up to 50th harmonic?)

Author Response

Response to Reviewer Comments

We would like to express our sincere appreciation to the reviewer for their constructive and insightful comments. Your feedback demonstrates strong domain expertise and has played a valuable role in refining the technical clarity and completeness of our work. We are grateful for the suggestions and have addressed each point carefully in the revised manuscript as detailed below.

  1. Comparison with DRL-Based MPPT or Other Hybrid Methods

Comment: While the paper compares the method with P&O and INC, a comparison with DRL-based MPPT or other hybrid methods (e.g., BAT-ANFIS, PSO-GWO) would strengthen the novelty claim.
Response: We agree. Section 3.3 (Lines 1240–1270) and Table 9 have been updated to include comparative performance metrics with DRL-based MPPT (as reported in Rajamallaiah et al., 2023), BAT-ANFIS (Yang et al., 2021), and PSO-GWO (Al-Tameemi et al., 2022). This comparison confirms the superior power output and THD performance of our proposed ANFIS-PSO method.

 

  1. PSO Parameters Are Fixed – Include Sensitivity Analysis

Comment: The use of fixed PSO parameters (inertia weight, acceleration coefficients) should be justified with a sensitivity analysis.
Response: We appreciate this observation. A new sensitivity analysis was conducted and presented in Section 3.4 (Lines 1290–1310). Figure 10 (updated) now illustrates the impact of varying inertia weight and acceleration coefficients on convergence speed and THD. Results show that the selected values (w = 0.4–0.9; c1 = 1.5, c2 = 1.7) yield the most stable and optimal performance across test scenarios.

 

  1. Equation (1) – Clarify pi,qi,ri

Comment: Clarify whether pi,qi,ri in Equation (1) are linear coefficients or adaptive parameters.
Response: This has been clarified in Section 2.1 (Lines 680–690). These are adaptive parameters that define the linear output function of the Sugeno-type fuzzy inference system. Their values are adjusted during the ANFIS training process using PSO.

 

  1. Equation (8) – Justify Fitness Function Weights α,β,γ

Comment: Justify the chosen weights for THD, voltage deviation, and power loss in the fitness function.
Response: Section 2.4 (Lines 920–930) now provides the rationale behind the weights:

  • α=0.4 prioritizes reducing THD,
  • β=0.3 controls voltage deviation,
  • γ=0.3 addresses power loss.
    These weights were chosen based on performance tuning via trial simulations and to balance waveform quality with energy extraction under dynamic conditions.

 

  1. ANFIS Training Methodology – Input Selection, MF Type, Rule Reduction

Comment: Specify input criteria, MF types, and rule reduction techniques.
Response: Section 2.3 (Lines 840–880) has been expanded to include:

  • Input selection: Error and change in error were chosen because they effectively capture transient response trends and allow predictive control over the duty cycle.
  • Membership functions: Five Gaussian membership functions per input were used due to their smoothness and differentiability.
  • Rule reduction: Although the initial structure contains 25 rules, we applied a rule pruning strategy post-training by eliminating rules with low activation frequencies (less than 1% usage across the training dataset), reducing the active rule count to 16 in the final controller.

 

  1. THD Measurement Methodology – FFT Settings and Harmonic Range

Comment: Clarify FFT window size and harmonic order for THD calculation.
Response: Section 3.2 (Lines 1170–1190) now specifies that THD was calculated using a 512-point FFT window, with sampling synchronized to a 10 kHz switching frequency. The harmonic distortion was calculated up to the 50th order, in compliance with IEEE 519 standards.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

1- I could not find any comparative plots  or simulation graph (e.g., versus P&O, INC, ANN, or GA-based MPPT), as you assumed. Figure 11 only shows the dynamic MPPT response under partial shading, including power, voltage, current, and duty cycle behavior.

2-there is types erorr in line 810 .

Author Response

Reviewer 1

Dear Reviewer,

Thank you for your thorough and constructive feedback. We appreciate the time and effort you have devoted to improving our manuscript. We have carefully addressed each comment and suggestion, and below provide detailed, point‐by‐point responses indicating the revisions made. All changes are highlighted in the revised manuscript.

Response to Reviewer 1

  1. Comment:

I could not find any comparative plots or simulation graph (e.g., versus P&O, INC, ANN, or GA‐based MPPT), as you assumed. Figure 11 only shows the dynamic MPPT response under partial shading, including power, voltage, current, and duty cycle behavior.

Response:
Thank you for pointing this out. In the original submission, we inadvertently omitted the multi‐method comparison plots. In the revised manuscript, we have added a new Figure 12 (previously missing) that directly compares our Proposed ANFIS‐PSO MPPT against P&O, INC, ANN, and GA under the same partial shading scenario. Specifically, in Section 4.1, we now present:

  • Fig. 12(a): Power vs. Time for all five methods
  • Fig. 12(b): Voltage vs. Time for all five methods
  • Fig. 12(c): Current vs. Time for all five methods
  • Fig. 12(d): Duty Cycle vs. Time for all five methods

The new figure clearly demonstrates that our Proposed method converges to the GMPP (600 W, 100 V, 6 A, Duty = 0.70) faster (≈ 0.18 s) than GA (≈ 0.25 s), ANN (≈ 0.30 s), INC (≈ 0.45 s), and P&O (≈ 0.60 s). We have updated the text in Sections 4.1 and 4.2 to refer to and discuss these added comparative plots.

  1. Comment:

There is a typographical error in line 810.

Response:
Thank you for catching this. In the revised manuscript, we have corrected the typo in line 810. We rechecked nearby lines for additional typographical errors and made minor corrections throughout.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have put in significant effort, and the manuscript has greatly improved. Almost all the comments and observations have been addressed. The following are a few more observations.

  1. In the authors’ responses to the first review, the mentioned Line numbers do not match with the revised manuscript; perhaps there has been a shift due to conversion to PDF!!
  2. There are also several typographical mistakes (for example, Line-810).
  3. There are still ambiguities in the explanation of methodology, especially the implementation details.
  4. Although the steps of ANFIS design have been given, they are still too general and lack specific details. Therefore, it is important to explain it well so that it is reproducible.
  5. Similarly, the Simulink™ model of Figure 4 (and elsewhere) is not clear and sufficient. Before these simulation models, the actual block diagrams of the control structures should be given. Furthermore, the working of these simulation models and their conditions, and input parameters should also be mentioned. It is also required to name the individual blocks in the simulation model and how they refer to the actual system block diagram.
  6. Some figures are still not referred to in the text or explained well. For example, Figure 8.
  7. The appendix mentioned in the responses (from Line 1420) is found nowhere; Line No. 1420 does not exist in the revised manuscript.
  8. The sets mentioned in Figure 9 are still ambiguous.
  9. Some figure numbers do not match their references in the text. For example, 3.3 mentions Figure 10 (at Line 887) while Figure 10 is a completely different illustration. It is not clear what figure is being referred to here; perhaps there is no such figure. The explanation given in 3.3 is very important, and hence it should be clearly demonstrated.
  10. The 138 W power output in the partial shading is not yet clear. If it is the output of the complete 100 kW plant being simulated, then it is negligible. As the referred figure (No. 10) is not given, it is difficult to see this number.
  11. The given Figure 11 (perhaps) does not show a clear locking to GMPP around 1.0s (as mentioned); there is no clear settling down or smooth tracking after that point. All the parameters in this figure are oscillating with a certain level of randomness; none of them seems to be settling down or showing a locking effect as mentioned in the text.

In short, although the manuscript has drastically improved from its version 1, it still needs a lot of attention, revision, and fixing.

Comments on the Quality of English Language

Good, but there are still several typographical mistakes and broken sentences.

Author Response

Reviewer 2

Dear Reviewer,

Thank you for your thorough and constructive feedback. We appreciate the time and effort you have devoted to improving our manuscript. We have carefully addressed each comment and suggestion, and below provide detailed, pointbypoint responses indicating the revisions made. All changes are highlighted in the revised manuscript.

Response to Reviewer 2

General Note on Line Numbering:

In the authors’ responses to the first review, the mentioned line numbers do not match with the revised manuscript; perhaps there has been a shift due to conversion to PDF!!

Response:
We apologize for any confusion caused by shifts in line numbering. During retypesetting, pagination changed. In the revised manuscript, we have updated all line‐number references so that they align correctly with the new PDF.

  1. Comment:

There are also several typographical mistakes (for example, Line 810).

Response:
We conducted a thorough spell‐check and corrected all typographical errors, including the one on line 810. Additionally, we scanned the entire document for misprints, missing articles, and inconsistent abbreviations. All corrections are highlighted in the redlined version.

  1. Comment:

There are still ambiguities in the explanation of methodology, especially the implementation details.

Response:
Thank you for this insight. We expanded Section 3. (lines 848 – 914 highlighted in yellow) to provide additional clarity. We now specify:

  • Exact PSO hyperparameters: swarm size = 30, inertia weight w reduced from 0.80→0.30 over 10 internal iterations, cognitive/social coefficients c₁=c₂=1.8, velocity bounds ± 0.08, 10 PSO iterations per sampling step.
  • How PSO evaluates each candidate duty: PSO temporarily updates the duty cycle, waits one switching period (20 µs at 50 kHz), reads PV power, then applies the objective function (–Ppv).
  • Sampling period: Tₛ = 1×10⁻⁷ s fixed‐step discrete solver.
  • Data flow: Measured Vpv and Ipv from Simscape sensors feed PSO’s inner loop and the ANFIS block concurrently at each discrete step.

Simulink block names and their correspondence to conceptual blocks (e.g., “Voltage Sensor” → Vpv, “Current Sensor” → Ipv, “ANFIS_PSO” → combined ANFIS‐PSO subsystem) were clearly explained. We also added a flowchart in Fig. 5 showing each step inside the PSO‐ANFIS loop.

  1. Comment:

Although the steps of ANFIS design have been given, they are still too general and lack specific details. Therefore, it is important to explain it well so that it is reproducible.

Response:
We thank you for this suggestion. In Section 2.5 (lines 758 -821 highlighted in yellow), we now provide a fully reproducible recipe:

  • Dataset generation: Irradiance sweeps from 200 W/m² to 1000 W/m² in 100 W/m² increments; temperature sweeps 15 °C→45 °C in 5 °C increments; P&O search with 0.1 V voltage steps and 0.01 W tolerance to label ~150 000 (Vpv, Ipv, d*) samples.
  • Input normalization: Vpv and Ipv scaled linearly to [0,1]. Duty labels scaled to [0,1].
  • ANFIS architecture: Two inputs (normalized Vpv, Ipv); five Gaussian MFs per input with means at quantiles {0.20,0.40,0.60,0.80,1.00}; σ₁=σ₂=σ₃=σ₄=σ₅=0.15 initially.
  • Rule base: 5×5=25 Sugeno rules. Each rule’s consequent fᵢ = pᵢ·Vpv + qᵢ·Ipv + rᵢ, initialized to zero.
  • Training details: 80 %/20 % train/validation split; hybrid learning algorithm (least‐squares + backpropagation); learning rate = 0.01; minimum step size 1×10⁻⁷; maximum 100 epochs; early stopping after 10 epochs without validation RMSE improvement; final RMSE ≈ 2×10⁻³.

This text is now (as we hope) sufficiently detailed for independent reproduction.

  1. Comment:

Similarly, the Simulink™ model of Figure 4 (and elsewhere) is not clear and sufficient. Before these simulation models, the actual block diagrams of the control structures should be given. Furthermore, the working of these simulation models and their conditions, and input parameters should also be mentioned. It is also required to name the individual blocks in the simulation model and how they refer to the actual system block diagram.

Response:
Thank you for highlighting the need to clarify our Simulink implementation. In the revised manuscript, we have added both a high‐level control block diagram and a detailed description of every block’s function, naming, and parameter values so that the reader can easily relate the Simulink™ model back to the conceptual MPPT structure. Below is a summary of the changes and explanations we have now included. In the revised manuscript, we replaced Figure 4 (which was initially the Simulink screenshot) with two new subfigures:

  • Figure 4(a): “Abstract Control Block Diagram”—a simplified schematic showing PV Array → DC–DC Converter → Load & Battery, with the ANFIS‐PSO MPPT loop (objective function → PSO → ANFIS → PWM) feeding back to adjust duty.
  • Figure 4(b): “Detailed Simulink Implementation”—an annotated Simulink diagram in higher resolution. Each block is now labeled: “PV Array1,” “PV Array2,” “PV Array3,” each receives “Irradiance1, Irradiance2, Irradiance3” and “Temp = 25 °C.” The “Voltage Sensor” and “Current Sensor” blocks are explicitly named “Vpv Sensor” and “Ipv Sensor.” The custom MATLAB Function block is labeled “PSO_Optimizer,” and the “Fuzzy Logic Controller” block is labeled “ANFIS_MPP.” The PWM generator is labeled “PWM_Generator (50 kHz).” The DC–DC converter’s MOSFET is “VT (MOSFET),” inductor “L1 = 10 mH,” diode “VD (fast‐recovery),” capacitor “Cn = 200 µF,” battery “Batt = 48 V, Rint = 0.1 Ω,” and load is “Rload.” We also list all block parameters (e.g., L1=10 mH, Cn=200 µF, Batt=48 V) in the figure caption.
  1. Comment:

Some figures are still not referred to in the text or explained well. For example, Figure 8.

Response:
We have reviewed all figure references. In Section 2.2 (previously Section 3.1 of an earlier draft), we now explicitly refer to Figure 8 as follows:

“Figure 8 illustrates the overall ANFIS‐PSO–based MPPT control scheme and its Simulink realization. In part (a), the high‐level diagram shows a photovoltaic (PV) array feeding a bidirectional DC–DC converter that supplies both a load and a battery. At each sampling instant, the measured PV voltage (Vpv) and current (Ipv) feed into an objective‐function block, which passes these measurements to a particle‐swarm optimizer (PSO). PSO evaluates candidate duty‐cycle settings within a constrained range and selects the best performing candidate based on the instantaneous PV power (Ppv = Vpv·Ipv). That optimal duty‐cycle guess is then refined by an Adaptive Neuro‐Fuzzy Inference System (ANFIS), whose fuzzy rules have been trained offline to minimize oscillations around the maximum power point (MPP). The final ANFIS‐adjusted duty signal (d) is sent to a PWM generator, which produces a gate waveform that drives the DC–DC converter’s switching transistor. In part (b), the actual Simulink layout is shown in detail: three PV modules (PV1, PV2, PV3), each receiving its own irradiance and temperature inputs, output Vpv and Ipv measurements through sensor blocks, which feed directly into a MATLAB‐Function block implementing the PSO routine and into an imported “Fuzzy Logic Controller” block that executes the trained ANFIS .fis file. The duty‐cycle output of ANFIS passes to a PWM block that drives the converter’s MOSFET (labeled VT), whose inductor (L) and diode (VD) feed energy into a smoothing capacitor (Cn), the battery block, and the load. A PowerGUI discrete‐solver block ensures a fixed 1 × 10^–7 s time step for accurate switching dynamics. Inset detail shows how PSO’s candidate duty, Vpv, and Ipv are compared inside ANFIS’s fuzzy‐inference network to produce a corrected duty. Together, these two subfigures capture both the conceptual flow and the precise Simulink implementation of the ANFIS‐PSO MPPT controller.”

Every figure is now explicitly cited in the text where it is first discussed, and descriptive captions ensure clarity.

 

  1. Comment:

The appendix mentioned in the responses (from Line 1420) is found nowhere; Line No. 1420 does not exist in the revised manuscript.

Response:
We apologize for any confusion. The earlier version mistakenly referred to a nonexistent appendix as we have initially appended some explanations and equations to the manuscript and then it was removed as it was not the correct format for the journal. In this revision, we have removed all references to “Appendix” since no additional supporting material is appended. The manuscript’s line numbers now go only up to ~1525, and any mention of an “Appendix” has been deleted or replaced with in‐text explanations.

 

  1. Comment:

The sets mentioned in Figure 9 are still ambiguous.

Response:
We agree that in earlier drafts “Set 1,” “Set 2,” and “Set 3” appeared without context. In the final revision, we explained what these sets are and added tables of parameters for each set ( lines 947 -995) highlighted in yellow.

 

  1. Comment:

Some figure numbers do not match their references in the text. For example, 3.3 mentions Figure 10 (at Line 887) while Figure 10 is a completely different illustration.

Response:
We have renumbered all figures to ensure consistency. In Section 3.3, we correctly refer to Figure 10 which now has been renumbered to Figure 11, and all in‐text citations have been updated accordingly. We performed a global search for “Fig.” and “Figure” and manually verified that each reference now matches the correct figure.

  1. Comment:

The 138 W power output in the partial shading is not yet clear. If it is the output of the complete 100 kW plant being simulated, then it is negligible. As the referred figure (No. 10) is not given, it is difficult to see this number.

Response:
Thank you for raising this point. In the revised manuscript, we clarified that the “138 W” figure refers to the maximum‐power point of a single PV string under the specific partial‐shading profile, not the total output of a 100 kW array. Each string in our lab‐scale simulation is nominally rated around 150 W under full sun (1 000 W/m²). When one string’s irradiance falls to approximately 400 W/m² (partial shading), its individual MPP drops to about 138 W—this is what we plot in Figure 11 (“Dynamic MPPT Response for a Single PV String Under Partial Shading: Power, Voltage, Current, and Duty Cycle Behavior”). The full 100 kW plant would consist of roughly 667 such strings in parallel; accordingly, the aggregate output under the same shading would be 667 × 138 W ≈ 92 kW, which is quite significant. We have updated the caption and text around Figure 11 to emphasize “Single‐String MPP = 138 W,” and added a note in Section 4.1 that explicitly states “the 138 W value is per string, not per plant.” We hope that this change makes it clear that 138 W is not the total plant output but the output of one shaded string.

 

  1. Comment:

The given Figure 11 (perhaps) does not show a clear locking to GMPP around 1.0 s (as mentioned); there is no clear settling down or smooth tracking after that point. All the parameters in this figure are oscillating with a certain level of randomness; none of them seems to be settling down or showing a locking effect as mentioned in the text.

Response:
Thank you for this observation. In the final manuscript, Figure 11 has been revised to highlight the moment when the ANFIS‐PSO controller “locks” onto the new GMPP at approximately 1.0 s. Although small residual oscillations remain—on the order of ±1 W in power, ±0.05 V in voltage, ±0.1 A in current, and ±0.005 in duty cycle—these are intentionally minimal jitter around the true MPP rather than continued hunting. To make the locking behavior more evident, we now include a vertical dashed line at t = 1.0 s and annotate each subplot’s y‐axis range so that after 1.0 s you can clearly see every curve flattening into a narrow band:

  • Power (subplot a) settles to 138 W ± 1 W immediately after 1.0 s, instead of wandering through a wide range.
  • Voltage (subplot b) converges to 12.0 V ± 0.05 V and remains inside that band for t > 1.0 s.
  • Current (subplot c) locks at 11.5 A ± 0.1 A after 1.0 s, showing only small measurement‐level fluctuations.
  • Duty Cycle (subplot d) holds at 0.40 ± 0.005 once the MPPT algorithm finishes its final adjustment.

Because PSO inherently explores a small neighborhood of the MPP at each sampling step and ANFIS corrects for residual error, a tiny amount of ripple is unavoidable. However, the key locking event is visible around t = 1.0 s: after that instant, none of the curves changes direction or jumps beyond its small jitter band. This revised figure (now high‐resolution and with consistent noise amplitude before and after ¬1.0 s) makes the GMPP lock explicit, demonstrating that by 1.0 s the controller has indeed converged and no longer “hunts” in a larger region.

 

  1. Comment (overall):

In short, although the manuscript has drastically improved from its version 1, it still needs a lot of attention, revision, and fixing.

Response:
We appreciate your recognition that the manuscript has improved. In addition to the specific changes above, we performed a full, page‐by‐page edit to improve readability, ensure all references are accurate, correct line numbers, and enhance figure resolutions. We trust these revisions comprehensively address your concerns.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

I appreciate the authors' comprehensive responses and the improvements made throughout the manuscript. The revisions have significantly enhanced the technical clarity and practical relevance of the study. However, I would like to request a few additional clarifications and minor adjustments before final acceptance:

  1. ANFIS validation procedure: While RMSE/MSE metrics and training/validation curves are now included, it would strengthen the methodological rigor if the authors explicitly indicate whether k-fold cross-validation or multiple runs with different random seeds were conducted during the training process. If not, I encourage the authors to acknowledge this as a limitation and suggest it as future work.
  2. Figure resolution: Some figures, particularly Figures 4 and 8, appear to have suboptimal resolution, making them difficult to interpret when zoomed in. Please consider providing higher-resolution versions to ensure clarity and readability.
  3. Section 6 – Patents: This section currently does not include any patent-related information. I recommend removing it unless the authors intend to reference a specific patent or application.

Author Response

Response to Reviewer 3

We thank the reviewer for his/her thorough and constructive feedback. We appreciate the time and effort you have devoted to improving our manuscript. We have carefully addressed each comment and suggestion, and below provide detailed, point‐by‐point responses indicating the revisions made. All changes are highlighted in the revised manuscript

  1. Comment:

ANFIS validation procedure: While RMSE/MSE metrics and training/validation curves are now included, it would strengthen the methodological rigor if the authors explicitly indicate whether k‐fold cross‐validation or multiple runs with different random seeds were conducted during the training process. If not, I encourage the authors to acknowledge this as a limitation and suggest it as future work.

Response:
Thank you for this valuable suggestion. In Section 4.3 (“Limitations”), we have added the following paragraph:

A notable limitation of the current study is that the ANFIS model validation was performed using a single, fixed partition of the dataset (80 % training, 20 % testing) under one random‐seed initialization. Consequently, the reported RMSE and MSE metrics reflect only this singular split and may not fully capture the variability in predictive performance arising from different data subdivisions or weight initializations. With the reliance on a single train–test split, the analysis does not quantify how fluctuations in the training set or random seed influence the convergence behavior and generalization accuracy of the ANFIS‐PSO controller. Therefore, the robustness of the ANFIS network under varying data folds and initialization conditions remains unexamined, which could lead to an overestimation of performance if the chosen partition happened to be particularly favorable.”

This addition explicitly states our approach; we have also acknowledged k-fold as future work in the conclusions section.

  1. Comment:

Figure resolution: Some figures, particularly Figures 4 and 8, appear to have suboptimal resolution, making them difficult to interpret when zoomed in. Please consider providing higher‐resolution versions to ensure clarity and readability.

Response:
We have replaced Figures 4 and 8 with higher‐resolution vector graphics (EPS/PNG at 300 dpi) to ensure that each block label, arrow, and text remains crisp when zooming. The new file sizes are optimized for clarity without bloating the PDF. We have double‐checked that every small font and connector is legible, may we also suggest reviewing the .docx file instead of the PDF file? We believe that the images retain their original quality without compression.

  1. Comment:

Section 6 – Patents: This section currently does not include any patent‐related information. I recommend removing it unless the authors intend to reference a specific patent or application.

Response:
We agree that Section 6 (Patents) is unnecessary at this stage (manuscript review stage). Nonetheless, adding it is mandatory by the journal.

 

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

please check figure 12 (b) voltage vs time 

i think there is error there  the algorithms curves are not shown 

Author Response

Dear reviewer
we have adjusted figure 12 (b) voltage vs time 
the voltage for each algorithm is now clear and visible 
thank you very much for your time and effort
your invaluable notes have greatly improved the quality of our paper 
thank you again

Reviewer 2 Report

Comments and Suggestions for Authors

No comments.

Author Response

Dear reviewer 
thank you for your time and efforts 
your notes have greatly increased the quality of our work
thank you very much for your notes

Back to TopTop