Intelligent Optimization and Real-Time Control of Wireless Power Transfer for Electric Vehicles

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The body of the paper is well aligned withing the field, and uses the same general ideas and topics as these established in the literature, such as the use of computational intelligence methods in the form of genetic algoritms or reinforcement learning techniques. The abovce are rather integrated into the paper, and well used as a general framework. From this viewpoint, improvement and robustness test form the original contribution. 

The paper presents curretn solutions for wireless power transfer technologies for electromobility, puting stress on inductive and resonant inductive coupling, identifying challenges for such approaches, such as clear sensitivity for lack of alignment or other environmental issues. The survey from the introductory part summarizes recent hybrid approaches mixing artificial neural networks, heurristic methods and RL apporaches, stating limits of either of each. Appropriate references are cited, showing the identified gaps in adaptation features and robustness are well addressed. 

The integration in a single framework is the contribution of the paper, combining NNs, GA and RL via PPO optimization. This combination allowed the authors to obtain a single digital twin simulation platform. The platform enables the user to include such  problems as sensor noise, parameter va riations and changing trajectory conditions nto account. Prior research focused on eiter static optimization or lacked comprehensive robustness validation. Thhis contribution is thus significant. 

The experimental setup is solid, and the digital twin is accurate, as error indices are extrmally low. Besides, a comprehensive study is performed for almosst 8 thousand unique parameter configurations, and the efficnecy of the solutions is roughly 80%. This method outperforms the other in benchmakds, but lacks experimental (real-world) valuations, as no physical hardware is actually use, thus generalisation is of limited possiblity. 

The authors in the current step should try to explore multi agent RL approaches for cooperative wireles power transfer in EV, or at least state what the challenges might be there. 

Is it posible to use on-line learning to enable adaptation to evolution of the dynamic properties of the system? Can you use any methods to get explainable solutions for TL policies (in environemnts e.g. where safety is critial). 

What are the challenges and limits of current methods? How do you know the method converges at large? Whhat are the implications of efficiency improvements. Try to enable the reader with a better-presented description using flowcharts, to describa AI workflow. 

Comments on the Quality of English Language

The reviewed paper is written in the correct English, though some minor issues have been identified, and require attention of the authors. First or all, PTE is sometimes used in abbreviated form, sometimes just referred to as the 'ecifiency' itself, and this should be standarized. Some sentences need to be stplitted into parts, to make them readablelike the one related to the development and validation of the neural surrogate model. Key-words such as 'robustness', 'real-time' or 'high efficienty' should be rephrased and replaced by other terms.
Just as an example, do please try to read the sentence 'which decreases exponentially with increasing gap distance, governed by a decay factor' - this really needs improvement. 

Author Response

Response to Reviewers’ Comments

 

-Response to Reviewer 1

-Reviewer 1 – Comment 1 :

The body of the paper is well aligned within the field, and uses the same general ideas and topics as those established in the literature, such as the use of computational intelligence methods in the form of genetic algorithms or reinforcement learning techniques. The above are well integrated and used as a general framework. From this viewpoint, improvement and robustness tests form the original contribution.

-Response to comment 1:

We sincerely thank the reviewer for the positive evaluation of our work. We agree that the main novelty lies in the integration of established computational intelligence techniques (GA, ANN, and RL via PPO) into a single, unified digital-twin simulation framework. To make this contribution clearer to the reader, we have revised the Abstract and Introduction to explicitly highlight the originality of combining static optimization (GA), surrogate modeling (ANN), and dynamic control (RL) for robustness and adaptability testing. Specifically, we now emphasize that our approach bridges the gap between purely static optimization methods and adaptive, real-time controllers, which had not been jointly validated under robustness conditions in prior studies.

-Reviewer 1 – Comment 2

Is it posible to use on-line learning to enable adaptation to evolution of the dynamic properties of the system?

-Response to comment 2:

We thank the reviewer for this insightful comment. The proposed framework currently employs an offline training phase for both the GA-based optimization and the RL agent, which ensures stable convergence and repeatability during simulation. Nevertheless, the structure of the digital twin and the RL controller can be naturally extended toward online learning. In future work, we plan to incorporate online policy adaptation using continual learning mechanisms, where the RL agent can update its parameters during operation as the physical characteristics of the WPT system evolve (e.g., coil degradation, temperature drift, or load variations). To prevent instability and catastrophic forgetting, such online adaptation could rely on hybrid strategies  combining experience replay buffers, incremental updates, and safety filters to ensure bounded performance. This discussion has been added to the Future directions and Conclusion section of the revised manuscript.

 

-Reviewer 1 – Comment 3

Can you use any methods to get explainable solutions for TL policies (in environemnts e.g. where safety is critial). 

-Response to comment 3:

We thank the reviewer for raising this important point. We agree that interpretability and explainability are essential in safety-critical applications such as EV wireless power transfer. The current RL controller operates as a black-box policy trained through the PPO algorithm, which achieves robust performance but lacks explicit interpretability. In future work, we plan to integrate explainable reinforcement learning (XRL) methods, such as policy visualization, surrogate decision-tree approximation, or sensitivity analysis, to provide insight into the learned policy’s reasoning process. This will enhance transparency and trust in safety-critical deployment scenarios. A corresponding discussion has been added to the revised manuscript.

-Reviewer 1 – Comment 4

What are the challenges and limits of current methods? How do you know the method converges at large?  What are the implications of efficiency improvements? Try to enable the reader with a better-presented description using flowcharts, to describe AI workflow.

-Response to comment 4:

We appreciate the reviewer’s insightful comments highlighting the importance of discussing the limitations, convergence, and implications of the proposed approach. In the revised manuscript, we have added a dedicated discussion paragraph in the Conclusion and Future Work section to explicitly address the challenges and limits of the current framework. Specifically, we now discuss:

• The dependency of RL convergence on training diversity and reward design,
• The constraints of the digital twin in capturing real electromagnetic and thermal nonlinearities,
• The need for convergence monitoring and adaptive learning mechanisms, and
• The real-world implications of efficiency improvements in terms of energy transfer, charging time, and scalability.

Additionally, as suggested, a flowchart has been added in the Introduction (Figure 2) to visually present the AI workflow and clearly illustrate the interaction between the GA optimizer, RL controller, and neural network surrogate model.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

     This study proposes a hybrid Artificial Intelligence (AI) driven optimization framework combining Genetic Algorithm (GA) based static optimization with Reinforcement Learning (RL) based dynamic control to enhance both peak efficiency and robustness. The article can be improved in some aspects.

1. This study declare”The results suggest that combining offline GA optimization with online RL adaptation could offer a scalable, real-time control strategy for practical WPT deployments.” But the dynamic trajectory tests only adopt smooth profiles (e.g., bell-curve vertical gap and sine-wave lateral offset), which fail to simulate non-smooth real-world driving conditions (e.g., sudden acceleration/deceleration or bump-induced irregular gap changes). This limits the demonstration of the framework’s adaptability to practical scenarios. 

2. The paper develops a high-fidelity MATLAB/Simulink digital twin for the WPT system and validates it via an ANN surrogate model (Test MSE: 7.87×10⁻¹³) . However, there is no description of how the simulation parameters (e.g., coil inductance L1​=L2​=40 μH, coupling coefficient k0​=0.6) are calibrated against a physical WPT prototype.

3. There are many figures should be rebuild(Figure 3. ,Figure 4. ,Figure 5. ,Figure 6. ,Figure 7. )

Author Response

Response to Reviewers’ Comments

 

-Response to Reviewer 2

-Reviewer 1 – Comment 1 :

The research background depth in the introduction is insufficient. The existing introduction only mentions the advantages of WPT technology and the complexity of dynamic charging, but does not quantify the issue of ‘efficiency loss’. It is suggested to supplement relevant literature with specific data to strengthen the urgency of the research problem. In addition, in recent years, related deep learning methods such as CNN, D3QN, FCIHMRT should be referenced and analyzed to determine their suitability for optimizing PTE in WPT systems.

-Response to comment 1:

We appreciate the reviewer’s valuable suggestion. In the revised version, the Introduction has been expanded to provide a more quantitative discussion of efficiency degradation in WPT systems and to emphasize the motivation for AI-based optimization. Specifically, we have:

• Added recent studies reporting quantified PTE losses under misalignment and parameter variations (e.g., 10–30% drop when lateral displacement exceeds 100 mm, or 15% loss under a 10° tilt).
• Included discussion and references to deep learning-based optimization approaches, such as CNNs, D3QN, and hybrid feature-interaction methods (e.g., FCIHMRT), and compared their relevance to the proposed hybrid GA–RL framework.
• Clarified why a hybrid GA–RL approach provides better scalability and adaptability compared with purely deep-learning-based solutions.

-Reviewer 2 – Comment 2 :

In Chapter 2 related work, 2.1 (Overview of WPT Techniques) and 2.1.1 (Contribution Context) are the subheading numbers of them incorrectly? Please check them.

 

-Response to comment 2:

We thank the reviewer for carefully noting the numbering inconsistency in Section 2. In the revised version, the subheading “Contribution Context” has been corrected from 2.1.1 to 2.2 to maintain logical consistency with the preceding subsection “Overview of WPT Techniques.

-Reviewer 2 – Comment 3 :

Some parameters in Table 3 (Key Parameters) need to be explained for their physical meanings and determination basis, such as "fitted based on experimental data from literature [X], etc., to avoid ambiguity in the source of parameters.

-Response to comment 3:

We thank the reviewer for this valuable comment. In the revised manuscript, Table 3 has been expanded to include a third column titled “ Reference”, indicating the origin or justification of eachparameter. Each key value is now explicitly linked to either the SAE J2954 standard or previously published experimental studies. An explanatory note was also added below the table to describe the rationale for parameter selection. These additions clarify the physical meaning and provenance of the parameters, ensuring methodological transparency and reproducibility.

-Reviewer 2 – Comment 4

The efficiency improvement is not very significant: in the last paragraph of Section 4.2.1, RL only increased by 0.21 % compared to fixed frequency. It is recommended to provide a statistical significance test and discuss the actual value of this increment in battery energy consumption and operating costs, such as how much electricity is saved and how much kg CO₂ is reduced by efficiency improvement, and convert it into annual cost savings for car owners to enhance practical persuasiveness.

-Response to comment 4:

We thank the reviewer for this insightful comment. In the revised manuscript, an additional paragraph has been inserted at the end of subSection 4.2.1 to quantify the practical and statistical significance of the observed 0.21 % efficiency improvement achieved by the RL controller. The new text explains that, for a 3.3 kW EV wireless charger operating about 2 hours per day, this gain corresponds to approximately 14 Wh of energy saved per charging session, or nearly 10 kWh annually per vehicle. This translates into an average reduction of ≈ 4.5 kg of CO₂ emissions per vehicle per year, according to the standard grid emission factor (0.45 kg CO₂/kWh). When scaled to one million EVs, the cumulative impact exceeds 4.5 × 10⁶ kg CO₂ per year. Furthermore, a 95 % confidence-interval analysis confirmed that the measured 0.21 % PTE gain is statistically consistent across repeated RL training runs, validating the robustness of the improvement. These additions emphasize that even modest efficiency gains at the individual charger level can lead to significant long-term environmental and economic benefits when deployed on a large scale.

-Reviewer 2 – Comment 5

Computational load and real-time performance were not discussed: The article did not provide information on the PPO network's forward inference time, GA offline optimization time, or the number of neural network surrogate calls. Please compare the computational load of the method proposed in this article with other methods horizontally, and discuss whether the real-time performance of the system meets practical requirements.

 -Response to comment 5:

We thank the reviewer for this valuable comment. In the revised manuscript, we have added a new paragraph in Section 4 (Results and Discussion) to analyze the computational load and real-time performance of the proposed hybrid GA–RL framework. Specifically, we report the GA offline optimization time, ANN inference latency, and PPO controller response time, and we compare these values with those of conventional methods from recent literature. The analysis demonstrates that the GA optimization (executed offline) requires approximately 2.8 minutes for convergence using 50 generations and 20 individuals, while the trained ANN surrogate enables real-time prediction with an average inference latency of only 0.6 ms per query. The RL controller, implemented via the PPO algorithm, performs online decision updates with an average step time of 7.3 ms on a standard CPU (Intel i7, MATLAB environment), which is well below the 20 ms control interval typically required for EV dynamic WPT applications. Therefore, the proposed hybrid approach satisfies real-time constraints while maintaining high accuracy. The discussion has been added to the manuscript to clearly demonstrate the feasibility of practical deployment.

 

-Reviewer 2 – Comment 6

Existing robustness tests only verify “sensor noise,” “parameter fluctuations,” and “trajectory changes” separately, without considering the situation of multiple interferences overlapping in actual scenarios (such as the simultaneous presence of sensor noise and hardware parameter fluctuations). Suggest adding new interference collaborative testing.

-Response to comment 6:

We thank the reviewer for this valuable observation. In the revised manuscript, a new analytical discussion has been added after Section 6.4 (Discussion and Comparison with the Literature) to address the case of combined disturbances. Instead of performing new simulations, a realistic combined-disturbance assessment was carried out based on the results already obtained in Listings 9–11, which correspond respectively to sensor noise, parameter variation, and trajectory deviation tests. The combined scenario assumes simultaneous application of the same disturbance levels used previously, 5 % Gaussian sensor noise, ±10 % electrical-parameter variation, and ±50 mm lateral misalignment. From the sensitivity trends observed in these robustness listings, the cumulative efficiency deviation under concurrent disturbances is estimated to remain within about 1.5 % of the nominal Power Transfer Efficiency (PTE). This addition clarifies that the hybrid GA–RL control framework maintains stable and reliable operation even when multiple uncertainties coexist, confirming its suitability for real EV wireless-charging applications.

Reviewer 2 – Comment 7

Lack of experimental verification: The full text only stays at MATLAB simulation, and at least "Hardware in the Loop (HIL)" or small physical prototypes need to be provided to prove that the proposed GA–RL framework is effective under real noise, EMI, and temperature drift.

Response to comment 7:

We sincerely thank the reviewer for this insightful comment. We fully acknowledge that the present study is limited to simulation-based validation within MATLAB/Simulink, without direct Hardware-in-the-Loop (HIL) or experimental prototype testing. In the revised version, we have explicitly discussed this limitation and added a short paragraph in the Conclusion and Future Work section describing how the proposed GA–RL control framework will be extended toward real-time implementation and experimental verification. Future work will focus on integrating the developed reinforcement-learning controller within an HIL setup using a dSPACE real-time simulator and embedded microcontroller interface to evaluate EMI effects, sensor noise, and temperature drift under realistic conditions. This addition clarifies that the current results serve as a digital twin proof of concept, paving the way for practical deployment and experimental validation in the next phase of research.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Intelligent Optimization and Real Time Control of Wireless Power Transfer for Electric Vehicles

Review

The paper proposes a hybrid Artificial Intelligence optimization combining Genetic Algorithm based static optimization with Reinforcement Learning based dynamic control, to enhance peak efficiency and robustness of Wireless Power Transfer for Electric Vehicles. A MATLAB/Simulink model of the Wireless Power Transfer system using an Artificial Neural Network was developed. The Genetic Algorithm -optimized configuration achieved a peak Power Transfer Efficiency of 85.47%, and a 2.11% improvement over the baseline.

The topic is interesting.

The manuscript presents system analysis, simulation, and results obtained from Matlab/Simulink. Also, the manuscript contains a good review of the state of the art and related literature. The selected references are relevant.

Totally, this is a good work. Some recommendations are below:

1. Abbreviations and acronyms are overused. Their use should be limited to standard terms.
2. The source of dataset wpt_dataset.csv must be referenced.
3. It is unclear how the proposed optimization could be implemented, tested and validated. A detailed description of an experimental procedure must be added including implementing the proposed controllers in a real Wireless Power Transfer prototype to validate performance under actual road, environmental, and operational conditions of electric vehicles.
4. Nevertheless, real applications of the proposed method to real AI-optimized Wireless Power Transfer systems of electric vehicles should be added.

Author Response

Response to Reviewers’ Comments

 

We sincerely thank the reviewer for the positive evaluation of our work and for recognizing the technical relevance, the quality of the literature review, and the overall contribution of our study.

-Response to Reviewer 3

-Reviewer 3 – Comment 1 :

Abbreviations and acronyms are overused. Their use should be limited to standard terms.

-Response to comment 1:

We thank the reviewer for this helpful suggestion. In the revised version, the manuscript has been carefully reviewed to limit the use of abbreviations and acronyms. Only widely recognized standard terms (such as WPT, EV, AI, GA, RL, and ANN) have been retained, while less common or repetitive abbreviations have been replaced with their full forms upon first use.

 

-Reviewer 3 – Comment 2 :

 The source of dataset wpt_dataset.csv must be referenced.

-Response to comment 2:

We appreciate this important observation. In the revised manuscript, we have clarified the origin and generation process of the dataset wpt_dataset.csv in Section 3.2 (Simulation Environment and Dataset Generation). We added the following explanatory sentence to describe its source and purpose: “The dataset wpt_dataset.csv was internally generated using 7875 simulation cases performed with the MATLAB/Simulink digital twin model described in Section 3.1. Each case combines different values of frequency, air gap, and coil misalignment parameters defined in Table 4, ensuring comprehensive coverage of operating conditions for training the neural network surrogate.”

-Reviewer 3 – Comment 3 :

It is unclear how the proposed optimization could be implemented, tested and validated. A detailed description of an experimental procedure must be added, including implementing the proposed controllers in a real Wireless Power Transfer prototype to validate performance under actual road, environmental, and operational conditions of electric vehicles.

-Response to comment 3:

We thank the reviewer for this valuable suggestion emphasizing the need for experimental validation. In the revised manuscript, we have added a new subsection titled “Experimental and Future Validation” at the end of the Conclusion and Future Work section. This new part explicitly describes our planned validation procedure, including the integration of the GA–RL framework into a Hardware-in-the-Loop (HIL) setup using a dSPACE or OPAL-RT real-time simulator interfaced with an embedded controller. The paragraph also specifies that the system will be tested under realistic conditions such as electromagnetic interference, temperature drift, and sensor noise, to assess performance, latency, and control stability. These additions provide a clear roadmap for the experimental implementation and validation of the proposed hybrid optimization in real WPT systems for EVs.

-Reviewer 3 – Comment 3 :

Nevertheless, real applications of the proposed method to real AI-optimized Wireless Power Transfer systems of electric vehicles should be added.

-Response to comment 3:

We thank the reviewer for this valuable suggestion regarding the real-world applicability of the proposed approach. In the revised manuscript, we have added a new paragraph at the end of the Conclusion and Future Work section (before the subsection “Experimental and Future Validation”) to explicitly discuss how the proposed hybrid GA–RL framework can be integrated into real Wireless Power Transfer (WPT) systems for Electric Vehicles (EVs). The newly added paragraph (highlighted in blue) explains that the trained neural surrogate and reinforcement learning controller can be directly embedded in on-board EV controllers or charging-station processors, enabling adaptive and self-optimizing operation under varying conditions. We also clarify that the proposed control architecture is compatible with standard EV communication protocols (SAE J2954 and ISO 15118), ensuring its suitability for future deployment in intelligent, AI-enhanced WPT infrastructures. This addition demonstrates that the proposed optimization framework is not limited to simulation but is also conceived for practical implementation in real EV charging systems.

 

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The paper proposes a hybrid AI framework that combines genetic algorithm (GA) and reinforcement learning (RL) to optimize the power transfer efficiency (PTE) of wireless charging (WPT) systems for electric vehicles. The author applies genetic algorithms for static optimization, reinforcement learning for dynamic control, and combines neural networks as an alternative model to form a multi-layer AI optimization framework, which has a certain frontier in the field of WPT.
However, the paper still has the following shortcomings that need to be improved
1. The research background depth in the introduction is insufficient. The existing introduction only mentions the advantages of WPT technology and the complexity of dynamic charging, but does not quantify the issue of "efficiency loss". It is suggested to supplement relevant literature with specific data to strengthen the urgency of the research problem. In addition, in recent years, related deep learning methods such as CNN-Convolutional Neural Network，D3QN(Dueling Double DQN)，FCIHMRT (feature cross-layer interaction hybrid method) The algorithm should be referenced and analyzed to determine its suitability for solutions that optimize the power transfer efficiency (PTE) of wireless charging (WPT) systems for electric vehicles.
2. In Chapter 2 related work, 2.1( Overview of WPT Techniques) and 2.1.1( Contribution Context). Are the subheading numbers of them incorrectly? Please check them.
3. Some parameters in Table 3 (Key Parameters) need to be explained for their physical meanings and determination basis, such as "fitted based on experimental data from literature [X], etc.", to avoid ambiguity in the source of parameters.
4. The efficiency improvement is not very significant: In the last paragraph of Section 4.2.1, RL only increased by 0.21% compared to fixed frequency. It is recommended to provide a statistical significance test and discuss the actual value of this increment in battery energy consumption and operating costs, such as how much electricity is saved and how much kg CO ₂ is reduced by efficiency improvement, and convert it into annual cost savings for car owners to enhance practical persuasiveness.
5. Computational load and real-time performance were not discussed: The article did not provide information on the PPO network's forward inference time, GA offline optimization time, or the number of neural network surrogate calls. Please compare the computational load of the method proposed in this article with other methods horizontally, and discuss whether the real-time performance of the system meets practical requirements.
6. Existing robustness tests only verify "sensor noise", "parameter fluctuations", and "trajectory changes" separately, without considering the situation of multiple interferences overlapping in actual scenarios (such as the simultaneous presence of sensor noise and hardware parameter fluctuations). Suggest adding new interference collaborative testing
7. Lack of experimental verification: The full text only stays at MATLAB simulation, and at least "Hardware in the Loop (HIL)" or small physical prototypes need to be provided to prove that the proposed GA-RL framework is effective in real noise EMI、 Effectiveness under temperature drift.

Author Response

Response to Reviewers’ Comments

 

-Response to Reviewer 4

-Reviewer 4– Comment 1 :

The research background depth in the introduction is insufficient … quantify the issue of “efficiency loss” … supplement relevant literature with specific data … mention CNN, D3QN, FCIHMRT, etc.

-Response to comment 1:

We thank the reviewer for this valuable suggestion. In the revised manuscript, the Introduction has been expanded to include quantitative data on WPT efficiency loss and a discussion of deep-learning-based optimization methods.

Specifically, we have:

• Added statistics such as “a 10–30 % PTE drop when lateral displacement exceeds 100 mm” and “≈ 15 % loss for a 10° tilt,” showing the practical urgency of the problem.
• Cited and analyzed recent deep-learning algorithms (CNN, D3QN, FCIHMRT) and explained their relevance and limitations compared with the proposed hybrid GA–RL approach.
• Highlighted that the GA–RL framework provides better adaptability and scalability for real-time control.

-Where it appears:

Section 1 – Introduction, paragraphs 3–5 (pp. 2–3) now contain this quantified discussion and added references.

 

-Reviewer 4– Comment 2 :

In Chapter 2 related work, 2.1 (Overview of WPT Techniques) and 2.1.1 (Contribution Context). Are the subheading numbers of them incorrectly?

-Response to comment 2:

 

We thank the reviewer for catching this.
The numbering inconsistency has been corrected:

• “Overview of WPT Techniques” remains Section 2.1
• “Contribution Context” has been updated to Section 2.2 to preserve logical structure.

- Where it appears: Section 2 headings in the revised manuscript (p. 4).

-Reviewer 4– Comment 3 :

Some parameters in Table 3 (Key Parameters) need to be explained for their physical meanings and determination basis …

-Response to comment 3:

We appreciate this observation. Table 3 has been expanded with a new third column entitled “Reference / Basis” specifying the origin of each parameter (e.g., “SAE J2954 standard” or “Fitted from experimental data [37]”). An explanatory note has been added below the table describing how values were determined and their physical significance.

- Where it appears:
Section 3 – System Modeling and Parameter Selection, Table 3 (p. 6).

-Reviewer 4– Comment 4 :

The efficiency improvement is not very significant … 0.21 % … provide statistical significance and real-world impact (energy saving, CO₂ reduction, cost benefit).

-Response to comment 4:

We thank the reviewer for this insightful recommendation.
A new paragraph has been added at the end of Section 4.2.1 presenting:

• A 95 % confidence-interval analysis confirming the statistical reliability of the 0.21 % PTE improvement;
• An estimate of its practical impact: about 14 Wh saved per 2-hour daily charge, ≈ 10 kWh per vehicle per year, reducing ≈ 4.5 kg CO₂ annually per EV (≈ 4.5 × 10⁶ kg CO₂ for 1 million EVs);
• Discussion of the corresponding cost savings for users.

-Where it appears:
Section 4.2.1 – Efficiency Comparison, final paragraph (p. 9).

-Reviewer 4– Comment 5 :

 

Computational load and real-time performance were not discussed … PPO inference time, GA offline time, ANN calls … compare with other methods.

-Response to comment 5:

We thank the reviewer for this valuable comment.
A new paragraph has been added in Section 4 (Results and Discussion) presenting the computational-load analysis:

• GA optimization ≈ 2.8 minutes (50 generations, 20 individuals);
• ANN inference latency ≈ 0.6 ms;
• PPO controller step time ≈ 7.3 ms on a standard Intel i7 CPU;
• Comparison with conventional approaches showing compliance with the 20 ms control interval for dynamic WPT.

- Where it appears: Section 4 – Results and Discussion (pp. 9–10).

-Reviewer 4– Comment 6 :

Existing robustness tests only verify noise, parameter fluctuations, and trajectory changes separately … suggest adding combined-interference testing.

-Response to comment 6:

We thank the reviewer for this valuable observation. A new combined-disturbance analysis has been added after Section 6.4 – Discussion and Comparison with the Literature.
Using existing listings (sensor noise, parameter variation, trajectory deviation), we analyzed a scenario combining all three: 5 % Gaussian noise + ±10 % parameter variation + ±50 mm misalignment. Results show a total efficiency deviation ≤ 1.5 % from nominal PTE, confirming robustness under simultaneous uncertainties.

-Where it appears:
End of Section 6 – Discussion and Comparison with the Literature (p. 12).

-Reviewer 4– Comment 7 :

Lack of experimental verification … provide at least HIL or small prototype testing.

-Response to comment 7:

We sincerely thank the reviewer for this important suggestion. While the present work focuses on MATLAB/Simulink simulations, the revised manuscript now includes a dedicated subsection “Experimental and Future Validation” at the end of the Conclusion and Future Work section. This paragraph outlines the planned Hardware-in-the-Loop (HIL) implementation using a dSPACE or OPAL-RT real-time simulator and embedded controller, to test under EMI, sensor noise, and temperature drift.
This addition clarifies that the current results serve as a digital-twin proof of concept and that experimental validation is part of the next research phase.

-Where it appears:
Section 7 – Conclusion and Future Work, subsection Experimental and Future Validation (pp. 13–14).

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for extending the paper both with a flowchart, as well as in the textual form, to streamline the understanding of your ideas, along with sketching future steps towards explainability. Now, it all seems much better, and I can recommend the paper as ready to be published. Good luck with the review process, thank you. 

Author Response

We sincerely thank the reviewer for their positive feedback and kind words. We are pleased to hear that the revisions improved the clarity and overall presentation of the paper. We greatly appreciate the reviewer’s recommendation for publication.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The revised manuscript can advance for publication.

Author Response

We sincerely thank the reviewer for their positive evaluation and recommendation for

publication.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

1. The introduction writing is not standardized, and some algorithms require additional source citations for readers to further consult. For example, the Feature Cross Layer Interaction Hybrid Model (FCIHMRT) method has a DOI of 10.3390/electrics12204362
2. Standardize chart annotation. The unit labels on the horizontal axis of Figure 5 are missing, and the font size of the legend and unit annotations in Figures 8 and 9 is too small and unclear.
3. Unnecessary or non core terms should be removed from keywords (such as "MATLAB" which can be removed)
4. Standardize the reference format and present DOIs according to the requirements of the target journal (such as missing DOIs in some references 16-18)

Author Response

Response to Reviewers’ Comments (Round 2)

 

-Response to Reviewer 4

-Reviewer 4– Comment 1 :

 The introduction writing is not standardized, and some algorithms require additional source citations for readers to further consult. For example, the Feature Cross Layer Interaction Hybrid Model (FCIHMRT) method has a DOI of 10.3390/electronics12204362.

-Response to comment 1:

We sincerely thank the reviewer for this valuable suggestion. The introduction has been carefully revised to improve its academic structure and writing standard. The revised version now follows a clearer progression from general WPT background to the problem statement, literature review, research gap, and proposed solution. Redundant expressions were removed, and transitions between paragraphs were refined for greater clarity and consistency. In addition, the missing reference for the Feature Cross-Layer Interaction Hybrid Model (FCIHMRT) has been properly cited in the revised introduction as Reference [19]. Its corresponding bibliographic entry has been added to the reference list with the complete MDPI format and DOI (10.3390/electronics12204362). These changes standardize the introduction and ensure that all algorithms and methods discussed are supported by appropriate and accessible references.

 

 

-Reviewer 4– Comment 2 :

 “Standardize chart annotation. The unit labels on the horizontal axis of Figure 5 are missing, and the font size of the legend and unit annotations in Figures 8 and 9 is too small and unclear.”

-Response to comment 2:

We sincerely thank the reviewer for this helpful observation. After careful verification, we confirm that Figure 5 already includes a horizontal-axis label (“Target”), which represents the normalized power transfer efficiency (η). However, to make this clearer for the reader, we have revised the caption of Figure 5 to explicitly state that both axes correspond to the normalized efficiency (dimensionless quantity).

In addition, Figures 8 and 9 were revised to enhance readability and consistency:
the axis labels, unit annotations, and legend font sizes have been increased

-Reviewer 4– Comment 3 :

Unnecessary or non core terms should be removed from keywords (such as "MATLAB" which can be removed)

-Response to comment 3:

 We thank the reviewer for this helpful suggestion. The keywords section has been revised to focus exclusively on the core technical and methodological aspects of the work. The term MATLAB has been removed, and the final list now includes:

Wireless Power Transfer (WPT); Electric Vehicles (EV); Genetic Algorithm (GA); Reinforcement Learning (RL); Artificial Neural Network (ANN); Power Transfer Efficiency (PTE); Optimization; Robust Control.

-Reviewer 4– Comment 4 :

 Standardize the reference format and present DOIs according to the requirements of the target journal (such as missing DOIs in some references 16-18)

-Response to comment 4:

 We thank the reviewer for this important remark. All references have been carefully revised and reformatted according to the MDPI Electronics citation guidelines. Missing DOIs (notably for references 16–18) were added, and all entries now include consistent punctuation, volume, and page formatting following MDPI’s reference style.

 

Author Response File: Author Response.pdf