Forward and Backpropagation-Based Artificial Neural Network Modeling Method for Power Conversion System
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe manuscript proposes an ANN-based modeling scheme for PCS, but the following issues must be addressed:
1.Specify the minimum input set required at inference, and conduct ablation studies to produce quantitative curves of accuracy–sensor count–cost (bill of materials/PCB area/power). This should demonstrate a verifiable cost–performance trade-off.
2.The current evaluation uses only a series–parallel resistor network, which does not represent the time-varying and nonlinear behavior of switching converters. Please add at least one real power converter (e.g., a buck converter or single-phase inverter), covering dynamic operating conditions (input/load steps, switching-frequency changes), temperature/part-tolerance sweeps, and noise/EMI robustness, and report worst-case and percentile errors.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe presented work proposes a forward/backpropagation ANN that learns the nonlinear input–output behavior of a power conversion system, enabling prediction of key variables without explicit sensors. The authors argue the PCS can be simplified and costs lowered, and they support the claim with simulation and laboratory results. The paper provides some useful information, which is supported by experimental work. However, extensive modifications are needed to improve the work. Here are some comments:
- The abstract should begin by addressing the problem statement. Then, delving into the details of the paper. The first sentence is repeated. Thus, the abstract needs to be rewritten for better clarity.
- To strengthen the introduction, the authors should expand the literature review with more detail on comparable approaches and include comparisons to the most recent publications in the field.
- Section 2.1 is not written properly. The section should first verify the validity of using the circuit in Figure 1. Then, a discussion of the variations of the resistance values follows. Please check Equation (3).
- All equations derived from external sources must be supported by appropriate citations to the original references.
- More details about implementing the presented ANN into PSIM should be given.
- Some figures need to be redrawn for better clarity. Please redraw Figures 6 and 7 with larger fonts. Add an inset zoom of the first period (e.g., 0–5 ms or 0–10 ms).
- For both the simulations and the experiments, ANN prediction errors must be calculated and addressed.
- The training process is not adequately represented. More details are required. Training performance should be given. Also, MSE should be calculated.
- Inputs/targets are chosen to predict only V4, I4 of one branch; broader PCS variables or multi-node predictions are not demonstrated.
- No explicit tests for sensor noise or temperature changes are given.
- The proposed model should be evaluated on a more practical PCS model.
- Benchmark the proposed approach against published baselines to clarify its advantages and limitations.
- The conclusion needs to be rewritten. Please concentrate on the key findings, quantify the gains, and state clear future work.
The text should be proofread to minimize typographical and grammatical errors.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for Authors- Please explain "the proposed ANN modeling method can accurately predict system variables without sensors or mathematical modeling". When there are no sensors or mathematical modeling, how to control the power conversion system with closed loops.
- The circuit in Fig. 1 is too simple, that can't present a typical power conversion system. Please use the conventional power converters, such as buck converter and inverters, to verify your proposed method.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe manuscript presents a technique using a forward and backpropagation-based artificial neural network (ANN) modeling method for power conversion systems (PCS) . The methodology of the paper is described well, but the simulation and experimental results are not sufficient to prove the accuracy and reliability of the proposed technique . The following comments should be considered during the revision process.
1. You should be specific about the work contributions in the abstract after "by eliminating the need for sensors, the system structure can be simplified, and the overall cost significantly reduced."
2. It is necessary to offer a literature review in the introduction section about previous work regarding the design and control of the power conversion system , discussing the different techniques, advantages, and limitations of each, and then introduce your novelty by building upon the previous studies.
3. To increase the quality of the citations , please check and update the references used in the manuscript ; more references are outdated, such as Refs. 6, 8, 17, and 21.
4. Are the equations from (4) to (32) newly obtained equations or sourced from references ?
5. The resolution of the simulation results figures is very low ; the figure resolution and text fonts should be increased for clear observation.
6. Details about the modeling or the simulation circuit constructed in the PSIM simulation should be included in the simulation section.
7. How did you select the parameters in Table 1 for the number of neurons in different layers and learning rate (i, j, k, l, η) ?
8. The simulation results are not enough to prove that you used and applied the proposed algorithm to obtain such results (the authors only provide the voltage and current waveforms ) . The characteristics and the values of the weights and the different parameters optimized in the algorithm should be presented in the simulation results for a clear understanding of the modifications and the algorithm's operation .
9. In Section 5, based on experimental results, it should include a table describing the main components used in the developed prototype , along with the description of the control system and board used in this setup .
10. You should extend the explanation and results from the experimental platform to prove the reliability of the proposed technique; only one operating condition is not enough to clearly describe the accuracy of the proposed technique.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsThis article addresses the well-known topic of applying neural networks to modeling power electronics systems. The novelty lies primarily in combining classical forward and backward propagation algorithms with a simple reference system as proof of concept. However, no significant theoretical innovations or new ANN architectures are presented. The originality can be described as moderate – more in the application than in the methodology.
Suggestions for improvement for the article
1. Expand the discussion of existing methods for modeling PCS using ANNs, especially newer approaches (physics-informed neural networks, hybrid AI–analytical models).
2. More clearly define the research gap the article aims to fill—the current justification for this new feature is too general.
3. Add a comparison with related publications, highlighting differences in architecture or scope of application.
4. Provide additional information on:
the number of training, validation, and test samples,
the software and hardware platform used,
the training termination criteria, and the model validation method.
5. Consider presenting a flowchart of the entire training process (including input data, normalization, and validation steps).
6. Include a comparison with another method (e.g., a classical mathematical model or another neural network) to demonstrate the advantages of the proposed approach.
7. Add a quantitative assessment of accuracy (MSE, RMSE, R², percentage relative error).
8. Supplement figure descriptions with units, scales, and legends.
9. Present the results in tabular form next to the graphs to facilitate comparative analysis.
10. Indicate limitations of the method, e.g., the impact of hyperparameter selection or generalization limitations.
10. Expand the interpretation of the results beyond the statement "high accuracy."
11. Discuss potential applications in real-world power electronics systems (inverters, DC-DC converters).
12. Add suggestions for directions for further research, e.g., applications in systems with switching nonlinearities.
13. Shorten excessively wordy sentences and standardize technical terminology ("forward/backward propagation" instead of alternating forms).
14. Entrust proofreading to a native technical speaker.
15. Simplify the syntax in mathematical sections while maintaining the clarity of the equations.
16. Improve the aesthetics of the graphs (clear axes, units, reference lines).
17. Add information about the sampling rate, sensor accuracy, and environmental conditions to the experiment parameters table.
Comments on the Quality of English LanguageThere are numerous repetitions ("the proposed ANN modeling method...") and long, complex sentences that complicate the flow of the reading. The syntax is imprecise in places, and the punctuation is irregular. The editorial style needs to be improved, sentence structure simplified, and terminology consistent (e.g., between "forward and backpropagation-based" and "forward and backward propagation-based").
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe manuscript proposes a sensorless modeling method using a forward and backpropagation-based ANN. However, the paper suffers from significant limitations regarding novelty and practical applicability. First, the "proposed" method is essentially a standard Multilayer Perceptron (MLP) using standard backpropagation; the extensive mathematical derivation provided is fundamental textbook knowledge rather than a novel research contribution. Second, although the title claims application to a "Power Conversion System" (PCS), the validation is performed solely on a linear series-parallel resistor circuit. This setup does not represent the non-linear switching dynamics, harmonics, or control challenges of actual PCS hardware (e.g., converters or inverters). Validating an ANN on a static linear circuit is insufficient to prove its effectiveness for dynamic power electronics applications. Therefore, the manuscript lacks the sufficient technical depth and novelty required for publication.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have addressed most of the issues raised in the previous review, and the manuscript has been substantially revised and improved. Only a few points remain to be clarified before the paper can be considered for possible acceptance:
In Figure 7, the focus should be from 0 to 5 ms (not from 5 to 10 ms)
Comments on the Quality of English Language
The text should be proofread to minimize typographical and grammatical errors.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsI would like to thank the authors for offering the revised version of the manuscript . However, some of the previously mentioned comments are not properly answered.
- You didn 't answer comment#2, about adding a literature review ; you just added three references and a small paragraph . You should make a literature review table regarding the design and control of the power conversion system, discussing the different techniques, advantages, and limitations of each, and then introduce your novelty by building upon the previous studies.
- Also, you didn't properly answer comment #4 . I didn't ask you about the purpose of the equations; I asked you about the source of these equations . Are they novel equations, or if they are sourced from references, you have to properly cite those references?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsThe article in its current form meets the publication criteria.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 1 Report
Comments and Suggestions for Authorsargee
