Next Article in Journal
Wind Integration and Power System Planning: Challenges, Policies, and Governance in Italy
Next Article in Special Issue
The Impact of Fixed-Tilt PV Arrays on Vegetation Growth Through Ground Sunlight Distribution at a Solar Farm in Aotearoa New Zealand
Previous Article in Journal
Building Energy Performance Modelling and Simulation
Previous Article in Special Issue
Toward Sustainable Urban Transport: Integrating Solar Energy into an Andean Tram Route
 
 
Article
Peer-Review Record

Methodology for Assessing the Technical Potential of Solar Energy Based on Artificial Intelligence Technologies and Simulation-Modeling Tools

Energies 2025, 18(19), 5296; https://doi.org/10.3390/en18195296
by Pavel Buchatskiy 1, Stefan Onishchenko 1, Sergei Petrenko 1,2 and Semen Teploukhov 1,*
Reviewer 1:
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Energies 2025, 18(19), 5296; https://doi.org/10.3390/en18195296
Submission received: 25 August 2025 / Revised: 1 October 2025 / Accepted: 3 October 2025 / Published: 7 October 2025
(This article belongs to the Special Issue Solar Energy, Governance and CO2 Emissions)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents a hybrid methodology that leverages artificial intelligence (primarily LSTM neural networks) and simulation modeling tools to assess and forecast the technical potential of solar energy in distributed energy systems. However, the lack of direct benchmarking with alternative AI models, incremental methodological novelty, limited experimental depth and scale, incomplete discussion of related cutting-edge work, and insufficient reproducibility/ablation results collectively undermine its potential for acceptance. 

Concerns:

  1. A lot of closely related recent papers employ deep learning for solar (and hybrid) resource assessment, advanced simulation environments, or AI for power plant forecasting, sometimes with larger-scale benchmarks or architectural/algorithmic improvements beyond LSTM. The presented work would benefit from a more explicit comparison against these, particularly those targeting direct integration of AI with simulation workflows.
  2. The AI model is only compared against itself (actual vs. forecasted), with no head-to-head benchmarks against alternative time series models (ARIMA, GBM, DLinear, hybrid CNN approaches, etc.) or ablations showing the incremental value of the LSTM approach. This omission is problematic, as competitive benchmarking is standard for AI-based forecasting in this domain. For example, the paper cites the use of multiple models (Page 14), but no direct experimental evidence is provided for why LSTM outperforms others for these tasks.
  3. Aside from using real datasets (NSRDB) for training, the simulation remains largely virtual and focused on a hypothetical single PV module/consumer setting. A true end-to-end validation (e.g., live deployment or real-world operational scenario) is missing, which limits the demonstration of practical impact and adoption.
  4. While high-level methodology and some implementation details are given, there is limited information on hyperparameters, training/validation splits, code availability, or full dataset disclosure. For the SimInTech simulation elements, schematic block diagrams (Figures 11, 13, 14) give insight, but parameters or initialization details are sparse. This impedes full reproducibility or third-party validation.

 

Author Response

Point 1: A lot of closely related recent papers employ deep learning for solar (and hybrid) resource assessment, advanced simulation environments, or AI for power plant forecasting, sometimes with larger-scale benchmarks or architectural/algorithmic improvements beyond LSTM. The presented work would benefit from a more explicit comparison against these, particularly those targeting direct integration of AI with simulation workflows..

Point 2: The AI model is only compared against itself (actual vs. forecasted), with no head-to-head benchmarks against alternative time series models (ARIMA, GBM, DLinear, hybrid CNN approaches, etc.) or ablations showing the incremental value of the LSTM approach. This omission is problematic, as competitive benchmarking is standard for AI-based forecasting in this domain. For example, the paper cites the use of multiple models (Page 14), but no direct experimental evidence is provided for why LSTM outperforms others for these tasks.

 

Response 1, 2: The paper provides a review of sources with different models used for forecasting. It also cites a publication (86. Rajagukguk, R.A.; Ramadhan, R.A.; Lee, H.-J. A review on deep learning models for forecasting time series data of solar irradiance and photovoltaic power. Energies 2020, 13, 6623) detailing the rationale for choosing an LSTM architecture for solar radiation forecasting.

This paper did not aim to provide a comprehensive review of neural network models and architectures. The main premise is that the methodology can be modified as needed. Therefore, a modified LSTM model for working with the specified spatial data (93. Jeong, S.-H., Lee, W. K., Kil, H., Jang, S., Kim, J.-H., & Kwak, Y.-S. (2024). Deep learning-based regional ionospheric total electron content prediction—Long short-term memory (LSTM) and convolutional LSTM approach. Space Weather, 22, e2023SW003763.) was chosen as an acceptable tool that demonstrates good accuracy and training time.

 

Point 3: Aside from using real datasets (NSRDB) for training, the simulation remains largely virtual and focused on a hypothetical single PV module/consumer setting. A true end-to-end validation (e.g., live deployment or real-world operational scenario) is missing, which limits the demonstration of practical impact and adoption.

 

Response 3: The current work is primarily focused on simulation and demonstrates the methodology using a hypothetical single PV module/consumer scenario. This choice was intentional, as the main goal of the study was to develop and validate a methodological framework for assessing the technical potential of solar energy by integrating intelligent forecasting with simulation modeling. At this stage, the use of real datasets (NSRDB) ensures that the inputs to the model reflect realistic solar resource conditions, which allows us to validate the adequacy of the proposed hybrid approach at the theoretical and technical potential assessment levels.

We acknowledge that a true end-to-end validation in a real-world deployment would significantly strengthen the practical impact of the work. Such validation is part of our future research agenda. Specifically, we are planning to expand the model towards integration with experimental testbeds and pilot distributed energy systems, where both actual PV installations and consumer load data will be utilized. This will allow us to demonstrate the adaptability of the methodology under operational conditions and confirm its applicability for demand-response oriented distributed systems.

 

Point 4: While high-level methodology and some implementation details are given, there is limited information on hyperparameters, training/validation splits, code availability, or full dataset disclosure. For the SimInTech simulation elements, schematic block diagrams (Figures 11, 13, 14) give insight, but parameters or initialization details are sparse. This impedes full reproducibility or third-party validation.

Response 4: In accordance with the comment, more detailed descriptions of intellectual and simulation models were added, as reflected in the work:

 

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

My main concern regarding the proposed method is that—if I understand correctly — the training of the LSTM model has to be performed for each new location; and because the LSTM training is a delicate part of the setup process, I wonder how practical such training would be in real-world situations, particularly in the hands of general practitioners. I think this is an aspect that should be addressed in the Discussion.
Sections 1 and 3 are thorough and clear, and are a pleasure to read, but I have the impression that, up to Figure 6, the pace  would be more suited to a book. For the purpose of this article, those parts could be leaner and tighter, so as to give more space to the sections that follow.
Also, the research that follows is specifically on solar potential, and this also makes me feelthat the treatment of renewable energy sources integration that comes in sections 1 and 2 is disproportionately thorough.
The major issue in the development of the study is that the discussion (within section 4) is very short and dry and does not allow for elaboration. I feel that the article’s impact could benefit from giving the discussion arguments more depth and breadth. In addition, a section on the limitations of the study could be integrated into the discussion.

More specific comments on specific parts follow below:

Line 25. Why "As a result"? Should it rather be "To this purpose"?

Lines 188–189. When the text says "For the convenience of designing such energy systems, a group of authors in their study...", it could instead read: "Analysing the most convenient ways to design such energy systems, a study..."

Line 211. "In the following we..." could be "In the following study we..."

Line 226. "It was possible to estimate the possible..." could be replaced with "produced an estimation of the possible..."

Line 231. "Allows." Should this be "Allowed"?

The contents of Figure 10 should be presented and commented on in the text of section 3, at least in their general aspects, as they are of fundamental importance for the study. Also, the text in the figure could be made larger and more easily readable.

Line 297. "Is." Should this be "Was"?

Line 368. "However, ..." could be changed to "Within this set of approaches, ...".

In Figure 20, the predicted values and the monitored ones could be plotted in the same chart, for better comparability, as has been done in the other figures.

Author Response

Response to Reviewer 2 Comments

 My main concern regarding the proposed method is that—if I understand correctly — the training of the LSTM model has to be performed for each new location; and because the LSTM training is a delicate part of the setup process, I wonder how practical such training would be in real-world situations, particularly in the hands of general practitioners. I think this is an aspect that should be addressed in the Discussion.

 

 Changes have been made to the maintenance of the article

 

 Sections 1 and 3 are thorough and clear, and are a pleasure to read, but I have the impression that, up to Figure 6, the pace  would be more suited to a book. For the purpose of this article, those parts could be leaner and tighter, so as to give more space to the sections that follow. Also, the research that follows is specifically on solar potential, and this also makes me feelthat the treatment of renewable energy sources integration that comes in sections 1 and 2 is disproportionately thorough.

 

We agree that, as currently presented, the introductory discussion on renewable energy integration in Sections 1 and 2 provides a broad background that may appear more comprehensive than necessary in the context of a research article focused specifically on solar technical potential.

Our rationale for devoting significant attention to the general integration of renewable energy sources was to establish a strong contextual foundation for the methodology. Since the proposed approach is designed to be scalable and potentially applicable to distributed energy systems with multiple RES types, we considered it important to emphasize the broader framework before narrowing the focus to solar energy. This structure was chosen to underline that solar energy assessment, although central to this study, fits into the wider perspective of distributed generation and demand-response–oriented energy management.

 The major issue in the development of the study is that the discussion (within section 4) is very short and dry and does not allow for elaboration. I feel that the article’s impact could benefit from giving the discussion arguments more depth and breadth. In addition, a section on the limitations of the study could be integrated into the discussion.
After the last revision, Section 4 was expanded to include a more detailed presentation of the material. Model parameters and characteristics of the SimInTech environment are now included. The final limitations of the methodology are presented before the conclusion, so we decided not to include a new "Discussion" section.

 

Line 25. Why "As a result"? Should it rather be "To this purpose"?

Lines 188–189. When the text says "For the convenience of designing such energy systems, a group of authors in their study...", it could instead read: "Analysing the most convenient ways to design such energy systems, a study..."

Line 211. "In the following we..." could be "In the following study we..."

Line 226. "It was possible to estimate the possible..." could be replaced with "produced an estimation of the possible..."

Line 231. "Allows." Should this be "Allowed"?

Line 297. "Is." Should this be "Was"?

Line 368. "However, ..." could be changed to "Within this set of approaches, ...".

All these comments are corrected in the article.

 The contents of Figure 10 should be presented and commented on in the text of section 3, at least in their general aspects, as they are of fundamental importance for the study. Also, the text in the figure could be made larger and more easily readable.
Figure 10 essentially depicts the main processes of the proposed approach, which were previously described in general terms (including the approaches used to forecast solar radiation). The section on presenting the simulation model is described in detail in the section dedicated to implementing this model's structure in the SimInTech environment. It should also be noted that the figure is high-resolution, allowing the reader to examine it in detail.

In Figure 20, the predicted values and the monitored ones could be plotted in the same chart, for better comparability, as has been done in the other figures.

Figure 23 (Figure 20 before adjustments) is high resolution, and the graphs are not displayed in a single window. This is because the presented solar insolation values are input data for the simulation model, and the quality of the model's forecast was discussed previously using well-known quality assessment metrics. Figure 24, however, presents the final forecast of generation volumes for a specific power system configuration.

 

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has merit in its practical approach but requires substantial enhancements: expand validation, add comparisons, detail implementations, and include economic analysis. Resubmit after addressing these for potential acceptance in a suitable journal.

-Only one dataset/location and PV type; no comparison with other forecasting models (e.g., CNN-LSTM hybrids) or simulation tools. The forecasting horizon is short-term but not specified (e.g., hours/days?). Economic/market benefits are claimed but not analyzed. Scalability is asserted without multi-system tests.

-Implementation details for LSTM (e.g., code framework, training duration, handling of missing data beyond smoothing). Why NSRDB data from Washington, USA, for a Russian-affiliated study? How does the model handle uncertainties like weather variability? The transition from forecasting to simulation lacks details on data interfacing (e.g., file formats, automation).

-What are the specific LSTM hyperparameters (e.g., number of layers, learning rate, batch size) and why were they chosen?

-Why was SimInTech selected over MATLAB/Simulink, and how does it improve accuracy or efficiency?

-How was the forecasting horizon determined, and what is the model's performance on longer horizons (e.g., weekly/monthly)?

-Can the authors provide a sensitivity analysis for key parameters like panel efficiency or inverter losses?

-Why focus on a single US location—does this generalize to other climates, such as those in Russia?

Author Response

Point 1: Only one dataset/location and PV type; no comparison with other forecasting models (e.g., CNN-LSTM hybrids) or simulation tools.

 

Response 1: We acknowledge that the current study employs data from a single dataset (NSRDB) and focuses on a representative PV configuration for an individual consumer. This design choice was intentional, as the objective of the present work was to demonstrate the feasibility of the proposed methodology within a controlled test case before scaling up to more diverse configurations. We agree that testing across multiple datasets, geographical locations, and PV technologies would provide stronger evidence of generalizability. We plan to expand this work in future studies by conducting comparative analyses across varied climates and PV designs to validate the robustness of the proposed approach.

 

Point 2: Economic/market benefits are claimed but not analyzed / Scalability is asserted without multi-system tests.

 

Response 2: We appreciate the reviewer’s remark on the economic and market aspects. Indeed, economic feasibility and market integration are crucial factors for the large-scale adoption of renewable energy solutions. However, in the scope of the present study, the primary objective was to demonstrate a methodological framework for assessing the technical potential of solar energy by combining forecasting techniques with simulation modeling tools.

Given that the case presented in this article is illustrative in nature, a detailed economic assessment was considered beyond the intended scope. Performing such an analysis would require case-specific data, including system size, local tariff structures, policy incentives, and investment conditions, all of which vary significantly depending on location and system configuration.

That said, we agree with the reviewer that the methodology we propose can and should be extended with techno-economic assessments in future work. The flexibility of the framework allows for integration of economic modules at later stages, enabling evaluation of payback periods, levelized cost of electricity (LCOE), and market competitiveness for specific system designs. We will revise the manuscript to clarify this point and to highlight the importance of economic and market considerations in future applications of the methodology.

In the manuscript, scalability is discussed as one of the strengths of the proposed approach, but we acknowledge that no explicit multi-system validation has been conducted within the present work. This is primarily because the article was intended to demonstrate the feasibility of the methodology through a single illustrative case study rather than provide an exhaustive validation across different system scales.

It is important to emphasize, however, that both components of the proposed framework — the forecasting module and the dynamic simulation environment — are inherently adaptable. The forecasting model can be retrained with new datasets corresponding to different locations, system sizes, or PV technologies, while the simulation blocks can be reconfigured to represent more complex or large-scale energy system architectures. Therefore, the scalability assertion is based on the methodological flexibility rather than on explicit multi-system test cases.

 

Point 3: Implementation details for LSTM (e.g., code framework, training duration, handling of missing data beyond smoothing).

Point 6: What are the specific LSTM hyperparameters (e.g., number of layers, learning rate, batch size) and why were they chosen?

 

Response 3: Changes have been made to the maintenance of the article

 

Point 4: How does the model handle uncertainties like weather variability.

 

Response 4: The uncertainty of daily / seasonal cycles and short-term fluctuations is reduced by using exponential smoothing and a spatially adapted LSTM model.

 

Point 5: The transition from forecasting to simulation lacks details on data interfacing (e.g., file formats, automation).

Response 5: Changes have been made to the maintenance of the article

 

Point 7: Why was SimInTech selected over MATLAB/Simulink, and how does it improve accuracy or efficiency?

Response 7: Changes have been made to the maintenance of the article

 

Point 8: How was the forecasting horizon determined, and what is the model's performance on longer horizons (e.g., weekly/monthly)?

 

Response 8: Changes have been made to the maintenance of the article

 

Point 9: Can the authors provide a sensitivity analysis for key parameters like panel efficiency or inverter losses?

 

Response 9: In the current version of the paper, sensitivity analysis for parameters such as PV panel efficiency and inverter losses has not been explicitly included, as the primary objective was to demonstrate the methodological link between intelligent forecasting and simulation-based estimation of technical potential.

We fully agree, however, that such sensitivity analysis is an important step in validating the robustness of the proposed approach. Since the simulation environment (SimInTech) allows flexible adjustment of equipment parameters, conducting a parametric study for different efficiency levels of PV modules and inverter losses is technically straightforward. For example, small variations in panel efficiency (±2–3%) or inverter losses (±3–5%) can be expected to directly scale the resulting technical potential estimates.

 

Point 10: Why NSRDB data from Washington, USA, for a Russian-affiliated study? Why focus on a single US location—does this generalize to other climates, such as those in Russia?

 Response 10: Changes have been made to the maintenance of the article

 

 

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Accept after minor revisions. The manuscript contributes meaningfully to the field of renewable energy assessment and aligns well with the journal's scope.

-A few grammatical errors and awkward phrasings (e.g., in the abstract and introduction) should be corrected for better readability.

-Ensure all references are consistently formatted and up-to-date.

Author Response

Hello! The following changes have been made to improve readability:

Line 12 - (RES)

Line 25 - For this purpose

Line 37-38 - Its implementation is impossible without predictive models.

Line 64 - thereby reducing

Line 67 - greenhouse gases, including carbon dioxide

Table 1 - Wind farms are relatively inexpensive

I've also checked the source links to ensure they are correctly placed.

Thank you very much for your feedback.

Author Response File: Author Response.pdf

Back to TopTop