Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors conducted a performance comparison of LSTM and Echo State Networks to predict the solar power generation. While the study contributes to the state of knowledge by investigating a promising candidate for this task (ESN), there are several major concerns to be addressed, before this study could be considered for publication. Additional experiments/simulations are definitely needed for a positive evaluation:

Major concerns: 

• There is no information to be found on the origin of the solar power data. Is it a simple physical representation of the parameters measured (ambient temperature, solar irradiation, etc.) or was an actual solar cell employed for retrieving the data of the study? There is no discussion of regional intricacies. The solar power signal appears to be very periodic, assuming that the weather conditions are also quite stable. This indicates that the developed models are inappropriate to use in regions with more frequent climatic changes (clouds, rain, etc.). The authors must discuss these aspects relevant to an application of their models. A simple comparison of two models for a very specific regional use case currently seems too weak to warrant publication.
• There is no provided information on how the data was split into training and test periods. What was the forecast horizon? How was the gridsearch conducted? Was the statistical robustness assessed by performing cross validation/Monte Carlo analysis? This has to be considered. 
• For a more comprehensive comparison, it would be recommended to add a Transformer based forecasting method, as well as a conventional method (SVR, Gradient Boosting Regression). This would strengthen the overall message of the manuscript, as it allows to better put the results in perspective.
• Currently, there are no comparisons of the achieved results to literature. This is also important for a reader to better put your results in perspective.
• The overall figure quality must be improved to better convey the essence of your results. 3D-graphs are unnecessary and often depict redundant information. 
  • Figure 1: the color bar indicating different times (hours) is not sensible here. The color should rather indicate variations in power. Subplots with V-I lineplots for various time scales would be a more appropriate option. 
  • Figure 2: Add weights and biases according to the equations 3-8 into the sketch.
  • Figure 4: The authors should consider showing only 5-7 days instead of a whole month. This would make it easier to compare the individual models. Same goes for Fig. 5. 
  • Figure 6: Scatter plots are more appropriate in a subplot layout. The 3D-plot is frankly not informative like that. 
• The methods section should be named Methods and Materials and it should be subdivided into sensible subsections. Currently, the entire manuscript appears not very well structured. 

Minor remarks:

• Indices and units in equations are NOT to be written in italic.
• Numeric results should be given in the abstract.
• Variable x is never properly introduced in equations 9-12.
• Introduction of RMSE actually belongs into the Methods section. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

• The topic of solar power forecasting using LSTM and ESN is relevant, but the paper in its current form does not sufficiently highlight novelty. Please clarify what is new compared to previous studies that have already compared these models.

• The dataset used (30 days of hourly data) is very limited. Results cannot be generalized unless larger, more diverse datasets (covering seasonal variations and multiple sites) are tested.

• The ESN model was carefully tuned, while the LSTM seems to be used with fixed hyperparameters. For a fair comparison, similar optimization efforts should be applied to both models.

• Reported RMSE values for ESN (close to 0.0068) appear unrealistically low. Please provide details on data preprocessing, normalization, and whether results were validated on out-of-sample or cross-validation datasets.

• The section on I–V and P–V curves of solar cells is mostly standard background material and does not directly contribute to the forecasting experiment. Consider shortening this section or moving it to supplementary material.

• Figures require improvement: axis labels, units, legends, and scales should be added for clarity. Comparative results should be shown side by side with statistical measures.

• The discussion section needs to go beyond reporting lower RMSE. Please explain why ESN may have performed better in this case, and under what conditions LSTM could still be preferable.

• Practical implications for grid operation and forecasting practice are not addressed. Please link the findings to real-world applications and discuss how results could be extended or benchmarked.

• The conclusion currently overstates the findings. Please moderate claims and emphasize the limitations, particularly the small dataset and limited validation.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed the majority of the author's concerns. Even though they refrained from training additional models (which could've improved the merit of the manuscript further), the reviewer believes it is justifiable, as the results of the ESN are quite good. 

Publication of the manuscript can be warranted, after the following minor issues (style and formatting related have been addressed: 

• Fig. 1 (c and d): For the winter period the 3D-plot is just not suitable. It is impossible to distinguish lines in the front from lines in the back. I would still recommend the authors to change this to several 2D-plots of Current/Power vs. Voltage across different time periods (Current/Power vs. Voltage from 21-18 hours, from 18-15 hours,...). I believe that this would enhance clarity. If the authors are not doing these changes, I would at least also like to refer them to a book on figure styling, which I find extremely helpful to improve conveying messages via artwork: https://clauswilke.com/dataviz/index.html
• Formatting of Figures in general should be consistent. Same fontsizes should be applied everywhere. Figs. 5 and 6 are quite distorted and should be polished... Most of the figures are blurry as they have been imported to the manuscript in low resolution. All figures should be at least 600 dpi, please adjust. 
• In the equations and in-text formulae, a lot of indices are still written in italic. T_ref is incorrect, T_ref is correct. Please apply this convention consistently throughout the manuscript.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Updates in the manuscript are satisfactory.

Author Response

All issues have been resolved.