Weighted Average Ensemble-Based PV Forecasting in a Limited Environment with Missing Data of PV Power
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe article concerns the current scientific issue of PV power generation forecasting in the situation of missing data. This is a common problem and coping with it is necessary for effective management of PV sources, their integration with the power system and increasing their share in electricity production. Such solutions support the energy transformation process. The authors present the valuable and original approach of Weighted Average Ensemble-Based PV Forecasting. The study is well-written and conducted with high scientific quality using adequate methods. Nevertheless, I have several minor suggestions for its improvement listed below.
1. Figure 4: The quality of the Figure should be improved.
2. Line 399: There is no Table 5 in the article. The authors mean Table 4, I suppose.
3. Conclusion: Has the presented methodology any constraints or limitations affecting the obtaining results? If yes, please describe.
4. Conclusions: What are planned or predicted future research directions in this field of knowledge? Please, describe.
Author Response
Please refer to the attached file.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper focused on Weighted Average Ensemble-Based PV Forecasting in a Limited Environment with Missing Data of PV Power. The problem statement of the study centred around the uncertainty of forecasting caused by volatility around PV power. This necessitated modelling and there is another issue of missing data during the modelling. Although, there exist imputation models such as KNN, MICEs, GAINs with vary degree of success and set-back. Authors proposed four methods—linear imputation (LI), multivariate imputation by chained 133 equations (MICE), K-nearest neighbor (KNN), and generative adversarial imputation 134 networks (GAIN)—for imputing the missing PV power.
Comments for Authors:
Abstract section: The abstract looks good but the key results and conclusion need to be added. The word count is less than 200 words. It should be increased with the additions to about 250 words.
Introduction section: The first two paragraphs laid the foundation for the topic. However, they lack key dataset. There is no mentioned of key statistics related to the topic. The references are also “old”. Although, 2020 may not be old but for a topic of this nature, newer references should be used. The introduction contains about 24 citations with only about 3 of such having 2023 and one for 2024 and vast majority being from 2020. Including a key reference used ie reference “17 Lindig, S.; Louwen, A.; Moser, D.; Topic, M.; Outdoor PV system monitoring—input data quality, data imputation and filtering 475 approaches. Energies, 2020, 13, 5099” which is from 2020. Authors should use a reference from 2024 or 2023 to replace “17” and increase the number of references from 2023 to 2024 in the introduction.
Also in the introduction, authors asserted that there is limited study on “analysing of the impact of missing PV data on forecasting models”. However, there exist references such as: Kim, T., Ko, W. and Kim, J., 2019. Analysis and impact evaluation of missing data imputation in day-ahead PV generation forecasting. Applied Sciences, 9(1), p.204.; Li, Q., Xu, Y., Chew, B.S.H., Ding, H. and Zhao, G., 2022. An integrated missing-data tolerant model for probabilistic PV power generation forecasting. IEEE Transactions on Power Systems, 37(6), pp.4447-4459. Romero-Fiances, I., Livera, A., Theristis, M., Makrides, G., Stein, J.S., Nofuentes, G., de la Casa, J. and Georghiou, G.E., 2022. Impact of duration and missing data on the long-term photovoltaic degradation rate estimation. Renewable Energy, 181, pp.738-748. And many others from 2024 to 2023 including “Lee, D.S. and Son, S.Y., 2024. PV forecasting model development and impact assessment via imputation of missing PV power data. IEEE Access.; Yu, L., Li, M. and Liu, X., 2024. A two-stage case-based reasoning driven classification paradigm for financial distress prediction with missing and imbalanced data. Expert Systems with Applications, p.123745.”
I think the key emphasis by authors should be in combining multiple models and not lack of body of knowledge in the study field.
Methodology Section: The figure 1 needs to be re-arranged to enhance the readability of the graphs. Parts of Figure 4 x-axis label is missing. It should be included.
Discussion of result: The results are well presented. However, authors need to compare their results with references from 2023 to 2024 to ascertain contribution.
A recommendation section should be added by the authors.
Author Response
Please refer to the attached file.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for Authors1. A comprehensive proofreading must be conducted for the text of the manuscript by a technical native person.
2. In the abstract, the authors should give necessary discussions in deep about the innovative methodology and findings of the proposed combined CNN-GRU algorithm that your theoretical research is trying to present rather than the long problem definition.
3. The presentation is of the problem definition in the abstract is too wordy. Try to conclude the key novelties and contributions and convey them precisely to the readers. There is much more work for the authors regarding English presentation.
4. From publishing ethics, you have to mention XXXX et al [10], YYYY et al. [11], ….etc, instead of Reference [10], Reference [11], Reference [12]
5- The review section has limited bibliographic. I would recommend strengthening your literature review by adding recent papers highlighting the applications of artificial intelligence methods in different solar power system simulations : https://doi.org/10.1016/j.applthermaleng.2021.117055 and https://doi.org/10.1016/j.est.2023.109533
6. In the exiting introduction, it is a not clear what are the advantages and research gaps between the previous review studies and this survey study.
7. It is necessary to discuss the machine learning methods' pros and cons, such as computational time and other limitations.
8. It is better to present and explain the mathematical representation of the forecasting methods other than the general description.
9. How to select the parameters of CNN-GRU network, please give detailed explanation.
10. Please justify the obtained results comprehensively.
11. Comprehensive statistical analysis is highly required to confirm the accuracy the proposed regression model.
12. Please compare the results obtained in the proposed model with the findings from the literature for the previous single imputation models.
13. It is also better to support the conclusions with the most important quantitative outputs that show the improvements obtained in the proposed combined algorithm.
14. It can also be precious for readers to make suggestions to continue the research path in the next steps
Comments on the Quality of English Language1. A comprehensive proofreading must be conducted for the text of the manuscript by a technical native person.
2. In the abstract, the authors should give necessary discussions in deep about the innovative methodology and findings of the proposed combined CNN-GRU algorithm that your theoretical research is trying to present rather than the long problem definition.
3. The presentation is of the problem definition in the abstract is too wordy. Try to conclude the key novelties and contributions and convey them precisely to the readers. There is much more work for the authors regarding English presentation.
4. From publishing ethics, you have to mention XXXX et al [10], YYYY et al. [11], ….etc, instead of Reference [10], Reference [11], Reference [12]
5- The review section has limited bibliographic. I would recommend strengthening your literature review by adding recent papers highlighting the applications of artificial intelligence methods in different solar power system simulations : https://doi.org/10.1016/j.applthermaleng.2021.117055 and https://doi.org/10.1016/j.est.2023.109533
6. In the exiting introduction, it is a not clear what are the advantages and research gaps between the previous review studies and this survey study.
7. It is necessary to discuss the machine learning methods' pros and cons, such as computational time and other limitations.
8. It is better to present and explain the mathematical representation of the forecasting methods other than the general description.
9. How to select the parameters of CNN-GRU network, please give detailed explanation.
10. Please justify the obtained results comprehensively.
11. Comprehensive statistical analysis is highly required to confirm the accuracy the proposed regression model.
12. Please compare the results obtained in the proposed model with the findings from the literature for the previous single imputation models.
13. It is also better to support the conclusions with the most important quantitative outputs that show the improvements obtained in the proposed combined algorithm.
14. It can also be precious for readers to make suggestions to continue the research path in the next steps.
Author Response
Please refer to the attached file.
Author Response File: 
Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors advocate for the use of a weighted average ensemble model to address the missing data in PV power. The primary points raised are as follows:
1) As the authors gathered data from an operational PV plant, it is logical to utilize this real data to address the issue of missing data.
2) The missing data is categorized as random or continuous, with the paper only addressing random missing data. It is suggested to also evaluate the performance of the proposed missing value imputation for continuous missing scenarios.
3) The inclusion of weather forecast data in Figure 1 is noted. However, the utilization of weather forecast data for imputing missing data is not clearly explained. For instance, solar irradiance data may aid in the imputation process.
4) Since the primary novelty of this paper lies in the proposed weighted average ensemble model, it is important to thoroughly explain the weighted average ensemble method. Section 2.2.2 should be significantly expanded.
5) The main methodology has already been outlined in the IEEE Access paper: https://ieeexplore.ieee.org/document/10387673
Author Response
Please refer to the attached file.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsThe current state of this manuscript is improved, and reviewers' comments are met. Hence, it is recommended to be accepted
Comments on the Quality of English LanguageThe current state of this manuscript is improved, and reviewers' comments are met. Hence, it is recommended to be accepted
Author Response
Thank you for the kind review. We carefully checked the manuscript to fix typos and errors again.
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors have addressed most of the comments. My main concern is the practical implementation applicability of the proposed imputation model. As the weighted average ensemble method requires running several sub-models (i.e. LI, MICE, KNN, and GAIN), which is supposed to significantly increase the computational time. It is suggested to include the computational time required for each considered imputation method in the case study section.
In addition, what's the theoretical background supporting the selection of LI, MICE, KNN, and GAIN as the submodels? It is suggested to add the performance comparison between LI+KNN+GAIN+MICE (the one proposed in the paper) and LI+KNN+GAIN (the one proposed in the authors' IEEE Access paper).
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 3
Reviewer 4 Report
Comments and Suggestions for AuthorsThank you for addressing all of my comments. I have no further issues with the updated manuscript.
