A Deep Learning Approach for Peak Load Forecasting: A Case Study on Panama
Round 1
Reviewer 1 Report
The paper in general is well structured, well presented, and has an important research topic to be studied.
1 - The review on the adopted methods is shallow, invest here works were carried out with hybrid techniques that stand out from the models presented. I recommend a further review on the topic including hybrid forecasting techniques.
2 - Why was the dataset divided into 80% of data for training, is there a specific motivation for this?
4 - In subsections 6.1, 6.2, 6.3, and 6.4 you repeat meanings of “n” and “yi”. I recommend creating only one subsection presenting the metrics that were used to evaluate the network, with the presentation of all the meaning of the variables only once.
5 - There are studies that show that algorithms based on combinations such as ensemble have better performance than LSTM. It would be interesting for you to present a benchmarking to prove that LSTM is better for your application, or justify your choice in a specific way in relation to other techniques.
6 - The conclusion has only one paragraph, try to divide the ideas and explain them separately so as not to make the reader tired.
Finally, I think there are minor adjustments that can improve the paper. I leave as a suggestion for reading the papers: Hybrid Deep Learning for Power Generation Forecasting in Active Solar Trackers, Fault Detection in Insulators Based on Ultrasonic Signal Processing Using a Hybrid Deep Learning Technique, that present hybrid techniques, and the paper Electricity Price Forecasting Based on Self-Adaptive Decomposition and Heterogeneous Ensemble Learning that present the ensemble for a similar problem.
Author Response
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
Manuscripts topic relies on Journals’ scope and will interest readers. It could be published as it is.
Author Response
Dear reviewer,
The authors would like to thank you for the valuable feedback provided and for finding our paper suitable for the journal.
Regards,
Bibi Ibrahim
Corresponding author
Reviewer 3 Report
This paper presents forescasting models to predict the future peak demand in power grids taking into account the new loads paradigm, like EVs charging. It is an interesting and current topic and may deserve readers' attention. The state of the art, can be further investigated with some new references must be add.
Before presenting the case study (section 3), the authors must include a the section with the approach and in-depth explanation of the methodology and model used (section 5). It would be the most suitable structure for the paper.
The results shows the suitability of the models used, but it would be interesting to consider different kind of measures for long-term approach, like peak shaving including renewables and Energy Management System to do displacement of non-priority loads for off-peak times.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 4 Report
The paper presents the application of deep learning techniques to the forecasting of the monthly energy demand at national level. The evaluated approaches are based on convolutional neural networks (CNNs) and long-short term memory networks (LSTMs). The paper evaluates different features to feed the models, and comparatively assess the performance of the various solutions. The evaluation uses historical energy and other relevant information in Panama.
The paper is very well written and easy to read. It uses a good balance between mathematical notations and textual descriptions. The quality of the English language is very good.
From the technical viewpoint, the novelty resides in the application of deep learning techniques that are usually used in the field of image processing to the domain of time series forecasting (while LSTMs are instead usually targeted to this domain). Some proposed models prove to be rather accurate in the prediction. On the other hand, the forecasting is only 1 month ahead, which helps a lot in obtaining accurate results.
All the models are not able to predict the high peak that happened in August 2019. Is there a possible explanation for this behavior? It can be argued that the features used for the prediction are not capturing the reasons of the peak. Is there any insight about the motivation of the peak, which could explain why the models and the features are not able to predict it?
Overall, the paper has merits of being clear and providing a viable solution for short term forecasting. However, this does not match with the goal stated in the paper (abstract and introduction), where the authors claim to address a long term forecasting. This seems not the case of the results shown in the paper.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 5 Report
This paper proposes a deep learning method (CNN) for peak load forecasting. The authors used convolutional neural networks (CNNs) to make long-term forecasting model for accurate prediction of peak demand. Also, they surveyed diverse researches related to peak demand forecasting. To show validity of this paper, the authors carried out case study using the data from Panama’s power system.
The authors well explained their proposed methods, in addition, the experimental results represented the performance of the proposed methods. I expect this paper will contribute to various areas for peak demand forecasting.
Author Response
Dear reviewer,
We would like to thank you for the time and effort dedicated to providing valuable feedback concerning our manuscript. Once again, thank you for expressing your interest in our topic.
Regards,
Bibi Ibrahim
Corresponding author
Round 2
Reviewer 3 Report
The paper was substantially improved. All my comments have been addressed.
