Training and Testing of a Single-Layer LSTM Network for Near-Future Solar Forecasting

Round 1

Reviewer 1 Report

This article presents the design, training, and testing of a single-layer long-short-term-memory (LSTM) artificial neural network for day-ahead power forecasting of a single PV system in Cyprus. The reviewer has the following comments:

• The idea is interesting. However, several works on LSTM is recently published. How is this work different from other existing papers? From the paper, the only difference is the location of the PV system that is Cyprus. This does not seem a big contribution. Please clearly outline the contributions of the paper.
• The literature review presented in this paper is not enough. Please review more papers in the introduction section. Some of the good papers are given for authors consideration:

1. Khan, Irfan Ahmad, Adnan Akber, and Yinliang Xu. "Sliding window regression based short-term load forecasting of a multi-area power system." In 2019 IEEE Canadian Conference of Electrical and Computer Engineering (CCECE), pp. 1-5. IEEE, 2019.
2. Guesmi, Tawfik, Anouar Farah, Ismail Marouani, Badr Alshammari, and Hsan Hadj Abdallah. "A new chaotic sine cosine algorithm for chance-constrained economic emission dispatch problem including wind energy." IET Renewable Power Generation(2020).
3. Ellahi, Manzoor, Ghulam Abbas, Irfan Khan, Paul Mario Koola, Mashood Nasir, Ali Raza, and Umar Farooq. "Recent approaches of forecasting and optimal economic dispatch to overcome intermittency of wind and photovoltaic (PV) systems: A review." Energies12, no. 22 (2019): 4392.

• The results provided are not enough and don’t reach a convincing conclusion. Please add more results and draw a convincing conclusion to prove the effectiveness of the proposed forecasting scheme.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Report on the manuscript Training and testing of a single-layer LSTM network for near-future solar forecasting, authored by Naylani Halpern-Wight†, Alexandros Charalambedes, Maria Konstantinou, Angèle Reinders.

The paper presents two versions of a single-layer long-short-term-memory (LSTM) artificial neural network model for forecasting the output energy production of a PV. The authors compare the accuracy of their models with the one of another more complex ANN model, concluding that simplicity does not impair the results. The paper is interesting and might be of interest for the researchers working in the solar energy forecasting domain.

The paper is well written in general, the Introduction describes clearly the machine learning and in particular the ANN models. The Data and Methods paragraph is also very clear.

The Results, however, which is the most interesting part (a value of 10.7% for the nRMSE is a very good result), seems to be incomplete and not so well presented.

I recommend the authors to address the following observations:

1. First of all, in the Abstract (line 9) the authors state that “This article presents the design, training, and testing of a single-layer long-short-term-memory (LSTM) artificial neural network for day-ahead power forecasting of a single PV system in Cyprus” while in the Results section (line 162) they say that the model “forecasted up to 1.5 hours  into the future”. The time horizon of the forecast should be clear.

2. I suggest the authors to expand the Results part:

• For the assessment of the models' accuracy I recommend nMBE (normalized mean bias error), to be added to the statistical indicators chosen by the authors (RMSE and nRMSE).
• I strongly recommend a graphical representation of the results of the forecasting: Forecasted versus Measured data, or both Forecasted and Measured data as time series.
• I recommend the authors to compare their models with some other types of models, for instance persistence or some simple classical statistical models, which are also known to perform very well in the intra hour time horizon.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

None of the reviewer's concerns has been addressed yet. Please write my comments first followed by the authors' response to each of the reviewer's comments.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

I consider that the authors addressed all of my previous observations.

Although, I still have one important observation to make:  In tables 1 and 2, the units for the statistical indicators (RMSE, MBE, MAE, MSE and variance) should be specified. All of the indicators  except for the normalized one and r2 are dimensional quantities. 

With this correction I recommend the manuscript for publication.

Author Response

Dear Reviewer 2,

Thank you for catching the error regarding not putting units on the metrics in tables 1 and 2. Units have now been added.

Thank you for your review,

Sincerely,

Naylani Halpern-Wight

Round 3

Reviewer 1 Report

The authors addressed most of my comments. The paper may be accepted for publication.