Comparison of Feedforward Perceptron Network with LSTM for Solar Cell Radiation Prediction

Round 1

Reviewer 1 Report

Review of “Comparison of Feedforward Perceptron Network with LSTM 2 for Solar Cell Radiation Prediction”

 

 

 

1). Abstract – not very solid. The entry sentences are so basic as to be useless for a technical paper. Further, the abstract should have some concrete results, e.g., the best LSTM model yields an MAE of _____.  Additionally, the authors should describe what is unique about their study. There have been a lot of prior work in predicting the power output from solar PV panels.

With the advancement of technology, the use of renewable energy sources has increased 13 considerably in recent decades, and one of the most popular renewable energy sources is the solar 14 energy. Photovoltaic (PV) panel is used to convert the solar energy into electrical energy. Correct  prediction of solar radiation constitutes a very important step to take advantage of PV solar panels.  WHY? We propose an experimental study to predict the amount of solar radiation using classical artificial 17 neural network (ANN) and deep learning methods. PV panel, and solar radiation data are collected 18 at Duzce University in Turkey.

 

 

2). Introduction: The first paragraph in the introduction is useless. These are obvious comments.

 

Line 43 --- Why has solar become commonly used since 2016? Do you mean to say that there has been  a tremendous growth in solar energy production worldwide since 2016?

 

 

The paragraph that begins on the bottom of page 1 extends to the bottom of page 3.  When a new thought is brought in, you should begin a new paragraph. For example the 1^st sentence on p. 2 shifts the direction by addressing economic issues of solar. This should start a new paragraph.

 

Line 51 on p.2 describes the instability of solar produced energy. I suppose you mean to say intermittency. PV systems are actually fairly stable.

 

 

3). English – There are numerous English and grammatical error throughout the paper.  These need to be fixed.

 

 

4). Fundamentally, I don’t see see what the unique contribution for the paper is. As noted in the extensive literature review section, other researchers have employed the same Machine Learning algorithms to predict solar power production as a function of meteorological data and measured solar panel power output data.

 

The Methodology section simply reiterates how each of the known algorithms works. The authors have not adapted these techniques in any way that I can see.

The tuning of the hyper-parameters is a common approach in ML.

 

In order to have any plausibility for publication, the authors should create a table that compares what they are doing differently relative to what others have done.  The table should describe the meteorological data used and the solar power production used, the ML techniques employed, use of hyper-parameter tuning, etc…., anything to show that there is in fact something unique about their paper.

 

Moreover, if there have been improvements in the techniques relative to the work of others, these should be highlighted only in the Results section. The results should show the value of their contributions.

 

As it stands now, I cannot accept this paper.

 

 

 

Author Response

Dear Reviewer, 

Thank you for your valuable comments and intriguing questions which helped us to improve our work. I hope that the attached file with the responses to your comments would change your decision about our paper.

Thank you,

Author Response File: Author Response.docx

Reviewer 2 Report

In this paper, the authors predicted the amount of solar radiation using classical artificial neural network (ANN) and deep learning models. Daily solar radiation prediction was obtained by using meteorological data and data obtained from 3 different types of solar panels for the city of Duzce in Turkey.

In the following, there is a list of suggestions and questions that the authors should consider:

3. In this manuscript, the authors used ANN and LSTM to predict solar radiation. Why did you use only these models? There are many machine learning and deep learning models that forecast solar radiation well, such as CNN, MLP, RBFN, and GRU.
4. In chapter 3, the authors showed the results for radiation forecasting using an LSTM model. Unfortunately, the authors failed to convince that the presented manuscript contains complete results. The author placed a greater emphasis on training loss and testing loss, which does not yield better outcomes in solar radiation forecasting. I recommend that the authors provide additional findings and present the results in their actual value rather than their normalized value.
5. To better understand results, I recommend that the authors utilize more error measures such as relative root mean square error (rRMSE) and relative mean bias error (rMBE). Furthermore, it would be good if the authors could provide some descriptions to explain when the error is judged substantial, as well as when a model is good enough.
6. Many papers have been published that used ANN and LSTM to predict PV power and solar radiation. What makes this research unique?
7. In general, the result and the discussion sections are relatively short. Limitations of the approach are not discussed and issues related to practical implementation are not discussed. Furthermore, this manuscript appears to be incomplete since it has several errors, such as missing table 3 in the manuscript.

Author Response

Dear Reviewer,

Thank you for your valuable comments and insightful questions which helped us to improve the quality of our research. I hope you would like more the updated manuscript.

Thank you,

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript describes a study about using Artificial Neural Network (ANN) and Long-Short-Term-Memory (LSTM) methods to predict the solar radiation level in the Duzce region of Turkey.

The manuscript is an interesting one but in order to be considered for publication some improvement should be done, as is mentioned further:

 

In the abstract, the abbreviation is recommended to be avoided. The abbreviation should be introduced when it appears for the first time within the paper body. Also, in the Abstract section the obtained results should be better highlighted.

In the Introduction section, a better critical analysis of the recently literature (three-four years) should be provided. The citation of the literature as a list should be avoided, without mentioning the contribution of each cited reference.

At page 2, lines 95-98 it is mentioned: “Y. Jung et al. [19] used the data obtained from 164 PV plants such … The authors proposed … of PV solar power [16].” The two papers do not have the same authors. The authors should check and correct the cited papers or correct the text.

At page 11, line 278 x^2' is mentioned. I think it should be x'.

At Page 11, lines 380-394, there is a part of the template. The authors should delete it.

In my opinion, section 4 entitled “Discussion” is more like a Conclusion section. Therefore, it should be renamed and improved. Section 3 should be named Results and Discussions and should offer more details about the results obtained during this study (e.g. a comparison between the results obtained through this study and others from literature in terms of MSE and R² even if the geographical and weather are different, the evolution of the accuracy in terms of MSE and R² for 24, 48 and 72 hours…).

The cited references should be correctly edited using the journal stile (e.g. reference [11] is an article, but it is cited as a website, moreover the link doesn't even work).

Author Response

Dear Reviewer,

Thank you for your valuable comments and insightful questions which helped us to improve the quality of our research. I hope you would like more the updated manuscript.

Thank you,

Author Response File: Author Response.docx

Reviewer 4 Report

The accurate prediction of solar radiation is an important step to fully take advantage of PV solar panels. In this work, the classical artificial neural network (ANN) and deep learning methods are utilized to predict the amount of solar radiation, and the performances are compared. The research topic is interesting. Here are the detailed comments.

1) The manuscript contains lots of grammar mistakes, incorrect or missing words, etc. The text must be carefully proofread.

2) The LSTM has been widely used in prediction of energy system behavior. The review about the LSTM application should be further extended. Here are some examples which may be useful.

[1] Research on Simulation and State Prediction of Nuclear Power System Based on LSTM Neural Network, Science and Technology of Nuclear Installations, 2021.

[2] Prediction of remaining mileage of hydrogen energy vehicles based on LSTM, 4th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE), 2021

3) The preprocessing for the input data is usually required for the LSTM and ANN, such as the data normalization. The detailed information about the data preprocessing should be provided.

4) Which kind of ANN is used in this work? It should be specified, and the detailed information about the network should be illustrated.

Author Response

Dear Reviewer,

Thank you for your valuable comments and insightful questions which helped us to improve the quality of our research. I hope you would like more the updated manuscript.

Thank you,

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Review of Applied Sciences Paper # SSN 2076-3417

 

My previous review noted that the paper was well-written, but I just didn’t see any significant or even any insignificant contribution to the literature.  My position is unchanged after edits.

 

The authors have added the following to describe their contributions.

 

The main contributions of the proposed work can be summarized as: (i) conduct a comparison between the performance of the most common deep learning models in the  literature, (ii) build an LSTM to accurately predict the solar radiation at the city of Duzce in Turkey with the potential to be generalized to mode cities around the world, and (iii) Conduct a comparison between our results in terms of the coefficient of determination 163 (R2 ), root mean squared error (RMSE), mean biased error (MBE), and mean absolute error 164 (MAE).

 

 

Relative to (i) the authors have improved the literature review of ANN related to solar power prediction of PV solar panels. But my previous review suggested the inclusion of a table that documents for each researcher the data features used to predict the solar power generation, and details about the ANN architecture. This would be a way to more clearly differentiate relative to the work of others (or not, if there’s nothing unique, which, from my perspective is the case). Moreover, the claim of a comparison to the work of others isn’t accurate. Comparisons were made only to the work of references [58] and [59]. Frankly, it would be easy to apply the developed approach to each of the locations and for the noted time periods of each the referenced authors. This would truly represent a comparison.

 

Secondly, in the introduction the authors mention the intermittency of the power in Ankura. This might represent an interesting angle for the work; namely, how the solar production at scale could impact the intermittency. Right now developing a model for one location and offering that as a unique contribution is insufficient for publication.  The model could be applied in any location. The only thing that changes are the solar inputs.

 

I would urge the authors to consider providing Ankura power consumption curves, and show and discuss how the intermittency is improved for this location. This might offer some ability for others to extend to their locations.

 

Third, the abstract and background section  continues to add obvious information. For example, the abstract starts with the idea that solar energy is the most popular renewable energy. This isn’t needed in a technical paper.    I would build upon the idea of intermittency in Ankura and other locations around the world as a driver for the research.

 

Fourth, the description of ANNs and LSTM could be found in Wikipedia. There isn’t anything new here. Couldn’t much of this section be referenced to other work.

Author Response

Dear Reviver,

Thank you for your valuable comments.

Kindly find our responses to your comments attached to this email.

Regards,

Tugba Ozdemir

Author Response File: Author Response.docx

Reviewer 2 Report

1. The authors have to explain why ANN and LSTM are used to predict solar irradiation. The authors should discuss the strengths of these models in comparison to other machine learning models.

2. I appreciate that the authors added more results in terms of RMSE, MBE, and MAE to forecast solar irradiance, as well as a comparison to other similar models from other studies. However, these comparison results will confuse the reader. In terms of R², the author's model outperforms previous studies, but not in terms of RMSE, MBE, or MAE. I think this comparison does not demonstrate that this study is better than others.

3. Overall, I think that results and discussion are still very concise. In addition, there are too many statements that confuse readers. For example the authors explain the study's limitation in chapter 4 by noting that additional machine learning models such as SVM, RFM, and GBT were not investigated. If this is actually the limitation in this study, the authors should overcome this limitation first in order to determine which machine learning model is better.

Author Response

Dear Reviver,

Thank you for your valuable comments.

Kindly find our responses to your comments attached to this email.

Regards,

Author Response File: Author Response.docx

Reviewer 4 Report

No further comment

Author Response

Dear Reviver,

Thank you for your valuable comments, and for helping us to improve the quality of our work.

Regards,