Research on Combined Model Based on Multi-Objective Optimization and Application in Wind Speed Forecast
Round 1
Reviewer 1 Report
English should be improved. Sometimes it is difficult to understand what authors want to say. There are many sentences that are not correct. In some paragraphs the same word is repeated several times. This should be avoided.
The major contributions of the work described at the end of “1. Introduction” should be included in “7. Conclusion”.
It is not clear which is the model the authors propose. In lines 95-97 they stated that the model comprises several forecasting tools, but they do not specify which of then make up their model.
The model the authors propose should be clearly and briefly described in “2. Strategy of the Proposed Combined Model”.
Data denoising is described in lines 146-147 and, again, in lines 149-150.
The authors should describe why the four tools used in denoising are used. That is to say, why those four steps are needed.
The models described in chapter 5 are difficult to understand for a non-expert. They should be described in a more simple way. Some of the mathematical expression describing those models are complex and difficult to understand.
Which is the meaning of equation (3)? What is it used for in the neural model?
The model the authors propose is described in chapter 5. Nevertheless it is not clear which are the elements that make it up. This chapter is confusing.
A “Definition 3” appears in line 307 before a “Definition 1” appears in line 317.
Which is “Definition 2”? (line 323)
The author proposes three error indices to define a multiobjective optimization problem (MOP). Nevertheless they are merely three error measurements, therefore optimizing one implies that the other two are also optimized. A MOP demands that several different performance indices should be minimized, and here only a performance one (expressed in three different ways) is minimized.
Which parameters of the models used in the combined one proposed by the authors are adjusted with the NSGA-III algorithm?
From the data provided by the authors it may be inferred that the combined model they propose are only slightly better than CSBPNN or DE-OSELM.
What is the point of using the combined model when two more simple ones provide results as good as those provided by that combined one?
Therefore I suggest the authors to rewrite their work without the combined model they propose, because the results provided by models CSBPNN or DE-OSELM are good enough to deserve to be published.
Author Response
English should be improved. Sometimes it is difficult to understand what authors want to say. There are many sentences that are not correct. In some paragraphs the same word is repeated several times. This should be avoided.
The major contributions of the work described at the end of “1. Introduction” should be included in “7. Conclusion”.
Respond: Thanks for your suggestion. I have described the contribution in Conclusion of the paper (line 568-584 and marked in yellow).
It is not clear which is the model the authors propose. In lines 95-97 they stated that the model comprises several forecasting tools, but they do not specify which of then make up their model.
Respond: Thanks for your suggestion. I have rewritten the sentence (line 100-107 and marked in yellow).
The model the authors propose should be clearly and briefly described in “2. Strategy of the Proposed Combined Model”.
Respond: Thanks for your suggestion. I have rewritten this part. We presented the theory our proposed combined model based on and the branch models include in our proposed model (line 140-152 and marked in yellow).
Data denoising is described in lines 146-147 and, again, in lines 149-150.
The authors should describe why the four tools used in denoising are used. That is to say, why those four steps are needed.
Respond: Thanks for your suggestion. It is our fault; we have deleted the repeat part and describe this part more clearly (line 154-160 and marked in yellow).
The models described in chapter 5 are difficult to understand for a non-expert. They should be described in a more simple way. Some of the mathematical expression describing those models are complex and difficult to understand.
Respond: Thanks for your suggestion. We have moved the introduction of NSGA-III and pseudocode to the Appendix (line 592-617 and marked in yellow).
Which is the meaning of equation (3)? What is it used for in the neural model?
Respond: Thanks for your suggestion. We have explained the meaning of the equation in line 209-210 and marked it in yellow.
The model the authors propose is described in chapter 5. Nevertheless it is not clear which are the elements that make it up. This chapter is confusing.
Respond: Thanks for your suggestion. In this part, we describe the theory of combined model (line 314-325), MOP theory (line 327-337), simple introduction of NSGA-III (line 354-358) and objective functions (line 344-352). The branch models and the flow chart are presented in Figure 3 (line 312).
A “Definition 3” appears in line 307 before a “Definition 1” appears in line 317.
Which is “Definition 2”? (line 323)
Respond: Thanks for your suggestion. It is our fault; we reorder the definitions. (Definition 1 is in line 319-325, Definition 2 is in line 330-333, Definition 3 is in line 337, Definition 4-8 are in line 401-411 and we marked these in yellow)
The author proposes three error indices to define a multiobjective optimization problem (MOP). Nevertheless they are merely three error measurements, therefore optimizing one implies that the other two are also optimized. A MOP demands that several different performance indices should be minimized, and here only a performance one (expressed in three different ways) is minimized.
Respond: Thanks for your suggestion. As you said optimizing one implies that the other two are also optimized, but according to our Experiment V which ran the model 100 times, for example, the average MAPE value was lowed by the different error indices. However, in the single experiment it is not stable, the range of MAPE are large. But our proposed model with three objective functions has a small interval of MAPE
Which parameters of the models used in the combined one proposed by the authors are adjusted with the NSGA-III algorithm?
Respond: Thanks for your suggestion. In this paper, we optimize the weights of the branch models. And we also added this part in line 104-105, 149-151 and marked in yellow.
From the data provided by the authors it may be inferred that the combined model they propose are only slightly better than CSBPNN or DE-OSELM.
Respond: Thanks for your suggestion. In this paper, we are not only concentrate on the accuracy, but also the stability. The average results of CSBPNN or DE-OSELM may obtain a good result for accuracy. But due to the reason that the randomness of artificial methods, the results were not stabile which is shown in Table 5, Experiment IV (line 498-516 and marked in yellow). So we proposed a model which can achieve both high accuracy and strong stability.
What is the point of using the combined model when two more simple ones provide results as good as those provided by that combined one?
Therefore, I suggest the authors to rewrite their work without the combined model they propose, because the results provided by models CSBPNN or DE-OSELM are good enough to deserve to be published.
Respond: Thanks for your suggestion. In this paper, we are not only concentrate on the accuracy, but also the stability in wind speed forecasting problem. CSBPNN or DE-OSELM may achieve a good average results for accuracy, but because of the randomness of artificial methods, the results were not stabile which is shown in Table 5, Experiment IV (line 498-516 and marked in yellow). Due to this reason, we proposed a combined model to obtain both accuracy and stability. And we will consider your opinions in future work and write an article with CSBPNN or DE-OSELM.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This paper proposes a new combined model that integrates non-positive constraint theory, two nonlinear hybrid neural networks (CS-BPNN and DE-OSLEM), two linear models (ARIMA and HW), and NSGA-III. Applied to wind speed data from four wind farm sites. Results of three combined model versions with two objective functions are compared to test the performance.
The added value of this paper is the extraction of signals representing different components of the original time series, such as long-term trend signals, periodic signals and noise signals.
The proposed combined model is based on the MOP theory, which is more accurate and stable.
The combination of linear models (ARIMA and HW) and non-linear models (BPNN and OSELM) address the linear and non-linear properties of wind.
The paper is well structured by very poorly written. References are exhaustive and recent and both the methods and results are detailed.
The authors should have the paper revised by a native English speaking person or by an editing service.
Authors should summarize the methods and principal findings in the abstract
The List of abbreviations should appear at the end of the paper
Code (i.e. p 182 and p 375) should be in annex.
Improve visibility of small font text in figure 3.
Abstract and Conclusions must explicitly state (by quantifying it) the improvement of the proposed model against other models
The paper may be considered for publication after the major revisions mentioned above
Author Response
This paper proposes a new combined model that integrates non-positive constraint theory, two nonlinear hybrid neural networks (CS-BPNN and DE-OSLEM), two linear models (ARIMA and HW), and NSGA-III. Applied to wind speed data from four wind farm sites. Results of three combined model versions with two objective functions are compared to test the performance.
The added value of this paper is the extraction of signals representing different components of the original time series, such as long-term trend signals, periodic signals and noise signals.
The proposed combined model is based on the MOP theory, which is more accurate and stable.
The combination of linear models (ARIMA and HW) and non-linear models (BPNN and OSELM) address the linear and non-linear properties of wind.
The paper is well structured by very poorly written. References are exhaustive and recent and both the methods and results are detailed.
The authors should have the paper revised by a native English speaking person or by an editing service.
Respond: Thanks for your suggestion. We have done the English-editing(english-edited-7117).
Authors should summarize the methods and principal findings in the abstract
Respond: Thanks for your suggestion. I have modified the abstract of the paper (line 14-31 and marked in yellow).
The List of abbreviations should appear at the end of the paper
Respond: Thanks for your suggestion. I have moved the abbreviations to the end of the paper (line 586 and marked in yellow).
Code (i.e. p 182 and p 375) should be in annex.
Respond: Thanks for your suggestion. I have moved the code to the Appendix of the paper (line 591 and 616 and marked them in yellow).
Improve visibility of small font text in figure 3.
Respond: Thanks for your suggestion. We have improved Figure 3 (line 311)
Abstract and Conclusions must explicitly state (by quantifying it) the improvement of the proposed model against other models
Respond: Thanks for your suggestion. I have explicitly stated the improvement of our proposed model in Abstract and Conclusion (line 14-31 and 568-684 and marked in yellow).
The paper may be considered for publication after the major revisions mentioned above
Respond: Thanks for your suggestion.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
English should be revised again. In some paragraphs the same word is repeated too much times. For example (for the sake of simplicity only two are provided):
“model” is repeated nine times in the “Abstract”, four times in line 54, ten times in paragraph 56-68,…
”objective function” has been repeated three times in one line (95).
This should be avoided.
The major contributions of the work described at the end of “1. Introduction” should be removed from this chapter and included in “7. Conclusion”.
Equation (3) has not been explained as requested. Authors should explain what it is used for in the neural model.
Reviewer 2 Report
The authors have satisfactorily addressed my comments. The paper may be considered for publication after only minor language corrections.
Round 3
Reviewer 1 Report
The paper may be published in present form.
Reviewer 2 Report
The paper may be considered for publication
