Forecasting Models for Wind Power Using Extreme-Point Symmetric Mode Decomposition and Artificial Neural Networks
Round 1
Reviewer 1 Report
1. The first chapter does not elaborate on the advantages of the ESMD method compared to other signal decomposition methods. It is recommended to add the advantages and disadvantages of ESMD and other signal decomposition methods in Chapter 1.
2. In the second paragraph of the Introduction, the description of the smart method is too simple. Many of the artificial intelligence algorithms have been used in wind speed prediction. It is recommended that authors read more relevant literature and compare the advantages and disadvantages of the various algorithms in the method.
3. It is recommended to explain why three different prediction models are to be established for high frequency, low frequency, medium frequency subsequences. Please explain the basis for the different methods selected.
4. In Chapter 3, the structural description of the article is not detailed enough.
5. According to 4.1, the author only used the data fromChinafrom April 24 to May 9, 2016 for experiments. The use of one Chinese wind farm data does not provide a good indication of the effectiveness of the proposed model.
6. The author used 14 days out of the 15-day data as a training set and 1 day as a test set. The author is advised to cite authoritative references to illustrate the rationality of data set partitioning.
7. It is recommended that the author unify the meaning of the full-text alphabetic symbol. For example, X in 2.1 represents raw data, and yi in 4.2 represents raw data, which can be represented by one letter.
8. Figure 2, Figure 3, Figure 5, and Figure 6 each contain 2-3 sub-pictures, but the author does not mark the sub-pictures, and the sub-pictures do not describe the contents of the sub-pictures. This results in a picture with a corresponding text description that is not clear enough to be intuitive. For example, in Figure 6, it should be noted that the three sub-pictures respectively correspond to high, medium and low frequency wind speed subsequences. It is recommended to mark the submap.
9. In Figure 6, the experimental results are not clearly visible. It is recommended to modify the horizontal and vertical coordinates of the image to show the dimension values of the filter more clearly.
10. Figure 8 is a representation of the predictions for all methods. Due to the excessive number of predicted sequences, the advantages of seeing the proposed model cannot be removed. It is recommended to modify Figure 8. At the same time, the author's analysis and description of Figure 8 is too general and does not reflect the necessity of Figure 8.
Author Response
Dear Editors and Reviewers:
Thank you very much for sending us the Reviewers' reports on our manuscript (Sustainability-420436) entitled “Forecasting models for wind power using Extreme-point Symmetric Mode Decomposition and Artificial Neural Networks”. Particularly, we would like to thank the Reviewers for their valuable comments and criticisms.
According to the Reviewers and Editors' recommendations, we have revised carefully our manuscript with red font in Microsoft Office Word. The following is a detailed list of response to all comments and criticisms, and changes the authors have made. If any question arises, please let us know.
Point 1: The first chapter does not elaborate on the advantages of the ESMD method compared to other signal decomposition methods. It is recommended to add the advantages and disadvantages of ESMD and other signal decomposition methods in Chapter 1.
Response 1: We are sorry about the negligence of the advantages and disadvantages of ESMD and other signal decomposition methods in Chapter 1. Therefore, we have rewritten this part and this specific answer is in lines 126 to 139 of the article. At present, the most common signal decomposition methods are mainly EMD, EEMD and some traditional time-frequency transform methods and so on. Although these signal decomposition methods have improved the accuracy of wind power prediction to a certain extent, there are still some defects, such as the mode mixing problem in EMD and the residual noise in EEMD method. To overcome these defects of EMD and EEMD, a novel technique called ESMD proposed by Wang et al is employed for reducing the noise and uncertainty of wind speed and wind power series. Moreover, compared with some classical time-frequency transform methods, ESMD method proposed "direct difference (DI) method” for data, broking through the traditional concept of using integral changes. And, the ESMD method has been applied in many fields, such as climate change issues and seismology.
Point 2: In the second paragraph of the Introduction, the description of the smart method is too simple. Many of the artificial intelligence algorithms have been used in wind speed prediction. It is recommended that authors read more relevant literature and compare the advantages and disadvantages of the various algorithms in the method.
Response 2: The reviewer correctly noted that the description of intelligent methods in Introduction is too simple. After reading a large number of references, the description of the intelligent methods was rewritten as two paragraphs, mainly in lines 63 to 125 of the article. Firstly, we briefly introduce some applications of artificial intelligence methods in wind power and wind speed prediction, such as LSSVM, SVM and RBFNN. Secondly, we focus on the application of BPNN in wind power and wind speed prediction, and point out the advantages and disadvantages of BPNN. Subsequently, we compare some optimization algorithms (PSO, GA and SA) with the bat algorithm and find that bat algorithm can not only optimize BPNN to improve prediction performance, but also have the characteristics of less parameters and strong global optimization ability. Moreover, the mathematical model involved in Bat algorithm is relatively simple and computationally efficient. Finally, we describe the application of ENN in wind power and wind speed prediction, and explain the advantages and disadvantages of this method. To further enhance the prediction ability of ENN for nonlinear, chaotic and volatility data, the Adaboost algorithm is used. This algorithm has made a lot of contributions in wind power and wind speed prediction and achieved better prediction performance, which is mainly due to two aspects. On is that it can improve the prediction performance by combining multiple predictor models; another aspect is that Adaboost algorithm has the advantages of simple calculation and small error.
Point 3: It is recommended to explain why three different prediction models are to be established for high frequency, low frequency, medium frequency subsequences. Please explain the basis for the different methods selected.
Response 3: It is really true as reviewer suggested to explain why three different prediction models are to be established for high frequency, low frequency, medium frequency subsequences. By looking for references, we found that the data characteristics of each characteristic component are different. Among them, PHC is a nonlinear system whose fluctuations are nonlinear and abrupt; the fluctuation of PMC is nonlinear, chaotic and its volatility is relatively moderate; and, the fluctuation of PLC is relatively flat and nonlinear. However, Bat-BP, Adaboost-ENN, and ENN are all suitable for dealing with nonlinear, fluctuating and chaotic data, and have made great contributions to wind power and wind speed prediction. Therefore, to select the best prediction model for each wind power characteristic component, we performed a set of comparative experiments by using Bat-BP, Adaboost-ENN and ENN in each characteristic component. The results of this experiment show that Bat-BP model is the best for PHC prediction. Similarly, Adaboost-ENN model and ENN model are suitable for PMC and PLC prediction, respectively.
Point 4: In Chapter 3, the structural description of the article is not detailed enough.
Response 4: Based on the reviewer's recommendations, we describe in detail the framework of the combined model mentioned in this article. This specific description is in lines 323 to 346 of the article.
Point 5: According to 4.1, the author only used the data from China from April 24 to May 9, 2016 for experiments. The use of one Chinese wind farm data does not provide a good indication of the effectiveness of the proposed model.
Response 5: I also considered this issue mentioned by the reviewer, and I also tried to find data from other wind power plants to supplement the experiment. However, the data involved in this paper was obtained from others, and I tried to find data from other wind power plants, but did not get it. In addition, the data involved in this paper not only has wind power data, but also corresponding meteorological data at different times and altitudes, so the data requirements are relatively high. It can be seen that the acquisition of data from other wind farms is relatively difficult. But, it is worth noting that in the wind power and meteorological data we obtained from others, the data we studied during this time period is continuous, there is no missing, so it can reflect the real wind power. And the data of this wind power plant has also been studied by others (Short-Term Wind Power Forecasting: A New Hybrid Model Combined Extreme-Point Symmetric Mode Decomposition, Extreme Learning Machine and Particle Swarm Optimization), so it is more representative. Therefore, this Chinese wind farm data can provide a good indication of the effectiveness of the proposed model.
Point 6: The author used 14 days out of the 15-day data as a training set and 1 day as a test set. The author is advised to cite authoritative references to illustrate the rationality of data set partitioning.
Response 6: Referring to the reviewer's suggestion, we cited a reference to illustrate the rationality of data set partitioning. For specific modifications, see lines 355 to 357.
Point 7: It is recommended that the author unify the meaning of the full-text alphabetic symbol. For example, X in 2.1 represents raw data, and yi in 4.2 represents raw data, which can be represented by one letter.
Response 7: According to the reviewer's suggestion, we have unified the letters and changed them to.
Point 8: Figure 2, Figure 3, Figure 5, and Figure 6 each contain 2-3 sub-pictures, but the author does not mark the sub-pictures, and the sub-pictures do not describe the contents of the sub-pictures. This results in a picture with a corresponding text description that is not clear enough to be intuitive. For example, in Figure 6, it should be noted that the three sub-pictures respectively correspond to high, medium and low frequency wind speed subsequences. It is recommended to mark the submap.
Response 8: The reviewer correctly pointed out the problem with the Figure in this article. Therefore, we re-examine the Figure in the article and describe it in detail. See Figure 2 (lines 408 to 411), Figure 3 (lines 418 to 422), Figure 5 (lines 451 to 452), and Figure 6 (lines 466 to 468) in the article.
Point 9: In Figure 6, the experimental results are not clearly visible. It is recommended to modify the horizontal and vertical coordinates of the image to show the dimension values of the filter more clearly.
Response 9: According to the reviewer's suggestion, we have redrawn Figure 6, see lines 466 to 468 in the text.
Point 10: Figure 8 is a representation of the predictions for all methods. Due to the excessive number of predicted sequences, the advantages of seeing the proposed model cannot be removed. It is recommended to modify Figure 8. At the same time, the author's analysis and description of Figure 8 is too general and does not reflect the necessity of Figure 8.
Response 10: Considering Figure 8 as part of the results, we are not well represented. Therefore, we redrawn Figure 8 based on the experimental results of the four groups of experiments and described them. The specific modifications are shown in lines 488 to 493 of the article, and lines 565 to 572.
Reviewer 2 Report
The paper presents a method for forecasting the power produced by a wind farm. The method begins with the analysis of input data, in order to assess that ones which provide the maximum quantity of information that can be used for forecasting. A number of statistical procedures are applied to the selected series of data in order to generate three series of data, respectively at high, medium and low frequencies. Three independent forecasting are performed on the three series, by means of different prediction algorithms. Finally, the predictions are combined in order to obtain the final prediction of the power. The results show that the proposed algorithm outperforms methods retrieved from the literature in a short term prediction.
Remarks
The method is a bit intricate, but it is not clear which kind of advantage derives from each part of the method
The short term prediction is not the main issue about forecasting of wind power. Long-term prediction is more important, because it affects the energy storage system, or the proper size of the plant. Therefore, the validation would be more significant if it concerns this aspect
Between velocity and power of wind there is a cubic relation. Please clarify if the relation considered in the method concerns different altitutes or time instants. If this is not the case, even if the relation is not linear, velocity and power bring exactly the same information
Some parts of the method are not described in depth. Citations are widely used instead, which does not favour the reading of the paper.
The algorithm used to train the recursive neural network is not indicated
The Error Back Propagation algorithm has not been used for a long time. More classical references than [38] could be used
Author Response
Dear Editors and Reviewers:
Thank you very much for sending us the Reviewers' reports on our manuscript (Sustainability-420436) entitled “Forecasting models for wind power using Extreme-point Symmetric Mode Decomposition and Artificial Neural Networks”. Particularly, we would like to thank the Reviewers for their valuable comments and criticisms.
According to the Reviewers and Editors' recommendations, we have revised carefully our manuscript with red font in Microsoft Office Word. The following is a detailed list of response to all comments and criticisms, and changes the authors have made. If any question arises, please let us know.
Point 1: The method is a bit intricate, but it is not clear which kind of advantage derives from each part of the method
Response 1: The reviewer correctly noted that the descriptions of advantages from each part of this method are unclear. Therefore, we re-describe the advantages and disadvantages of the method in the Introduction, and explain its application in wind power and wind speed prediction. The specific modifications are reflected in lines 63 to 158 of the article.
Point 2: The short term prediction is not the main issue about forecasting of wind power. Long-term prediction is more important, because it affects the energy storage system, or the proper size of the plant. Therefore, the validation would be more significant if it concerns this aspect.
Response 2: This suggestion made by the reviewer can serve as a direction for follow-up research. Although this article only covers short-term prediction of wind power, it still has important practical value for achieving high-precision prediction of wind farm power generation and safety and economic dispatch. Therefore, accurate short-term wind power prediction is also important, and it is necessary to verify it. However, the recommendations mentioned by the reviewers will be adopted in subsequent studies.
Point 3: Between velocity and power of wind there is a cubic relation. Please clarify if the relation considered in the method concerns different altitudes or time instants. If this is not the case, even if the relation is not linear, velocity and power bring exactly the same information.
Response 3: It is really true as reviewer suggested that clarify if the relation considered in the method concerns different altitudes or time instants. This point mentioned by the reviewers is our negligence. The meteorological data we get from other people is related to different times and altitudes. We have already corrected this in the article, such as lines 382 to 396.
Point 4: Some parts of the method are not described in depth. Citations are widely used instead, which does not favour the reading of the paper.
Response 4: When we saw the reviewer's suggestion, we also found that the description in the method section was not deep enough. Therefore, we rewrote the introduction to explain the advantages and disadvantages of the method and its application. The revised description is in lines 63 to 158 of the article.
Point 5: The algorithm used to train the recursive neural network is not indicated.
Response 5: We are sorry about the negligence of the algorithm used to train the recursive neural network. Therefore, we have rewritten this part and pointed out its training algorithm. For details, please refer to lines 281 to 294 of the article.
Point 6: The Error Back Propagation algorithm has not been used for a long time. More classical references than [38] could be used.
Response 6: Based on the reviewer's recommendations, we cite more classic references, see lines 229 to 230 of the article.
Reviewer 3 Report
I have reviewed the manuscript "Forecasting models for wind power using Extreme-point Symmetric Mode Decomposition and Artificial Neural Networks", Manuscript ID: sustainability-420436. In this paper, the authors propose a combined model designed in order to improve the accuracy of short-term wind power prediction, which involves the grey correlation degree analysis, ESMD (Extreme-point Symmetric Mode Decomposition), Sample entropy (SampEn) theory, and a hybrid prediction model based on three prediction algorithms. In order to evaluate the prediction performance of the proposed combined model, the authors propose a case study and, by analyzing it, they state that the prediction results show that the combined model provides better performance regarding the short-term wind power prediction compared with other models. I consider that the article will benefit if the authors take into account the following remarks and address within the manuscript the signaled issues:
1) Most of the sections of the manuscript are not according to the ones recommended by the Sustainability MDPI Journal's Template. The authors must restructure their paper according to the Template, as follows: Abstract, Keywords, 1. Introduction, 2. Materials and Methods, 3. Results, 4. Discussion, 5. Conclusions (not mandatory), 6. Patents (not mandatory), Supplementary Materials (not mandatory), Author Contributions, Funding, Acknowledgments, Conflicts of Interest, Appendices and References.
2) The equations within the manuscript should be explained, demonstrated or cited, as there are some equations that have not been introduced in the literature for the first time by the authors and that are not cited.
3) In order to validate the usefulness of their research, in the "Discussion" section (that for the time being is missing completely from the manuscript), the authors should make a comparison between their study from the manuscript and other ones that have been developed and used in the literature for this purpose. In the "Discussion" section the authors should also highlight current limitations of their study, and briefly mention some precise directions that they intend to follow in their future research work. The paper will benefit if the authors make a step further, beyond their approach and provide an insight at the end of the "Discussion" section regarding what they consider to be, based on the obtained results, the most important steps that all the involved parties should take in order to benefit from the results of the research conducted within the manuscript.
4) Lines 19-23: "Then, the wind power sub-series obtained by ESMD method is reconstructed into three characteristic component, namely PHC (high frequency component of wind power), PMC (medium frequency component of wind power) and PLC (low frequency component of wind power). Similarly, the wind speed sub-series obtained by ESMD method is reconstructed into SHC, SMC and SLC." The authors must explain in a more detailed manner the significance of the three components of the wind speed, stating clearer the meaning of SHC, SMC, SLC just like they did in the case of the wind power sub-series.
5) Lines 306-311: "To validate the prediction performance of the proposed combined prediction model, 15-min wind power and meteorological data are collected from wind farm in China, which were generated from April 24, 2016 to May 9, 2016, a total of 15 days. Each dataset is divided into training set and testing set, in which the data of first 14 days is used as training set to train the prediction model, and the data of the last day are used as testing set to estimate the prediction performance of the model." The authors must provide more details regarding the dimension of the dataset along with the ones of the training and testing subsets. How did the authors solve the problems related to missing data or abnormal values if they are to occur?
6) The authors must provide specific details regarding the development environment in which the method has been programmed. What were the hardware and software configurations of the platform used to run the experimental tests/develop the case study?
7) Are the performed experimental tests relevant for the moment when their combined approach will be put in a real production environment? Is the training data set relevant for the huge amount of data that the developed method will have to process when it is put into practice?
8) How often does the network need to be retrained/updated and how did the authors tackle the need of retraining/updating the network? How is the data encountered stored for subsequent updates of the network?
9) The authors should pay more attention to the spelling, grammar and style as several errors have occurred, for example at Line 18: "three prediction algorithm" instead of "three prediction algorithms", at Line 20: "three characteristic component" instead of "three characteristic components", at Line 139: "three prediction algorithm" instead of "three prediction algorithms", at Line 282: " Figureure 1" instead of "Figure 1", at Line 357: " Figureure 2" instead of "Figure 2".
10) Line 59:"…and Kalman Filters (KF) [11], etc ", Line 71: "…(LSSVM) [19-20], etc.", Line 80: "…low convergence rate, etc [24].", Lines 160-161: "…climate change issues [35] and seismology [36], etc.", Line 301: "…ESMD-BPNN model, etc." In a scientific paper one should avoid using run-on expressions, such as "and so forth", "and so on" or "etc.". Therefore, instead of "etc.", the sentences should mention all the elements that are relevant to the manuscript.
Author Response
Dear Editors and Reviewers:
Thank you very much for sending us the Reviewers' reports on our manuscript (Sustainability-420436) entitled “Forecasting models for wind power using Extreme-point Symmetric Mode Decomposition and Artificial Neural Networks”. Particularly, we would like to thank the Reviewers for their valuable comments and criticisms.
According to the Reviewers and Editors' recommendations, we have revised carefully our manuscript with red font in Microsoft Office Word. The following is a detailed list of response to all comments and criticisms, and changes the authors have made. If any question arises, please let us know.
Point 1: Most of the sections of the manuscript are not according to the ones recommended by the Sustainability MDPI Journal's Template. The authors must restructure their paper according to the Template, as follows: Abstract, Keywords, 1. Introduction, 2. Materials and Methods, 3. Results, 4. Discussion, 5. Conclusions (not mandatory), 6. Patents (not mandatory), Supplementary Materials (not mandatory), Author Contributions, Funding, Acknowledgments, Conflicts of Interest, Appendices and References.
Response 1: We are sorry about the format of the manuscript. After receiving the reviewer's suggestion, we revisited the template of the Sustainability MDPI Journal and modified each part of the manuscript according to the template.
Point 2: The equations within the manuscript should be explained, demonstrated or cited, as there are some equations that have not been introduced in the literature for the first time by the authors and that are not cited.
Response 2: Based on the reviewer's suggestion, we cite the relevant references; see lines 303 to 304 in the article.
Point 3: In order to validate the usefulness of their research, in the "Discussion" section (that for the time being is missing completely from the manuscript), the authors should make a comparison between their study from the manuscript and other ones that have been developed and used in the literature for this purpose. In the "Discussion" section the authors should also highlight current limitations of their study, and briefly mention some precise directions that they intend to follow in their future research work. The paper will benefit if the authors make a step further, beyond their approach and provide an insight at the end of the "Discussion" section regarding what they consider to be, based on the obtained results, the most important steps that all the involved parties should take in order to benefit from the results of the research conducted within the manuscript.
Response 3: Based on the reviewer's suggestion, we revisited the paper and placed this part of the reviewer's recommendations in the Conclusion of the article. For specific additions, see lines 588 to 608 of the article.
Point 4: Lines 19-23: "Then, the wind power sub-series obtained by ESMD method is reconstructed into three characteristic component, namely PHC (high frequency component of wind power), PMC (medium frequency component of wind power) and PLC (low frequency component of wind power). Similarly, the wind speed sub-series obtained by ESMD method is reconstructed into SHC, SMC and SLC." The authors must explain in a more detailed manner the significance of the three components of the wind speed, stating clearer the meaning of SHC, SMC, SLC just like they did in the case of the wind power sub-series.
Response 4: According to the reviewer's suggestion, we elaborated the significance of the three wind speed characteristic components and further explained the meaning of SHC, SMC, SLC. The specific changes are reflected in lines 21 to 24 of the article.
Point 5: Lines 306-311: "To validate the prediction performance of the proposed combined prediction model, 15-min wind power and meteorological data are collected from wind farm in China, which were generated from April 24, 2016 to May 9, 2016, a total of 15 days. Each dataset is divided into training set and testing set, in which the data of first 14 days is used as training set to train the prediction model, and the data of the last day are used as testing set to estimate the prediction performance of the model." The authors must provide more details regarding the dimension of the dataset along with the ones of the training and testing subsets. How did the authors solve the problems related to missing data or abnormal values if they are to occur?
Response 5: The reviewer correctly noted that the missing of dimension of the dataset along with the ones of the training and testing subsets. Therefore, we provide a more detailed description of the dimensions of the dataset along with the ones of the training and testing subsets in Appendix A. Moreover, although wind power and meteorological data for 15 days are continuous without missing data points, there are still a small number of outliers. Therefore, we use fractal interpolation to process the outliers that are rejected (lines 359 to 362). The Appendix A is shown in lines 613 to 615 of the article.
Point 6: The authors must provide specific details regarding the development environment in which the method has been programmed. What were the hardware and software configurations of the platform used to run the experimental tests/develop the case study?
Response 6: According to the reviewer's suggestion, we added a chapter, which is the parameter setting of the prediction model. In this chapter, it is elaborated the hardware and software configurations of the platform used to run the experimental tests/develop the case study. Specific modifications are shown in lines 471 to 475 of the article.
Point 7: Are the performed experimental tests relevant for the moment when their combined approach will be put in a real production environment? Is the training data set relevant for the huge amount of data that the developed method will have to process when it is put into practice?
Response 7: The wind power and meteorological data used in this paper are all from the real data of wind power plants. The experiment was also carried out to simulate the wind power prediction of real wind power plants. Therefore, the performed experimental tests are relevant for the moment when their combined approach will be put in a real production environment. And, the training data set is relevant for the huge amount of data that the developed method will have to process when it is put into practice.
Point 8: How often does the network need to be retrained/updated and how did the authors tackle the need of retraining/updating the network? How is the data encountered stored for subsequent updates of the network?
Response 8: In order to answer the reviewer's question, we list the relevant parameters of the prediction model and show them in lines 471 to 475 of the article.
Point 9: The authors should pay more attention to the spelling, grammar and style as several errors have occurred, for example at Line 18: "three prediction algorithm" instead of "three prediction algorithms", at Line 20: "three characteristic component" instead of "three characteristic components", at Line 139: "three prediction algorithm" instead of "three prediction algorithms", at Line 282: " Figureure 1" instead of "Figure 1", at Line 357: "Figureure 2" instead of "Figure 2".
Response 9: We are sorry about the negligence of the spelling, grammar and style. After seeing the reviewer's comments, we revisited our paper and scrutinized the spelling, grammar, and style.
Point 10: Line 59:"…and Kalman Filters (KF) [11], etc", Line 71: "(LSSVM) [19-20], etc.",Line 80: "…low convergence rate, etc [24].",Lines 160-161: "…climate change issues [35] and seismology [36], etc.",Line 301: "…ESMD-BPNN model, etc." In a scientific paper one should avoid using run-on expressions, such as "and so forth", "and so on" or "etc.". Therefore, instead of "etc.", the sentences should mention all the elements that are relevant to the manuscript.
Response 10: It is really true as reviewer suggested that avoid using run-on expressions, such as "and so forth", "and so on" or "etc." Therefore, we examined the article carefully and made changes to the corresponding places.
Round 2
Reviewer 1 Report
I have no more comments about the paper.
Reviewer 3 Report
I have reviewed the revised version of the manuscript "Forecasting models for wind power using Extreme-point Symmetric Mode Decomposition and Artificial Neural Networks", Manuscript ID: sustainability-420436 and I can conclude that the authors have improved the manuscript.
