All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The authors have made revisions following the comments from the reviewers. In order to avoid further delay, the paper can be accepted.
[# PeerJ Staff Note - this decision was reviewed and approved by Monika Mortimer, a PeerJ Section Editor covering this Section #]
The reviewer raised major concerns. As a result, an essential revision is needed.
No comment.
No comment.
No comment.
If possible, could the authors please update more basic concepts and fundamentals of Short-Term Wind Speed Forecasting (STWSF), Empirical Mode Decomposition (EMD), Feature Selection (FS), and Support Vector Regression (SVR) based on the research topic in this paper?
In this paper, the descriptions of starting layer, convolutional layer, hidden layer, and pooling layer should be modified. How many convolutional layers and hidden layers have been determined in this paper, and what is the functionality of these layers in the experimentations and proposed methodologies? Meanwhile, if possible, could the authors please make an explanation for choosing the number of these layers mentioned above and update some descriptions and statements on the functionality of these aforementioned layers?
Specifically, it is worthy to point out, the various numbers of convolutional layers, hidden layers, and pooling layers might have either positive or negative impacts on the performances and characteristics of Short-Term Wind Speed Forecasting (STWSF) -- Empirical Mode Decomposition (EMD) -- Feature Selection (FS) -- Support Vector Regression (SVR), namely STWSF-EMD-FS-SVR, for more complicated industrial systems to some extent. Based on this reason, could the authors please provide an evidence to make some explanations of the availability, feasibility, and capability of the proposed method STWSF-EMD-FS-SVR based on the research topic in this paper?
The detailed information of training data and testing data should also be modified and respectively demonstrated the percentage of training data and testing data of the original or/and experimental dataset regarding case studies or/and experimentations based on the research topic in this paper for STWSF-EMD-FS-SVR.
Which are the most advantages of the proposed method in comparison with other existing algorithms of Empirical Mode Decomposition (EMD) in the areas of Short-Term Wind Speed Forecasting (STWSF), Feature Selection (FS), and Support Vector Regression (SVR) based on data-driven approach? A data-driven method of fault diagnosis has been demonstrated in a recent survey paper (https://doi.org/10.3390/pr9020300). Could the authors please explain which are the differences among Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), and Ensemble Empirical Mode Decomposition (EEMD) techniques? Meanwhile, evidence of using the proposed EMD framework rather than VMD or EEMD methods should also be supported. Alternatively, could the authors please provide either some TABLES or update Simulation or/and Experimental Results to make an explanation of the advantages of the proposed methodology EMD, as well as update some statistical criteria based on the research topic in this paper?
The proposed algorithm STWSF-EMD-FS-SVR should be demonstrated as the form of Algorithm Environment, which can be efficaciously followed by readers. Additionally, the flowchart of the proposed method mentioned in this paper should also be updated and included the determined dataset collections, pre-processing of the dataset, post-processing of the dataset, feature extractions, feature selections, and the performances and characteristics of the proposed algorithm STWSF-EMD-FS-SVR. A data-driven and supervised machine learning-based fault detection and fault classification has been addressed in a journal paper (https://doi.org/10.3390/pr8091066), which includes respectively the determined dataset collections, pre-processing of the dataset, post-processing of the dataset, feature extractions, and feature selections.
Alternatively, both different topologies of experimental samples and the various numbers of selected significant features might also have either positive or negative impacts on the performances and characteristics of STWSF-EMD-FS-SVR for more complicated industrial systems to some extent. Based on this reason, could the authors please support an evidence to make some explanations of the availability, feasibility, and capability of the proposed methodology STWSF-EMD-FS-SVR based on the research topic in this paper?
If possible, could the authors please have in comparison between with and without Additive White Gaussian Noise (AWGN) signals/environments (including the different values of signal-to-noise ratio [SNR]) for the system mentioned in this paper? As well as, the authors should contrast the simulation results between these scenarios mentioned above. Alternatively, could the authors please provide either some Tables or updated simulation results to make an explanation of the advantages of the proposed methodology of STWSF-EMD-FS-SVR in the fields of data-driven short-term wind speed forecasting and update some statistical criteria based on the research topic?
Two reviewers have consistent recommendations. The current version cannot be accepted, and a major revision is suggested.
[# PeerJ Staff Note: It is PeerJ policy that additional references suggested during the peer-review process should only be included if the authors are in agreement that they are relevant and useful #]
[# PeerJ Staff Note: Please ensure that all review comments are addressed in a rebuttal letter and any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate. It is a common mistake to address reviewer questions in the rebuttal letter but not in the revised manuscript. If a reviewer raised a question then your readers will probably have the same question so you should ensure that the manuscript can stand alone without the rebuttal letter. Directions on how to prepare a rebuttal letter can be found at: https://peerj.com/benefits/academic-rebuttal-letters/ #]
Figures presented in the manuscript should be improved:
1) From Fig.1, the reader might think that from IMF(1) (high-frequency) to IMF(n-1), the proposed hybrid approach utilizes FS+SVR combination for predicting the IMF component because of the 3 dots placed at the right of the pipeline processing IMF1. However, from the codes submitted, if the reviewer correctly understands, it seems that FS+SVR is applied only to the very first IMF, while LassoCV is used for all the remaining lower frequency components, including the residue. Is it possible to clarify this issue? If I correctly understood the proposed method, it is better to modify Fig. 1 to avoid this confusion.
2) Fig. 2-6. It is better to add the labels and scales to the X-axis. From .csv files, it is understandable that those axis represent time expressed in hours, but it can be good to put the labels and scales for X-axis.
The representation of some tables can be improved:
3) In the opinion of this reviewer, it can be better to present tables 4-5 (the improvement comparison) in form of tables 8 and 9.
The proposed methodology explanation and description should be elaborated more by the authors of the manuscript. Specifically:
1) The challenges of predicting high-frequency IMF and low-frequency IMF with residue are not well stated in the text of the manuscript. Why do the authors consider different models for predicting those components and thus, construct a hybrid approach? Why not use the same FS+SVR or LassoCV for all the extracted intrinsic modes? These questions should be better explained in the text of manuscript.
2) Feature Selection subsection. From this subsection, the definition of raw features is not very clear. Are these features comprise the statistical feature parameters of lags created from IMF components? Or the lags (i.e., time-series) are used as features? Please, clarify this in the text of the manuscript.
3) It is recommended to add more details in the text (formulations and explanations) regarding the step 3 from Fig.1, where the predictions of each IMF are combined in the ensemble and final prediction is made.
4) From experimental results the predictive performance of the proposed model looks good. But are there any limitations of the proposed model? Please, state them in the discussion section if they are any.
No comment.
The general comments can be split into the major and minor comments which are represented as below.
Major comments:
1) How exactly the author combined EMD with the referenced models for the second comparison in the Discussion section? Do those techniques utilize EMD in the same manner as the proposed method? In the opinion of this reviewer, this explanation should be elaborated more.
2) Fig.1. At the bottom of this figure, Step 3 shows the process of merging the prediction results obtained for IMF components into the ensemble. Is it possible to create an additional figure which shows how the proposed method predicts individual IMF components, for example, for a single 3-step prediction, and how these predictions then form the ensemble? In the opinion of this reviewer, this additional figure can be useful for the potential readers of the manuscript since it provides a better understanding of the intermediate results.
Minor comments:
1) In the Introduction section and along with the text of the manuscript, the author mentions short-, medium-, and long-term wind speed forecasting scenarios. In the Results section, the experimental results for the proposed and referenced models are obtained while testing 1-step, 2-step, and 3-step prediction strategies. As a reader, I am curious, whether these strategies used in the experiment cover those 3 scenarios (short-, medium-, long-term predictions) mentioned in Introduction or, for example, they all considered short-term or medium-term prediction scenarios? Is it possible to briefly clarify this moment?
2) In the Introduction section of the manuscript there are a lot of acronyms that are not defined in the text of the manuscript. Some of them are well-known, such as ANN, SVM, and ARIMA, but some of them are not. It is a good practice to define the acronyms before their use in the text.
3) Does the proposed method support online learning, i.e., the update of prediction models when the newly unseen instances of data arrive?
4) There are some minor typos in the manuscript, such as at lines 282-283. It is recommended to double-check the manuscript and fix them.
Dear Author, from my own observations,
If possible, could you please update more background and principles of Empirical Mode Decomposition (EMD), Feature Selection (FS), and Support Vector Regression (SVR) in the specific areas of wind speed forecasting based on your research topic.
The details of training data and testing data should be listed and respectively demonstrated the percentage of training data and testing data of the original dataset in terms of case studies or/and experimentations.
Which are the most advantages of your proposed method in comparison with other existing algorithms of EMD and SVR in the areas of Short-Term Forecasting (STF) based on wind speed data? If possible, could you please have in comparison between with and without noisy signals/environments for STF system and contrast the simulation results between the two scenarios mentioned above. Alternatively, could you please provide either some Tables or update simulation results to make an explanation the advantages of your proposed methodology of EMD and SVR in the fields of wind speed STF and update some statistical criterion based on your research topic.
The proposed algorithm should be demonstrated as the form of Algorithm Environment, which can be efficaciously followed by readers. Additionally, the flowchart of proposed method mentioned in this paper should also be updated and included the dataset collections, pre-processing of dataset, post-processing of dataset, feature extractions, feature selections, and the performances of forecasting, respectively.
Alternatively, for one thing, different types of experimental samples and the number of feature selection might have either positive or negative impacts for the performances of wind speed forecasting for more complicated industrial systems to some extent. Based on this reason, could you please provide an evidence to make explanations the availability and feasibility of the proposed methodology based on wind speed data for STF system?
For another thing, the characteristics of both healthy and faulty simulated wind speed data based on time-domain should be modified, and correspondingly the energy distribution of IMFs should also be analysed and discussed.
Recently, a new survey paper of wind turbine systems concentrates on fault diagnosis, prognosis and resilient control, which included model-based, signal-based, and knowledge-based (data-driven) techniques to demonstrate the characteristics of fault detection, classification, and isolation in various faulty scenarios. Additionally, the novel EMD or/and SVR techniques for STF have also been proposed in some papers.
If possible, could you please update these References for having in comparison with other algorithms in the specific field of your research topic in terms of wind speed data for STF system, as well as making any comments? The References are shown in the following below.
Reference 1: A. T. Eseye, et. al., “Short-Term Forecasting of Heat Demand of Buildings for Efficient and Optimal Energy Management Based on Integrated Machine Learning Models,” in IEEE Transactions on Industrial Informatics, vol. 16, no. 12, pp. 7743–7755, Dec. 2020.
Reference 2: “An Overview on Fault Diagnosis, Prognosis and Resilient Control for Wind Turbine Systems,” Processes, vol. 9, no. 2, p. 300, Feb. 2021.
Reference 3: M. Sajjad, et. al., “A Novel CNN-GRU-Based Hybrid Approach for Short-Term Residential Load Forecasting,” in IEEE Access, vol. 8, pp. 143759–143768, 2020.
Reference 4: Y. Fu, et. al., “Actuator and Sensor Fault Classification for Wind Turbine Systems Based on Fast Fourier Transform and Uncorrelated Multi-Linear Principal Component Analysis Techniques,” Processes, vol. 8, no. 9, p. 1066, Sep. 2020.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.