Combined Optimization Prediction Model of Regional Wind Power Based on Convolution Neural Network and Similar Days

Round 1

Reviewer 1 Report

The paper contains appropriate literature review, description of proposed methods, case study and discussion.
The authors could strengthen the paper by adding details about model combination (ensemble learning) methodology. Especially, it should be clearly stated why this particular combination type was choosen.
It is necessary because proposed models like similar days and CNN can be combined in a huge number of ways. For example, similar days output could be embedded as additional inputs of neural network, thus this data will be combined in a more complex non-linear way (this can lead to significant accuracy improvements). It should be cleared why authors choose simple weighting.
Also authors can improve paper quality by adding more details about their CNN hyperparameters tuning process.
I have no hesitation in recommending that it be accepted for publication after the details written above have been added.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper describes an optimization wind power prediction model which is applied to a case study. The content and the topic are interesting; however, the paper lacks of information for the input data and the analysis of the results is limited. There are specific points that should be taken into consideration:

1. At line 32, in what way the complexity of the model is an advantage?
2. KNN, NWP, ARMA, LSTM are some of the abbreviations that full form should also be given in the text.
3. Lines 81,82 are irrelevant with the content.
4. At line 105, a reference should be provided where Google proposal is mentioned.
5. Please provide the dates referred in Figure 2.
6. 4.1 and 4.2 subsections have the same title.
7. Please further explain how PE and QR evaluates the results. Furthermore, each equation should have a different numbering (equation (6)).
8. At the begging of the case analysis the authors should provide more details about the data they used. The only information they provide is the installed wind power. The presentation of data should be extensive.
9. I suggest presenting in Table 3 the errors in percentage rather than actual values.
10. The authors should provide more comments and discussion on their results in Table 5.
11. Section 6 is entitled Case Analysis. Is this what the authors intended to present?
12. At line 304 the authors mention a reference level. Can you please provide more details about what are you referring to?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this manuscript, the authors used convolutional neural network model, similar days model and combination model to predict the future short-term wind power curve. The experimental results showed that the combination model exhibited better prediction performance. This paper may be accepted in Machines with minor revision. The following points should be clarified:

1) In Section 5.1, please provide the Statistical error of CNN prediction (similar to Table 3).

2) Please provide both the training error and testing error for CNN model, similar days model, and combination model, respectively. Is there an overfitting problem?

3) There are some errors with Figure 4. In general, the original dataset are split into train dataset, validation dataset and test dataset. Sometimes, the original dataset are only split into train dataset and test dataset. The definitions are listed as follows: “Training Dataset: The sample of data used to fit the model. Validation Dataset: The sample of data used to provide an unbiased evaluation of a model fit on the training dataset while tuning model hyperparameters. Test Dataset: The sample of data used to provide an unbiased evaluation of a final model fit on the training dataset.” Test dataset does not belong to train dataset. Please modify this figure.

4) The name of Section 3.1 and Section 3.2 are same. Please check this typo.

5) In Line 231- 233, the authors mentioned that “ Ten years of data from 2010 to 2019 are used in similar day model for long-term law prediction, and the data of the last two years are used in CNN model for short-term law prediction, so as to achieve the prediction goal of the next four hours.” Different train datasets were used to develop the CNN model and similar day model, how can they compare the prediction accuracy between these two models in Section 5. Please explain this?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The response to the reviewers' comments was very short and in many cases, the authors chose to answer to comments rather than modifying the text and embed the answers to the manuscript. Their revised manuscript still misses information and elements that would improve the presentation to the final reader and would make it more comprehensible and interesting. According to my opinion, the authors failed to sufficiently respond to the reviewers' comments. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf