A Novel Hybrid Deep Learning Model for Day-Ahead Wind Power Interval Forecasting
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis manuscript presents a hybrid model based on deep learning for predicting wind power intervals one day in advance. The manuscript contributes to the field of research, but some points need to be revised for improvement.
- Explanation of the research design
a) The research questions and hypotheses are not explicitly stated, which could improve the clarity of the study. - Presentation of results
a) Although the results show clear improvements, it should be explained in more detail how each component of the model contributes to error reduction in practice.
b) There is no discussion of possible scenarios where the model could fail (e.g. extreme weather events not reflected in historical data). - References
a) In some places, general statements are made about the benefits of deep learning without citing specific studies to support them. For example, some sentences are presented without clear support:
"Deep learning methods have become the dominant approach to wind power forecasting."
Location: Page 2, line 56
"Combining techniques such as data decomposition, feature selection, and deep learning has proven to be an effective strategy for improving the accuracy of wind power forecasts."
Location: Page 3, line 93
"The proposed model achieves higher accuracy compared to the interval forecasts generated by the NPKDE method."
Location: Page 19, line 26
- Relationship between Results and Conclusions
a) It is mentioned that the proposed model is "more efficient and accurate", but the computational cost and implementation requirements are not analyzed in depth.
b) No scenarios are presented where the model may have limitations (e.g. extreme weather conditions or insufficient data).
Comments for author File: Comments.pdf
Author Response
1. Explanation of the research design
a) The research questions and hypotheses are not explicitly stated, which could improve the clarity of the study.
Response-We sincerely thank the reviewer for his/ her insightful suggestion. The research questions and hypotheses have been explicitly stated in Section “1. Introduction”, the specific content related to this is as follows:
“
Wind power forecasting has made significant advancements through extensive research. Nevertheless, challenges remain in improving model accuracy and addressing the uncertainty in forecasting results. To tackle these issues, this paper proposes a day-ahead wind power interval forecasting model that integrates the Gaussian mixture model (GMM), feature selection (FS), empirical wavelet transform (EWT), convolutional neural networks (CNN), bidirectional gated recurrent unit (BiGRU), and multi-head self-attention mechanism (MHSAM). The GMM is employed to cluster numerical weather prediction (NWP) and wind power data with similar daily variation patterns. Feature selection is then applied to identify the most influential NWP features affecting wind power output. EWT decomposes the data into frequency components with time information, isolating high-frequency elements that represent randomness and volatility. The CNN-BiGRU-MHSAM model is constructed by combining the strengths of CNN, BiGRU, and MHSAM to capture both spatial and temporal correlations, thereby enhancing forecasting accuracy.
2. Presentation of results
a) Although the results show clear improvements, it should be explained in more detail how each component of the model contributes to error reduction in practice.
Response-We would like to thank the reviewer for pointing out this. The role of each component of the model in reducing wind power forecasting errors has been elaborated in detail and added to the manuscript, highlighted in blue font. The specific content is as follows:
“
As shown in Table 3, the forecasting accuracy of the GMM-CNN-BiGRU-MHSAM model is higher than that of the CNN-BiGRU-MHSAM model. It can be inferred that clustering daily data with similar distribution characteristics using the GMM model can effectively improve the accuracy of wind power forecast. The forecasting accuracy of the GMM-FS-CNN-BiGRU-MHSAM model is higher than that of the GMM-CNN-BiGRU-MHSAM model, indicating that selecting appropriate meteorological features for wind power forecast is also an effective way to enhance forecasting accuracy. The forecasting accuracy of the GMM-EWT-CNN-BiGRU-MHSAM model is higher than that of the GMM-CNN-BiGRU-MHSAM model, proving that using EWT to decompose numerical weather prediction data and wind power data into frequency data containing temporal information, and extracting high-frequency components that represent randomness and volatility in the data, is also an important strategy for improving wind power forecasting.
”
b) There is no discussion of possible scenarios where the model could fail (e.g. extreme weather events not reflected in historical data).
Response- As the respected reviewer has suggested, we have discussed the possible scenarios where the model could fail. The relevant content has been added to the manuscript and highlighted in blue font. The specific content is as follows:
“
Based on the case study analysis above, it can be seen that the forecasting model proposed in the paper has good forecasting performance in time series data forecast. However, we must point out that when these time series data are significantly missing, the forecasting accuracy of the proposed model will be affected, and its forecasting accuracy may even be lower than that of conventional machine learning models. Additionally, since this paper uses the EWT to remove high-frequency components from numerical weather prediction data and wind power data, the prediction accuracy of the model is bound to decrease in the event of extreme weather conditions.
”
3. References
a) In some places, general statements are made about the benefits of deep learning without citing specific studies to support them. For example, some sentences are presented without clear support:
"Deep learning methods have become the dominant approach to wind power forecasting."
Location: Page 2, line 56
Response- We would like to thank the learned reviewer for his/her constructive suggestion. To support the statement that "Deep learning methods have become the dominant approach to wind power forecasting," we cited the review paper "Offshore Wind Power Forecasting Based on WPD and Optimised Deep Learning Methods" in our manuscript, the literature reference number is [15].
"Combining techniques such as data decomposition, feature selection, and deep learning has proven to be an effective strategy for improving the accuracy of wind power forecasts."
Location: Page 3, line 93
Response- We would like to thank the learned reviewer for his/her constructive suggestion. To support the statement that"Combining techniques such as data decomposition, feature selection, and deep learning has proven to be an effective strategy for improving the accuracy of wind power forecasts." We cited the review paper " Improving Short-term Offshore Wind Speed Forecast Accuracy Using a VMD-PE-FCGRU Hybrid Model " in our manuscript, the literature reference number is [33].
"The proposed model achieves higher accuracy compared to the interval forecasts generated by the NPKDE method."
Location: Page 19, line 26
Response- As correctly pointed out by the wise reviewer, we have compared to the interval forecasts generated by the NPKDE method in Section 7.3.
4. Relationship between Results and Conclusions
a) It is mentioned that the proposed model is "more efficient and accurate", but the computational cost and implementation requirements are not analyzed in depth.
Response- As the respected reviewer has suggested, the computational cost and implementation requirements have been analyzed in depth. The relevant content has been added to the manuscript and highlighted in blue font. The specific content is as follows:
“
To demonstrate the effectiveness of the GMM-FS-EWT-CNN-BiGRU-MHSAM forecasting model, this section compares its performance with that of the GMM-EWT-CNN-BiGRU-MHSAM, GMM-FS-CNN-BiGRU-MHSAM, GMM-CNN-BiGRU-MHSAM, and CNN-BiGRU-MHSAM models. The computational resources are as follows: the computer processor has a clock speed of 1.8 GHz, with 2 GB of RAM, and the software used for computation is MATLAB 2021. Since GMM, FS, and EWT all belong to machine learning methods, the computation time for these machine learning methods can complete relevant tasks within a few seconds. Therefore, the training and forecasting time of the GMM-FS-EWT-CNN-BiGRU-MHSAM model is primarily determined by the training and forecasting time of the CNN-BiGRU-MHSAM component. It can be inferred that the training and forecasting times for these five forecasting models are almost identical.
”
b) No scenarios are presented where the model may have limitations (e.g. extreme weather conditions or insufficient data).
Response- As the respected reviewer has suggested, we have discussed the possible scenarios where the model could fail. The relevant content has been added to the manuscript and highlighted in blue font. The specific content is as follows:
“
Based on the case study analysis above, it can be seen that the forecasting model proposed in the paper has good forecasting performance in time series data forecast. However, we must point out that when these time series data are significantly missing, the forecasting accuracy of the proposed model will be affected, and its forecasting accuracy may even be lower than that of conventional machine learning models. Additionally, since this paper uses the EWT to remove high-frequency components from numerical weather prediction data and wind power data, the prediction accuracy of the model is bound to decrease in the event of extreme weather conditions.
”
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper proposes a hybrid deep learning model for day-ahead wind power interval forecasting. The paper is well-written, however, I have the following comments:
- The paper includes too many acronyms and abbreviations. I would recommend adding a nomenclature to ease the readability of the paper. For example, it was hard to read the following: “To demonstrate the effectiveness of the GMM-FS-EWT-CNN-BiGRU-MHSAM forecasting model, this section compares its performance with that of the GMM-EWT-CNN-BiGRU-MHSAM, GMM-FS-CNN-BiGRU-MHSAM, GMM-CNN-BiGRU-MHSAM, and CNN-BiGRU-MHSAM models”.
- For Figure 3(b), the y-axis label and scale are not shown for all results. Also more clarifications and demonstrations on the obtained results are recommended to be added.
- In figure 10, it is recommended to represent the x-axis in terms of time quantities, not points, for better readability or to show us how much time is equivalent to 96 points. Is it 24 hours?
- Also, in Figure 10, using the yellow and blue colors to present different results makes it hard to read. Please use different colors.
- Moreover, the figure numbering has to be revised.
- One more thing: I think it is useful to provide the readers with a paragraph about the computational time taken for each method to analyze and develop the result.
Author Response
The paper includes too many acronyms and abbreviations. I would recommend adding a nomenclature to ease the readability of the paper. For example, it was hard to read the following: “To demonstrate the effectiveness of the GMM-FS-EWT-CNN-BiGRU-MHSAM forecasting model, this section compares its performance with that of the GMM-EWT-CNN-BiGRU-MHSAM, GMM-FS-CNN-BiGRU-MHSAM, GMM-CNN-BiGRU-MHSAM, and CNN-BiGRU-MHSAM models”.
Response- We would like to thank the learned reviewer for his/her constructive suggestion. A nomenclature has been added into manuscript and highlighted in blue font, the relevant content is as follows:
“
Nomenclature
GMM |
Gaussian mixture model |
EWT |
empirical wavelet transform |
CNN |
convolutional neural network |
BiGRU |
bidirectional gated recurrent unit |
FS |
feature selection |
MHSAM |
multi-head self-attention mechanism |
GMM-FS-EWT-CNN-BiGRU-MHSAM |
A hybrid model with GMM, FS, EWT, CNN, BiGRU and MHSAM |
GMM-EWT-CNN-BiGRU-MHSAM |
A hybrid model with GMM, EWT, CNN, BiGRU and MHSAM |
GMM-FS-CNN-BiGRU-MHSAM |
A hybrid model with GMM, FS, CNN, BiGRU and MHSAM |
GMM-CNN-BiGRU-MHSAM |
A hybrid model with GMM, CNN, BiGRU and MHSAM |
CNN-BiGRU-MHSAM |
A hybrid model with CNN, BiGRU and MHSAM |
NWP |
Numerical weather prediction |
VMD |
Variational mode decomposition |
GRU |
Gated recurrent unit |
WPD |
Wavelet packet de-composition |
WPF |
wind power forecasting/ forecast |
”
For Figure 3(b), the y-axis label and scale are not shown for all results. Also more clarifications and demonstrations on the obtained results are recommended to be added.
Response-As correctly pointed out by the wise reviewer, Figure 3(b) has been revised to show the y-axis label and scale, and more clarifications and demonstrations on the obtained results have been added in manuscript.
The revised content is as follows:
“
Among them, Subsequence A1 is the high-frequency component, Subsequence A2 is the medium-frequency component, and Subsequence A3 is the low-frequency component. As shown in Figure 3(b), the high-frequency components primarily describe the stochasticity of NWP and wind power data, while the low-frequency components mainly describe the fluctuations in NWP and wind power data.
(a) Wind power reconstruction error (b) Three level subsequences decomposed by EWT
Figure 3. EWT decomposition results.
”
The revised content is as follows:
“
Among them, Subsequence A1 is the high-frequency component, Subsequence A2 is the medium-frequency component, and Subsequence A3 is the low-frequency component. As shown in Figure 3(b), the high-frequency components primarily describe the stochasticity of NWP and wind power data, while the low-frequency components mainly describe the fluctuations in NWP and wind power data.
(a) Wind power reconstruction error (b) Three level subsequences decomposed by EWT
Figure 3. EWT decomposition results.
”
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper proposes a day-ahead wind power interval forecasting method based on Gaussian Mixture Model (GMM), feature selection, Empirical Wavelet Transform (EWT), Convolutional Neural Network (CNN), and Bidirectional Gated Recurrent Unit (BiGRU), and verifies the validity and feasibility of the proposed method through case studies. The calculation results demonstrate that this day-ahead wind power interval forecasting approach exhibits certain novelty and practical applicability. However, the paper has several areas that require revision, including the following points:
(1) There are formatting issues present in parts of the paper, such as on pages 4 and 5, among others.
(2) Eq. (16) should be aligned to the right.
(3) The English writing quality of the paper needs further improvement.
(4) While the combination of various models is innovative, there is limited discussion on why certain models or techniques were chosen over others, and further elaboration on this rationale is necessary.
(5) Further validation on diverse real-world datasets (e.g., from different geographic regions and under varying seasonal/weather conditions) is recommended to comprehensively assess the model's generalizability.
(6) The conclusion section of the paper needs to further reflect the work that has been done.
Comments on the Quality of English LanguageThe English could be improved to more clearly express the research.
Author Response
(1) There are formatting issues present in parts of the paper, such as on pages 4 and 5, among others.
Response- We are sincerely thankful to the referee for the constructive suggestion. The formatting issues present in pages 4 and 5 has been revised.
(2) Eq. (16) should be aligned to the right.
Response- The authors are extremely thankful to the respected referee. Eq. (16) has been aligned to the right.
(3) The English writing quality of the paper needs further improvement.
Response- We are sincerely thankful to the referee for the constructive suggestion. The English writing quality of the paper has been further improved.
(4) While the combination of various models is innovative, there is limited discussion on why certain models or techniques were chosen over others, and further elaboration on this rationale is necessary.
Response- Thank you for pointing this out. The analysis of the advantages of these model combinations has been added to the manuscript, and highlighted in blue font. The relevant discussion content is as follows:
“
As shown in Table 3, the forecasting accuracy of the GMM-CNN-BiGRU-MHSAM model is higher than that of the CNN-BiGRU-MHSAM model. It can be inferred that clustering daily data with similar distribution characteristics using the GMM model can effectively improve the accuracy of wind power forecast. The forecasting accuracy of the GMM-FS-CNN-BiGRU-MHSAM model is higher than that of the GMM-CNN-BiGRU-MHSAM model, indicating that selecting appropriate meteorological features for wind power forecast is also an effective way to enhance forecasting accuracy. The forecasting accuracy of the GMM-EWT-CNN-BiGRU-MHSAM model is higher than that of the GMM-CNN-BiGRU-MHSAM model, proving that using EWT to decompose numerical weather prediction data and wind power data into frequency data containing temporal information, and extracting high-frequency components that represent randomness and volatility in the data, is also an important strategy for improving wind power forecasting.
”
(5) Further validation on diverse real-world datasets (e.g., from different geographic regions and under varying seasonal/weather conditions) is recommended to comprehensively assess the model's generalizability.
Response- We are sincerely thankful to the referee for the constructive suggestion. The data used for the example analysis comes from a real wind farm, and the dataset is a complete year of operating data, which can fully support the model validation task. The introduction related to data is as follows.
“
The NWP and wind farm power data used in this study were obtained from a wind farm located in northwest China, with an installed capacity of 200 MW. The wind turbines have a hub height of 70 m, a rotor diameter of 120 m, and an individual turbine capacity of 1.5 MW. Data were collected over the course of a year, from January 1 to December 31, 2019, at a resolution of one data point every 15 minutes, yielding 96 points per day and a total of 35,040 points for the year. Each data point contains 15 features, including temperature, humidity, air pressure, and wind speed and direction at 10 m, 30 m, 50 m, and 70 m above ground level.
”
(6) The conclusion section of the paper needs to further reflect the work that has been done.
Response- We would like to thank the learned reviewer for his/her constructive suggestion. The conclusion section has been revised, and the revised content is as follows:
“
(1) Performing feature selection on NWP data and then clustering NWP data and wind power data using the GMM model can effectively improve the forecasting accuracy of day-ahead wind power.
(2) The use of empirical wavelet transform to decompose NWP and wind power data into frequency components with time information allows for the extraction of high-frequency data, which enhances the forecasting accuracy.
(3) A CNN is used to extract spatial correla-tions and meteorological features, while the BiGRU model captures temporal dependencies within the data sequence. A MHSAM is incorporated to assign greater weight to the most influential elements.
(4) The GMM-FS-EWT-CNN-BiGRU-MHSAM model reduces interval width while maintaining the coverage rate of the forecasting intervals, and the examples demonstrate that the method proposed in this paper has good forecasting performance.
”
Author Response File: Author Response.pdf