A Hybrid Deep Learning Framework for Wind Speed Prediction with Snake Optimizer and Feature Explainability
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsMy comments are attached
Comments for author File: Comments.pdf
Some Suggestions for Improvements:
- may be more 50 % need a Technical Consistency
- Removed redundant phrases such as " according to where citations suffice"
- Impact Enhancement: Consider adding one sentence about practical applications (e.g., "Theses improvements directly support more efficient energy market participation...")
Author Response
Response to Reviewer 1
We thank the reviewer for the thoughtful and constructive comments. We have carefully considered each point and revised the manuscript accordingly. Below are detailed responses to each comment:
Comment 1:
“On page 1, in the Introduction section, the paragraph structure is rather disorganized...”
Response:
Thank you for the suggestion. We have thoroughly reorganized the Introduction section. The revised structure begins with the global context of energy demand and sustainability, followed by challenges in wind energy integration, a concise review of forecasting methods, and concludes with a clear statement of novelty and contributions. Furthermore, the UK and US statistics were merged into a unified paragraph titled “Global Wind Energy Development Background” for coherence and flow.
Comment 2:
“Lines 36–37 lack clarity regarding the projection of a 2.5-fold increase in electricity generation...”
Response:
We appreciate this observation. The sentence has been revised to clarify both the geographical scope and baseline year. It now reads:
“Global projections by the International Energy Agency (IEA) indicate that by 2050, electricity generation needs will increase 2.5-fold relative to 2020 levels to meet the rising demand for goods and services, despite anticipated improvements in energy efficiency.”
Comment 3:
“U.S. wind power statistics are cited without sources or years...”
Response:
All statistical claims in the Introduction have been revised to include proper citations and publication years. For instance, the statement about the UK’s renewable energy target now references the UK Government’s official “Net Zero Strategy (2021).” Similarly, U.S. data is now cited from the U.S. Energy Information Administration (EIA, 2022).
Comment 4:
“The narrative on forecasting models lacks coherence...”
Response:
We have restructured the discussion of forecasting models in the Introduction and Literature Review to follow a thematic classification:
-
Traditional Time-Series Models (e.g., ARMA, ARIMA)
-
Machine Learning Models (e.g., SVR, kNN, RF)
-
Deep Learning Models (e.g., LSTM, GRU, BiGRU)
-
Hybrid and Optimization-Based Models (e.g., LSTM-GRU with SOA)
Each category now includes a brief description of strengths and limitations, followed by the motivation for our proposed hybrid and explainable deep learning approach.
Comment 5:
“Excessive citation stacking and lack of critical evaluation...”
Response:
We have reduced citation stacking and inserted transitions to connect concepts logically. A summary table (now Table 1 in the revised manuscript) has also been added, comparing previous studies by forecasting method, dataset, and key findings. This improves clarity and critical engagement with existing literature.
Comment 6:
“Doctoral dissertation mention lacks context...”
Response:
We agree with the reviewer. The reference to the doctoral dissertation has been relocated to the Author Contributions section and revised to state that the present study extends earlier work on renewable energy forecasting frameworks developed during the author’s PhD research.
Comment 7:
“Literature Review lacks thematic structure...”
Response:
The Literature Review has been fully restructured to align with methodological themes as recommended:
-
Traditional ML Models
-
Deep Learning Models
-
Optimization Algorithms
-
Hybrid/Ensemble Models
Within each section, comparisons have been drawn to highlight the progression in performance and accuracy. This enhances the clarity and logical progression of the review.
Comment 8:
“The review is descriptive; a comparative table is recommended...”
Response:
We have added a new comparative table (Table 2) summarizing previous forecasting models by dataset, performance metrics (e.g., RMSE, MAE), and applications. This enhances the analytical depth and helps readers interpret the field’s progression.
Comment 9:
“Citation placeholder ‘[?]’ appears on lines 219–221...”
Response:
This error has been corrected. The placeholder “[?]” has been replaced with the proper citation number [43] in both the main text and the reference list.
Comment 10:
“Wind power forecasting datasets are not clearly described...”
Response:
We thank the reviewer for this point. The Proposed Methodology section now includes detailed descriptions of the three wind datasets from the NREL Wind Integration National Dataset (WIND) Toolkit. For each dataset, we have specified:
-
Source and collection period
-
Variables used (e.g., wind speed, direction, pressure)
-
Geographic coverage and climatic diversity
-
Preprocessing steps including normalization and missing value handling
Comment 11:
“Validation strategy and performance metrics lack detail...”
Response:
We have expanded the validation section to describe dataset selection rationale, evaluation metrics (MSE, RMSE, MAE, R²), and their mathematical formulations. Furthermore, a cross-validation technique has been considered to enhance robustness, and model generalization has been discussed in terms of performance across three cities with distinct climates.
Reviewer 2 Report
Comments and Suggestions for AuthorsReview Comments
1.On page 1, in the Introduction section, the paragraph structure is rather disorganized. I recommend reorganizing the content in a more logical order: begin with the global energy context, then discuss challenges in wind energy, followed by a review of forecasting methods, and conclude with the novelty and contributions of this study. Additionally, the UK and US wind power data can be merged into a single paragraph under a unified theme such as “Global Wind Energy Development Background.”
2.On page 1, in the Introduction section, lines 36–37, the sentence “Projections indicate that by 2050, the growing demand for goods and services will require a 2.5-fold increase in electricity generation” lacks clarity. It is unclear whether this projection is global or region-specific. Moreover, the reference period for the “2.5-fold increase” is not specified. I suggest revising this sentence to clarify both the geographical scope and baseline year.
3.On page 2, in the Introduction section, U.S. wind power statistics are cited without clear sources or corresponding years. For example, in lines 43–44, the sentence “By 2050, the goal is for renewable energy to account for three-quarters of the UK’s total electricity generation” should be supported by a specific data source. I recommend adding accurate references for all statistical statements.
4.In the Introduction section, there are repeated mentions of models such as ARMA, ARMA-ANN, ANN, and SVM, but the narrative lacks coherence and does not explain the unique strengths or evolution of these models. I suggest restructuring the literature according to methodological categories—e.g., traditional time series models, machine learning, deep learning, and hybrid models—while summarizing their advantages, limitations, and the motivation for adopting a hybrid deep learning + optimization + interpretability approach in the current study.
5. In the Introduction section, paragraphs 3–5 include excessive citation stacking without smooth transitions or critical evaluation. For example, the sentence “The electricity market, together with the gas market and different real-world applications...” includes too many bracketed references with inconsistent formatting. I recommend unifying the citation format and adding transitional commentary. Furthermore, a summary table of prior studies—highlighting their methods, datasets, and key findings—would enhance clarity.
6.On page 3, in the final paragraph of the Introduction, the authors suddenly mention the doctoral dissertation, which is not clearly connected to the main study. If the dissertation is directly relevant, please briefly explain its relation to the present work. Otherwise, consider removing this part or moving it to the author contribution statement.
7.In the Literature Review section, although a wide range of methods is discussed, the structure lacks thematic clarity. Many models are listed sequentially without comparing their strengths or applications. I recommend grouping the literature by methodology—e.g., traditional machine learning (SVR, RF, kNN), deep learning (LSTM, Bi-GRU), optimization (PSO, GA, ALO), and hybrid/ensemble methods—and summarizing key insights and gaps in each group.
8. In the Literature Review section, the review is largely descriptive and lacks in-depth comparative analysis. I suggest adding a table comparing different forecasting models in terms of datasets used, performance metrics, and application contexts to enhance analytical depth and reader comprehension.
9. In the Literature Review section, lines 219–221, the sentence “Research in [?] proposes a cooperative wind power forecasting method...” contains an incomplete citation. Please replace “[?]” with the correct reference number and verify that all references are properly formatted and listed in the bibliography.
10.In the Proposed Methodology section, the NSRDB dataset is introduced, which is primarily associated with solar energy. However, details of the dataset(s) used for wind power forecasting are missing. I recommend specifying the data source, time span, features, and preprocessing steps for wind datasets. If three datasets were used for validation, please explain how they capture different climatic or geographic conditions, or justify the dataset selection.
11. In the Proposed Methodology section, the description of the validation process is vague. It is unclear how the datasets were selected or whether they reflect different operating conditions. Additionally, evaluation metrics such as RMSE or MAPE are not explained in detail. I suggest elaborating on the validation strategy and performance metrics, and considering a more rigorous cross-validation approach to enhance the robustness of the results.
Comments for author File: Comments.pdf
Author Response
Response to Reviewer 2
We sincerely thank the reviewer for their insightful and technically detailed feedback. All suggestions were carefully considered, and the manuscript was revised accordingly to improve clarity, structure, and technical rigor. Please find our point-by-point responses below.
1. Areas Needing Improvements
i. Technical Clarity & Specificity
-
Dataset Details:
Response: We have added detailed descriptions of each dataset in the “Proposed Methodology” section. This includes time resolution (5-minute intervals), turbine specifications (hub height of 100m, rated capacity 1.5–3.0 MW), and feature set (wind speed, direction, temperature, pressure, air density). -
Preprocessing:
Response: Preprocessing steps are now described in more detail, including missing data imputation using linear interpolation and forward fill, outlier detection using Z-score analysis (|z| > 3), and normalization via MinMaxScaler. -
SOA Implementation:
Response: We now explicitly define the hyperparameters optimized using SOA:-
LSTM units: {32, 64, 128}
-
Dense units: {16, 32, 64}
-
Learning rate: {0.001, 0.0005, 0.0001}
-
Dropout rate: {0.1, 0.2, 0.3}
-
Batch size: {16, 32, 64}
These were evaluated over 10 iterations with five snakes (population size). This clarification is included in Section 3.
-
-
Physical Models:
Response: A sentence has been added to explain the limitations of physical models:
“These models, although informative, are computationally intensive due to the need to solve partial differential equations and often underperform at the local level because of the coarse spatial resolution of NWP outputs.” -
Problem Statement Transition (Lines 43–55):
Response: We have added the suggested bridging sentence:
“To address these integration challenges, researchers have developed forecasting models aimed at mitigating intermittency risks through improved predictive accuracy.” -
Physical vs. Data-Driven Models:
Response: This section was reformatted and clarified with improved structure. We now distinguish both model types using clear subheadings and elaboration. -
"Functional Relationships" Clarification:
Response: This phrase has been clarified as:
“These models infer functional relationships—such as linear regression, decision trees, or neural networks—between historical weather patterns and corresponding power outputs.” -
ARMA Limitations Quantified:
Response: We have added the following clarification:
“Forecasting errors increase significantly over longer horizons; for example, RMSE can increase by more than 30% for forecasts beyond 6 hours using ARMA models alone.”
ii. Comparative Baseline
-
Benchmark Models Added:
Response: The experimental evaluation section included the persistence and ARIMA models as baseline comparisons. The manuscript now reports that the proposed LSTM+SOA model achieved RMSE reductions of 52.5% over ARIMA and 38.3% over standard LSTM, averaged across all datasets.
iii. XAI Limitations
-
LIME Limitations Acknowledged:
Response: We now include a sentence recognizing LIME’s local interpretability limitation:
“While LIME provides local interpretability of individual predictions, it does not capture global model behavior. Future work may explore SHAP or Grad-CAM for more comprehensive explanations.”
2. Suggestions for Revision
i. Contribution Statement
-
Response: A new paragraph summarizing key contributions has been added at the end of the Introduction:
“The contributions of this study are threefold: (1) development of a hybrid forecasting framework combining LSTM, GRU, and BiLSTM-BiGRU architectures; (2) integration of the Snake Optimizer Algorithm (SOA) for hyperparameter tuning to enhance long-term forecasting accuracy; and (3) application of LIME for model explainability, ensuring transparency in energy prediction models.”
ii. Proofreading for Conciseness
-
Response: Redundant phrases have been removed throughout the manuscript. For example, the sentence “The research emphasis... intensified throughout the previous twenty years” has been revised to:
“Research on wind power forecasting has grown significantly since 2000 [6][7].”
We are grateful for your thoughtful review, which has significantly improved the manuscript. Please let us know if further refinements are needed.
Reviewer 3 Report
Comments and Suggestions for Authorsnice proposal
it is obvious and known that numerical simulation is a method more cheaper than experiment in reality, and it is used for forecasting and available comparisons.
to stress this idea and indicate the value of +/- of the possible errors, according to your proposal
Figures are not clear, try to review them all
make a clear and correct connection between abstract and conclusions
perhaps a nomenclature/for abbreviations/ list is suitable
to introduce speficic nomencalature in the article for all abbreviations, making thus more easy to understand the content/follow the ideas/results
introduce a special section (in abstract and introduction) to raise the importance and novelty braught througgh the article, and review the *hot* aspects. Explain in detail why the topic i relevant and what gap it covers
comment at least with one sntence more the figures and tables. As they are now, it is quite minimum offered.
Comments on the Quality of English Languageunderstandable, perhaps a nomenclature list is suitable
Author Response
Response to Reviewer 3
We want to thank the reviewer for the encouraging comments and valuable suggestions. We have carefully revised the manuscript based on the feedback provided. Below are our detailed responses to each point.
Comment 1:
“Nice proposal. It is obvious and known that numerical simulation is a cheaper method than experiments in reality, and it is used for forecasting and for available comparisons. To stress this idea, indicate the value of +/- of the possible errors according to your proposal.”
Response:
Thank you for the positive evaluation. We have incorporated a statement in both the Introduction and Methodology sections to emphasize the cost-effectiveness and practical relevance of numerical simulations in forecasting tasks, especially in the renewable energy sector. We also added a short discussion on forecast uncertainty ranges and the interpretability of error metrics (MSE, RMSE, MAE) to highlight the model’s performance under realistic scenarios.
Comment 2:
“Figures are not clear, try to review them all.”
Response:
We appreciate this observation. All figures have been reviewed for clarity and resolution. We have replaced low-resolution graphics with high-quality versions and ensured that axis labels, legends, and captions are clearly legible. Moreover, each figure has been rechecked for consistency with the text and better visual presentation.
Comment 3:
“Make a clear and correct connection between the abstract and the conclusions.”
Response:
We have revised both the abstract and conclusion sections to improve alignment. The key findings stated in the conclusions are now clearly anticipated in the abstract, including the model’s superior performance, the role of the Snake Optimizer Algorithm (SOA), and the contribution of LIME to model interpretability. This ensures a coherent narrative from beginning to end.
Comment 4:
“Perhaps a nomenclature/for abbreviations/ list is suitable.”
Response:
A comprehensive Nomenclature section has been added after the conclusion, listing all abbreviations and their definitions (e.g., LSTM, GRU, SOA, MAE, RMSE, etc.). This addition enhances the article's accessibility and readability for a broad audience.
Comment 5:
“Introduce a special section (in abstract and introduction) to raise the importance and novelty brought through the article, and review the hot aspects. Explain in detail why the topic is relevant and what gap it covers.”
Response:
We have revised both the abstract and the final paragraph of the introduction to state the novelty and importance of the work clearly. These include:
-
The integration of explainable AI (LIME) in wind speed forecasting
-
The application of SOA for hyperparameter optimization
-
We combine multiple DL architectures to capture complex temporal dependencies. Additionally, we describe the gap in the literature related to explainable and adaptable forecasting frameworks for long-term energy planning under climate variability.
Comment 6:
“Comment at least with one sentence more the figures and tables. As they are now, it is quite minimum offered.”
Response:
We have revised the main text to include more descriptive and interpretive commentary on all figures and tables. For each, we now provide at least one additional sentence explaining its significance, trends, or comparison outcomes to improve reader understanding and engagement.
Comment on English Language:
“Understandable, perhaps a nomenclature list is suitable.”
Response:
Thank you. We have added the suggested nomenclature list and performed a final language review for clarity, grammar, and technical precision. We believe the revised manuscript now meets high standards of academic writing.
Please let us know if further improvements are needed. Once again, we appreciate the reviewer’s helpful insights.