Intrusion Detection Method Based on CNN–GRU–FL in a Smart Grid Environment
Round 1
Reviewer 1 Report
The paper presents a distributed intrusion detection method for smart grid environments where business units are decentralized and independent. The method involves the use of a CNN-GRU-FL model with an attention mechanism and a trust-based node selection mechanism to improve convergence ability. The paper content sounds well. However, I have some comments.
1. It is important to include abbreviations in the abstract so that readers can understand the terms without having to refer to the details of the paper.
2. The abstract needs to be rewritten: a proposed rewrite could be like: “The aim of this paper is to address the current situation in which business units in smart grids (SG) environments are decentralized and independent, and there is a conflict between the need for data privacy protection and network security monitoring. To address this issue, we propose a distributed intrusion detection method based on Convolutional Neural Network-Gated Recurrent Unit-Federated Learning (CNN-GRU-FL). We design an intrusion detection model and local training process based on Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) and enhance the feature description ability by introducing an attention mechanism. We also propose a new parameter aggregation mechanism to improve the model quality when dealing with differences in data quality and volume. Additionally, we design a trust-based node selection mechanism to improve the convergence ability of federated learning (FL). Through experiments, we demonstrate that the proposed method can effectively build a global intrusion detection model among multiple independent entities and that the improved mechanism improves the performance and efficiency of parameter aggregation when there are differences in data quality.”
3. Some of the paragraphs in the introduction section are not of uniform length and may not be coherent. It is recommended to carefully rewrite the introduction in order to make the paragraphs more uniform in length and improve coherence.
4. It is not proper to mention references as "Ref. [11]". It is sufficient to simply include the number "[11]" without the use of "Ref."
5. After the first definition, it is necessary to use abbreviations rather than the full term. For example, after mentioning "federated learning" on line 267, it is appropriate to use the abbreviation "FL" rather than writing out the full term again.
6. Figure 6 needs to be replotted as it is not clear. This also applies to all of the other figures in the paper.
7. The authors did not mention any information about the validation of the proposed trained model. It would be helpful to include details on how the model was validated and the results of the validation process.
8. The formatting of the equations should be reconsidered as they are large and the clarification sentences following them are not clear enough. It may be helpful to adjust the formatting and add more detailed explanations to make the equations easier to understand.
9. It would be helpful to provide more information on the mechanism for implementing the proposed intrusion detection method for practical use. This could include details on the cost of implementation and the communication protocol that could be used. Providing this information would help to better understand the practicality and feasibility of the proposed method.
10. To make the novelty of the paper more clear, it is recommended to rewrite the last paragraph of the introduction in bullet point format. Some potential bullet points could include:
o The proposed method combines the advantages of RNN and CNN through algorithm improvement in order to address the specific situation of smart grids.
o Local detection methods based on CNN and GRU are deployed in multiple independent branch nodes and use an attention mechanism to extract important data.
o Federated learning is introduced to aggregate and optimize the parameters globally, resulting in a unified and efficient intrusion detection method.
o A node selection mechanism is designed to improve the convergence ability of federated learning in real environments.
o A new parameter aggregation mechanism is designed to improve the training effect of the intrusion detection model under federated learning, while also allowing for the efficient training of the model without direct aggregation of original data.
The overall presentation and the idea of the paper are good, and I suggest accepting the paper for publication subject to major English and language corrections. I would also thank the authors for their effort and hope to see their modifications very soon.
Best Regards
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors should discuss in detail the type of data and the amount of data to be processed and protected.
How accurate is the proposed method compared to other existing methods in the literature?
How fast is this method.
It lacks some concrete examples for example in the case of a smart grid or a microgrid.
In equation 3 what is the min and max value of offset?
Part 4.2 must have the same font and size.
Node selection is based on what criteria?What does TG>Teta mean in Figure 5?
The quality of Figure 6 is very poor.
The references are poor and this part must be improved.
The authors can use the references below to improve their article.
- - Performance evaluation of an inverter-based microgrid under cyberattacks, IEEE-SOSE2020.
- - Machine Learning Based Intrusion Detection System for Real-Time Smart Grid Security, 2021 13th IEEE PES.
- - Fault Diagnosis of Smart Grids Based on Deep Learning Approach, IEEE-WAC2021
- Intrusion Detection Method Based on SMOTE Transformation for Smart Grid Cybersecurity, 2022 3rd International Conference on Smart Grid and Renewable Energy (SGRE)
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper entitled “Smart Grid Intrusion Detection Method Based on CNN-GRU-FL” could be of interest concerning the journal. However, I have some comments for improvement.
1. The title needs revision.
2. Abstract section should be concise and present problem statements and rationale ought.
3. Typically, the introduction comprises four parts motivation and background, related work, contribute, and organization. This manuscript lacks the typical structure
4. Related work is very rudimentary and need to highlight research gaps and include the following papers to strengthen it.
1. https://doi.org/10.1016/j.apenergy.2020.114915
2. 10.1109/ACCESS.2020.2989316
3. https://doi.org/10.1016/j.apenergy.2021.117178
4. https://doi.org/10.3390/en13092244
5. It is unclear which is the novelty of the present work, which means which aspect of Intrusion Detection it is addressing and what open issue they are trying to solve.
6. A review of state of the art is completely missing. Authors shall present a review of previous work on the topic. I suggest adding a Related Works section where the authors state clearly which is the novelty wrt to previous works and which issues are trying to solve.
Check the following papers.
1. https://doi.org/10.1007/978-3-030-22354-0_4
2. https://doi.org/10.1007/978-3-319-93554-6_25
7. Some Smart Grid Intrusion Detection models in the following literature, which you take as reference or as examples.
8. The authors shall present briefly the same concepts in the introduction which is now too generic.
9. Current limitations, future scope, and scalability issues are missing in conclusion.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors did not answer all questions correctly. Please answer all the questions and improving the article and the references . The size of the formulas is very large. Table 2 does not represent much in terms of data type. Figure 2 must be redrawn, the writings must not go out of frame.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 3
Reviewer 2 Report
The authors have completed their article.
Author Response
请参阅附件。
Author Response File: 
Author Response.docx
