Research on a Method for Identifying Key Fault Information in Substations

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

SUMMARY
The research addresses a relevant topic for digital power substations. The authors proposed a method of combining scalable time window adjustment with PCA. The real-life data were used to train and test the proposed approach. However, several issues weaken the paper's quality. It needs major revisions before it can be considered for publication.

COMMENTS

1. The Related Works section is not critical. Please, provide a clear gap analysis.
2. Parameter selection (like threshold φ for alarm density) is not justified or tuned experimentally.
3. Please, point out, how exactly is Δalarm/Δt calculated when multiple alarms overlap?
4. The PCA process is standard. Please, provide explanation why PCA was selected. There are alternative dimensionality reduction techniques like t-SNE, UMAP, or Autoencoders that could better capture non-linearities.
5. The dataset size is low: 164 groups (120 training, 44 testing). It is not enough for a statistically meaningful machine learning application.
6. The dataset and the source should be described in details.
7. How much better is this method than existing ones? Please, use additional quantitative metrics (precision, recall, F1-score, confusion matrix, etc.).
8. Conclusions summarize steps without providing new scientific insights.

Comments on the Quality of English Language

1. Some sentences are long and verbose, making the manuscript hard to follow.
2. There are awkward phrases: "Time correction for group i will be obtained circularly" (should be "iteratively" not "circularly").

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

SUMMARY

This study proposes a fault-critical information identification method that integrates a scalable time window with Principal Component Analysis (PCA). The study demonstrates the operation of the method for data from a real substation (500 kV) and shows its effectiveness for diagnostics. In general, the level of research and paper is high, but there are a number of comments.

COMMENTS

1. In the introduction, the section on the main contributions could be improved by stating more specifically the main distinguishing features of the method.

2. It is recommended to add more Related Works by analyzing more contemporary sources.

3. How the predefined threshold φ is selected? (lines 164-165)

4. Why preset threshold t is 90%? (line 252) It can be analyzed how the accuracy of the result will change when this parameter is changed.

5. Has the stability of the method been analyzed?

6. A quantitative assessment of the quality of the method can be added (specify accuracy). Comparison with other methods can also be added.

7. What are the limitations of the method?

8. In Table 3 it is not clear why some parameters are used and some are not. It might be worth adding a bit more information about this, rather than just “YES” and “-”.

9. Figure 1 is difficult to read because of the quality and vertical orientation of the text.

10. It is recommended to present Figure 6 in a table rather than screenshots.

11. 44 test samples is not enough. It is recommended to increase, for example by adding syntactic data.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

In this paper, the research on a method for identifying key fault information in substations is presented. The topic has already been widely studied in the literature. The novelty of the proposed approach is not clearly stated, but it should be clarified. Following are some considerations:

1. What is the criteria used to define the fixed window length (10 seconds) and the alarm rate threshold?

2. The limitation of the paper is that there isn't any numerical comparison with another approach for a particular scenario. This makes it difficult to understand the real advantages of the proposed approach.

3. The results shown in Fig. 7 should be better discussed. How do you interpret the high contribution of the first principal component (over 40%) from an engineering perspective? Does this suggest that a single parameter predominantly influences system behavior during D6 faults? If so, which physical parameter is most strongly represented by this component? Have you evaluated the presence of multicollinearity or redundancy among the original parameters that might explain the rapid convergence of the cumulative variance (93.64% with only four components)? 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript describes a novel technique for fault detection in direct current electrical power stations. The Authors suggest using a data-driven approach with special methods for data extraction and processing. The paper is well-written, makes sense and can be interesting for practicing electrical plant designers and scholars in fault detection, power engineering, and control. The study fits well into the scope of the MDPI Computation journal.
However, I found several shortcomings and wanna make some remarks to help the Authors revise the manuscript. My most serious concern is that real fault-control systems are all about reliability and safety, and usually incorporate a hybrid approach combining mathematical, data-driven and expert levels to ensure the detection. This issue is not discussed in the manuscript. Please, find my more detailed comments as follows.
1. The introduction is devoted to the problems of DC power stations (HVDC, maybe power plant is the more correct term here). Can your method be applied to HVAC equipment as well? If yes, why do we need this lengthy introduction about HVDC? If not, the title should be somewhat like "A method for identifying key fault information in DC power substations"
2. The Authors mention computational complexity in the Abstract, but it is not measured during this study. This sounds a bit speculative.

3. Some alternative fault diagnosis techniques based on new generation of neural networks may be included into the literature review, e.g. reservoir artificial and spiking neural networks. 

4. Figure 3 represents a "scalable time window" concept. Meanwhile, what about hard \firm real-time systems? How the floating time part will work there?

5. I recommend replacing Figure 6 with corresponding listings, or moving it to appendix. The information contaned in this "figure" is obviously a text, not graphics.

6. I recommend expanding the list of references by mentioning some state-of-the-art fault detection techniques, e.g. abovementioned reservoir artificial and spiking neural networks.

7. In my opinion, the comparison with existing solutions is far from being complete.

8. The subsection Mathematical Model-based Methods lacks the mention of some modern approaches for obtaining mathematical models of power system components, e.g. algebraic methods for the reconstruction of partially observed nonlinear systems using differential and integral embedding and identifying empirical equations of electric circuit from data. These methods can potentially resolve a main drawback of Mathematical Model - Based Approach to fault detection by allowing to obtain precise and adequate plant models.

Nevertheless, my overall opinion is good and I can recommend this paper to be accepted for publication in MDPI Computation after necessary revisions.

Comments on the Quality of English Language

The manuscript requires some further proofreading, but is generally well-written.

I recommend making some variables written in italic font, e.g. at line 199: "...group i will be obtained". It will increase readability.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The paper has been improved

Reviewer 3 Report

Comments and Suggestions for Authors

All the improvements have been addressed.

Reviewer 4 Report

Comments and Suggestions for Authors

Thank you very much for taking into account my comments and revising your manuscript. I have no doubts that your study will be a great contribution to the MDPI Computation journal. I wish the Authors good luck in their future studies.

Sincerely,

Reviewer