A Power Transformer Fault Diagnosis Method Based on Improved Sand Cat Swarm Optimization Algorithm and Bidirectional Gated Recurrent Unit

Round 1

Reviewer 1 Report

After reviewing this paper, we found some concerns that need to be revised before the acceptance decision. These comments are given as follows. And the authors are encouraged to modify any part that needs to be improved.

The English form should be improved, typos are present, and the syntax needs to be revised to improve legibility.

I suggest the authors improve the readability of this paper to facilitate it for the readers and get more attention in the future. This can be started from the abstract section. This section can be improved by focusing on the main idea with the given contributions. As well,  the obtained results and the experiments overall.

Your ideas in the introduction section need to be more comprehensive.

As the work proceeds from a mathematical perspective, derivations are not necessary. The meaningful contribution to its research findings has to be stated well.

The research findings and contributions need to be stated clearly. As well as, the obtained results in this paper

Before the end of section one, the last paragraph should contain at least the following: The proposed methods, the main differences between the proposed algorithm and the others How the contributions were made. The problem that has been solved in this research. The datasets that have been used in the research experiments. The overall results that you have been got.

The related works section is concise, and no benefits from it. I suggest increasing the number of studies and adding a new discussion.

Show the advantage, disadvantages, and weaknesses of the studied works. The authors should discuss the literature review more profoundly and clearly.

The design of the proposed method is too general without enough links to the problem. The introduction to the method should directly be oriented on the specific problem.

For example, the following papers might be cited in your work.

Multi-fault diagnosis in three-phase induction motors using data optimization and machine learning techniques

Modified Sand Cat Swarm Optimization Algorithm for Solving Constrained Engineering Optimization Problems

Automatic fault diagnostic system for induction motors under transient regime optimized with expert systems

The discussion of the results needs to include the strengths and weaknesses of the proposed algorithm.

 The authors need to summarize all their findings at the end of the experiment and indicate how their findings can help future research.

The discussion of the results needs to include the strengths and weaknesses of the proposed algorithm.

Furthermore, the evaluation metrics should be briefly described in the experimental section. Moreover, add further details on how simulations were conducted. Similarly, system and resource characteristics could be added to Tables for clarity.

The conclusion section also needs significant revisions. It should briefly describe the findings of the study and some more directions for further research. The authors should describe academic implications, major findings, shortcomings, and directions for future research in the conclusion section.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

1. Line 34 - the figure such as 2, 4 and 6 should be presented as subscript.

2. Literatures related to fault diagnosis of power transformers are insufficient. More papers should be reviewed. Authors should also highlight the differences between their proposed work and the published works.

3. Problem statements that motivate the current study are not properly explained. Please rephrase for better clarity. 

4. Please summarize the technical contributions of current work.

5. All mathematical symbol and variables need to be presented in standardized way in italic form. 

6. One of my concerns with this paper is that the mathematical equations are not properly presented and distorted. The front sizes of these equations are either too large or too small. Please rectify this issue.

7. There are two H_n are observed in Eq. (1). Why is it so? Please explain.

8. It is still not clear how the feature selection process can be achieved through Eqs. (1) to (3). Authors need to provide further explaination.

9. FIgure 1 - Fronts inside are too small and difficult to read.

10. What are the differences between Pos_bc in Eq. (7) and Pos_b in Eqs. (8) and (9)? They seem to share the same definitions. Same concerns raised for Pos(t+1) in Eqs. (7)-(9) and X(t+1) in Eq. (10). Are these two vectors refer to same meaning? 

11. Section 3 is not presented comprehensively.  It is not clear which search operator governs the exploitation and exploration. Authors need to provide further elaboration in this section for better clarity. 

12. Why logistic chaos mapping is chosen for population initialization. What is the strength of logistic map as compared to other chaotic map?

13. The rationale of using Eq. (12) to adjust the value of rG need to be explained further.

14. Lines 153 to 154 - It is quite misleading to say higher inertia weight can lead to better search performance. Overemphasis on the exploration can slow down the convergence speed of algorithm. 

15. Eq. (13) is not adequately presented. What is the meaning of "i" inside this equation?

16. Eq. (15) seems to be incomplete. Is there any connection between the first and second row of equations?

17. Figure 15 - Please enlarge because it is difficult to read the front.

18 - Figure 16 - Typo of "Falut" is observed.

19 - Please include future works that can be derived from current study. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

General information

The article describes an algorithm for classifying defects in the paper-oil insulation system based on the assessment of gases dissolved in the oil analysis methd (DGA). The DGA method is still considered to be the basic diagnostic method that allows to make an initial assessment of the developing insulation problem without having to take the transformer out of service. So far, many algorithms have been developed to classify a defect based on the results obtained, so it is increasingly difficult to improve something in this area. The authors of the article showed that it is possible with the improved sand cat swarm optimization algorithm and bidirectional gated recurrent unit, which should be considered a considerable achievement.

In general, the subject matter certainly coincides with the subject of the journal and has an appropriate scientific level. The authors chose the right analysis tools and performed the research correctly. However, a one important issue was missing, which is discussed below.

In terms of editing and language the article is of a sufficiently standard, with small editorial errors listed below. The article also uses graphics of appropriate quality.

Editorial errors

1.  In line 34 numbers should be written as subscripts.

2.     In lines 80 and 82 remove extra spaces before dots.

3.   In line 272 should be: “Figure 14.”.

4.    In line 279 the sentence should start with a capital letter “There…”.

5.    In line 333 and in Table 3 and more you use term Partial Discharge, but is should be used the wording Partial Discharges (plural), like in the standard: IEC 60599 and similar.

 

Issues for discussion

In the article, the authors presented the algorithm in great detail and in a sufficient way. Unfortunately, the database that was used for both training and testing was vaguely described. The database on the basis of which the analysis was carried out is sufficiently numerous and this is not a problem. However, we do not know what the base refers to, what units (power transformers) and where it comes from, etc. Table 3 presents a summary, but it does not specify what. Is it some average value from the adopted case database? How the database verifies what type of damage we are dealing with, etc. In the aforementioned IEC 60599 standard, for the interpretation of the results of the DGA method, there are, for example, ranges of typical values (not single numbers, just ranges of possible parameters). I am asking for a broader comment and supplementing this part of the article, thanks to which it will gain much its value.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

The authors of this paper studied the fault diagnosis problem for power transformers considering five fault types including low energy discharge, high energy discharge, partial discharge, med-to low temperature overheating, and high temperature overheating. There are some issues that need to be addressed in the revisions.

1. You wrote that there are six fault types. However, this is problematic because 'normal' is not really a fault so it cannot be considered as a fault type.

2. The paper title is confusing and seems erroneous in grammar. Shall it not be "A Power Transformer Fault Diagnosis Method Based on Improved Sand Cat Swarm Optimization Algorithm and Bidirectional Gated Recurrent Unit"?

3. There are many abbreviations in the abstract unexplained at first use. This will make readers hard to follow and the paper has poor readability.

4. For fault diagnosis research, it is necessary to formally clarify the scientific question first before diving into solutions in section 2. By saying that, there is a lack of formal statement and formulation of scientific problems from the practical problems. 

5. Why did you adopt L-sIomap and ISCSO to solve the considered problem in the first place? What are the core motivations when selecting the methods over the many existing methods?

6. There are many other recent data-driven fault diagnosis methods to further discuss, eg, Optimized design of parity relation based residual generator for fault detection data-driven approaches.

7. In the experiment parts, in addition to accuracy, can you add other indicators, eg, others in the confusion matrix, to evaluate the performance more comprehensively?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I don't see the authors revise the paper accordingly.

Also, a list of responses should be prepared point by point

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors have addressed most of my comments. However, authors are advised to be more specific when indicating the changes made for each comment. It will make reviewer easier to track the changes made.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Thanks to the authors for the discussion and answers to all my doubts. I accept the article in its current, corrected form.

Author Response

Revision complete.

Reviewer 4 Report

• The paper has been well revised.

Author Response

Revision complete.

Round 3

Reviewer 1 Report

ok