Fault Location Method of Distribution Network Based on VGAE-GraphSAGE

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

First of all, I should mention that an effective study has been conducted using GNN, one of the current and popular topics of recent years. The study has been examined in detail and comprehensively. In this article, a distribution network fault detection method based on graph neural network is proposed. The distribution network is treated as non-Euclidean graph data; then Variational Graph Autoencoders (VGAE) are used to extract the underlying information of the nodes and improve the overall denoising performance of the fault detection method. Then, the GraphSAGE model is used to collect the neighbor information of the nodes, fully consider the influence of the surrounding lines on the target lines, and improve the output of the model to find the distribution network line where the fault occurs. Experimental sample analysis based on OpenDSS simulation software proves that the proposed method has high accuracy and anti-interference, and the accuracy reaches 97.81%. Moreover, the positioning result is still good in the new smart distribution network scenario with distributed power access, and the accuracy is 95.07% in the hybrid power generation scenario. The graph approach has been an effective solution in solving and analyzing complex problems. The article seems technically appropriate and successful. However, I would like to make the following contributions for the academic writing and organization of the article. In addition, 1-2 new strong references can be added.

1- The Introduction section should be revised by adding subsections 1.1. Problem statement 1.2 Main contributions and 1.3 Paper organization.

2- Section 2 should also be enriched by developing the literature. The following sources are directly related studies. Current articles from 2023 and 2024 can be added and presented in a table.

 

Das, B., Kutsal, M., Das, R., 2022. Effective prediction of drug – target interaction on HIV using deep graph neural networks. Chemometrics and Intelligent Laboratory Systems 230, 104676. https://doi.org/10.1016/j.chemolab.2022.104676

 

Das, R., Soylu, M., 2023. A key review on graph data science: The power of graphs in scientific studies. Chemometrics and Intelligent Laboratory Systems 240. https://doi.org/10.1016/j.chemolab.2023.104896

 

3- A title that includes a contribution to the approach developed in the article should be selected. The current title is weak in terms of article content.

 

4- A block diagram that includes all the processes of the proposed approach should be drawn. This can also be thought of as a graphical abstract.

 

5- Subsections come immediately after some section titles in the article. This is not correct. After a title, a paragraph of text containing that topic must be written.

As in Section 6 and 6.1. This should be corrected.

 

6- With the gradual expansion of the scale of the distribution grid system, the topology is more complex. Access to distributed power increases the randomness and volatility of the distribution grid system, causing the complex and variable operation of the distribution grid. Therefore, it is necessary to prevent faults as much as possible during the normal operation of the distribution grid. Additional comments on these parts should be given in the contributions section.

 

7- At the same time, after the fault occurs, the fault location can be found accurately and quickly, and timely guidance can be provided to eliminate the fault and reduce the fault loss. This paper conducts a comparative analysis among traditional distribution grid fault location methods. It is stated that these methods do not take into account the influence of the relationship between the distribution equipment in the distribution grid on the fault location accuracy. !!! If most of them are analyzed based on the assumption of the research, they should be cited with up-to-date and concrete references ???

Comments on the Quality of English Language

It is fine and it can be readable and followable. 

Author Response

Comments 1: 1- The Introduction section should be revised by adding subsections 1.1. Problem statement 1.2 Main contributions and 1.3 Paper organization.

Response 1: The Introduction section of the paper has been modified according to the above format. We agree with this comment, this structure makes the text look clearer and easier to understand.

 

Comments 2: 2- Section 2 should also be enriched by developing the literature. The following sources are directly related studies. Current articles from 2023 and 2024 can be added and presented in a table.

Response 2: Thank you for mentioning important references directly related to the research content of this paper. We will cite the above two literatures with reference 16 and reference 17 for the convenience of readers. Since the relevant content of the references has been introduced in the paper, listing the table separately will be slightly repetitive in the content, so we did not list it separately.

 

Comments 3: 3- A title that includes a contribution to the approach developed in the article should be selected. The current title is weak in terms of article content.

Response 3: Thank you for pointing this out. The VGAE-GraphSAGE model proposed in this article is aimed at the problem of power grid fault location. The title already contains the background of the problem and the proposed model, which well reflects the core content of the article. Therefore, we think this title is appropriate.

 

Comments 4: 4- A block diagram that includes all the processes of the proposed approach should be drawn. This can also be thought of as a graphical abstract.

Response 4: Maybe we were not accurate in our presentation. In fact, Figure 1 represents all the processes of the proposed approach, and we have amended the title to: Overall flowchart of VGAE-GraphSAGE distribution network fault location model.

 

Comments 5: 5- Subsections come immediately after some section titles in the article. This is not correct. After a title, a paragraph of text containing that topic must be written.

Response 5: We agree with this comment. Therefore, we added paragraph of text containing each topic when necessary, as in Section 6 and 6.1, Section 7 and 7.1.

 

Comments 6: 6- With the gradual expansion of the scale of the distribution grid system, the topology is more complex. Access to distributed power increases the randomness and volatility of the distribution grid system, causing the complex and variable operation of the distribution grid. Therefore, it is necessary to prevent faults as much as possible during the normal operation of the distribution grid. Additional comments on these parts should be given in the contributions section.

Response 6: Thank you for pointing this out. We have added the relevant content in the "Main contributions" section.

 

Comments 7: 7- At the same time, after the fault occurs, the fault location can be found accurately and quickly, and timely guidance can be provided to eliminate the fault and reduce the fault loss. This paper conducts a comparative analysis among traditional distribution grid fault location methods. It is stated that these methods do not take into account the influence of the relationship between the distribution equipment in the distribution grid on the fault location accuracy. !!! If most of them are analyzed based on the assumption of the research, they should be cited with up-to-date and concrete references ???

Response 7: Thank you for pointing this out. We have added content to the comparative analysis of traditional switching network fault location methods in Section 1.1, pointing out that these methods do not consider the impact of the relationship between switching devices in the switching network on fault location accuracy.

Reviewer 2 Report

Comments and Suggestions for Authors

In this study, a distribution network fault location method based on graph neural network is proposed which provides reduced dimension of input sample features and computational complexity, and good noise reduction. The organization and overall presentation is good. I have one minor and one bigger comment:

 

1-     What is FTU, PT, CFMI and etc? All abbreviations should be given with its original form when they are used first.

2-     It would be good if there is a comparison table provided comparing the proposed method and state-of-the-art methods from the literature.

Comments on the Quality of English Language

It is good.

Author Response

Comments 1: 1- What is FTU, PT, CFMI and etc? All abbreviations should be given with its original form when they are used first.

Response 1: Thank you for pointing this out. We have modified the content and added its original form.

 

Comments 2: 2- It would be good if there is a comparison table provided comparing the proposed method and state-of-the-art methods from the literature.

Response 2: Thank you for pointing this out. We have already provided specific supplements at the end of the “1.1 Problem statement” section.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper proposes a distribution network fault location method based on graph neural network where the distribution network is treated as non-Euclidean graph data, a Variational Graph Auto-Encoders(VGAE) used to mine information of nodes and a GraphSAGE model is finally used for fault identification. The manuscript is well written with respect to written language and scientific content. The following suggestions can improve the quality of the paper.

 

1.      What information is contained in the attribute feature X of the node in line 190?

2.      What information is contained in the low-dimensional abstract feature Z of each node in line 193. It will be beneficial if possible to provide the information in terms of electrical parameters (like adjacency matrix is given for  structure information )

3.      What are the more useful deep abstract features obtained in the output of the encoder in line 194

4.      Please explain what is meant by coding result Z of the distribution network in line 197. Also, Z was used before to represent low-dimensional abstract feature which is confusing.

5.      It will be helpful to explain what is meant by mutual influence of distribution network equipment in line 209 to have more clarity especially since distribution system are unbalanced with untransposed lines that can result in mutual impedances

6.      It will be beneficial if the four central characteristics of the node is explained briefly in line 231

7.      Please explain Fig 5, the voltage diagram diagram of IEEE123 node network. What is suggested here. IEEE 123 bus has ZIP loads .What is load model 1,2 and 3?

8.      Please explain Fig 6, the power distribution diagram of IEEE123 node network. What does 100% load level and 40% load level mean?

Author Response

Comments 1: 1- What information is contained in the attribute feature X of the node in line 190?

Response 1: Thank you for pointing this out. For a detailed explanation of the attribute feature X of the node, we have provided a detailed explanation on lines 229-239.

 

Comments 2: 2- What information is contained in the low-dimensional abstract feature Z of each node in line 193. It will be beneficial if possible to provide the information in terms of electrical parameters (like adjacency matrix is given for structure information )

Response 2: Thank you for pointing this out. For a detailed explanation of the low-dimensional abstract feature Z of each node, we have provided a detailed explanation on lines 245-246.

 

Comments 3: 3- What are the more useful deep abstract features obtained in the output of the encoder in line 194

Response 3: Thank you for pointing this out. For a detailed explanation of the output of the encoder we have provided a detailed explanation on lines 237-240.

 

Comments 4: 4- Please explain what is meant by coding result Z of the distribution network in line 197. Also, Z was used before to represent low-dimensional abstract feature which is confusing.

Response 4: Thank you for pointing this out. In the previous version of the article, it described “ the low-dimensional abstract feature Z of each node in the graph is obtained by sampling”. Since the sampling result Z is updated iteratively during the coding process, the output is also represented by Z. To avoid confusion, we have deleted the abstract representation Z of low-dimensional abstract features.

 

Comments 5: 5- It will be helpful to explain what is meant by mutual influence of distribution network equipment in line 209 to have more clarity especially since distribution system are unbalanced with untransposed lines that can result in mutual impedances

Response 5: Thank you for pointing this out. The current model primarily considers the direct connectivity between distribution network equipment, without accounting for the complexities introduced by unbalanced systems and untransposed lines, which can lead to mutual impedances. We will conduct research in this regard subsequently based on the suggestions.

 

Comments 6: 6- It will be beneficial if the four central characteristics of the node is explained briefly in line 231

Response 6: Thank you for pointing this out. Since the four features are common characteristics of the graph structure, and due to the length limit of the article, we will provide supplementary explanations by presenting relevant references.

 

Comments 7: 7- Please explain Fig 5, the voltage diagram diagram of IEEE123 node network. What is suggested here. IEEE 123 bus has ZIP loads. What is load model 1,2 and 3?

Response 7: Thank you for pointing this out. We have modified it to elaborate in more detail. Figure 5 shows the voltage diagram of the IEEE123 node network established based on OpenDSS simulation software. Three colors are used to distinguish the mode types of load changes with voltage. The load model is explained in detail on lines 372-380

 

Comments 8: 8- Please explain Fig 6, the power distribution diagram of IEEE123 node network. What does 100% load level and 40% load level mean?

Response 8: Thank you for pointing this out. We have modified it to elaborate in more detail. Figure 6 is a power distribution diagram showing the power levels in the lines connecting the network. Among them, the thickest line represents the maximum power level in the network, and the dotted line represents the minimum power level, which is roughly 40% load power.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all the revisions indicated correctly. They have completed the necessary corrections. They have especially highlighted the changes.

They have done an up-to-date study on the Graf topic, which has been popular and effective in recent years. I think that the study will be cited and will be of interest to readers. I will also ask my doctoral students in my research group to read and review it.

Thank you for the improvements.

Comments on the Quality of English Language

In singular and plural expressions, prepositions and prefixes, noticeable points can be corrected. For example; instead of "Distribution network is the main component of the power system" it should be "Distribution networks are the main component of the power systems".