Review Reports

Round 1

Reviewer 1 Report

In this study, the AM-Bi-LSTM network is presented for analyzing and detecting IOC faults using PD signals. The proposed model combines past and future information for fault detection, and assigns different weights to features based on their correlation with the final detection using AM. The performance of the model was evaluated using the ENET public dataset. Simulation results show that the proposed model outperforms existing LSTM, Bi-LSTM, and AM-LSTM models. The study also compares the influence of different features on classification results, and finds that combining statistical and entropy features leads to improved accuracy. Below are my comments.

- Please add one section to discuss relevant studies as only a few studies were highlighted in the introduction.

- The caption for Figs 2 and 3 requires additional information to provide a more comprehensive description of the figure.

- Please cite references for LSTM and BI-LASTM models.

- Line 173 ‘are all the parameters’ what do you mean? Please provide more details.

- Please provide a link/reference for the ENET dataset.

- Please make the description much clearer between metadata and signal data for the ENET dataset.

- Line 230, ‘which are smoother than the original signals.’ Please refer to Figures 9 and 10.

- Lines 248-249 ‘One-phase PD signal has 19 statistical features, so three-phase PD signals 248 have 57 dimensional features, which are shown in Fig.13.’ How do these extracted features become the signals shown in Fig13?

- Please add parameter setting values used with the proposed and compared models.

- Regarding the comparison between different algorithms, did the authors conduct simulations for the compare models, or it is based on some published results? Please highlight that in the manuscript. If it is based on some published results, please cite those works.

- Please add results as numbers as well to Figure 21.

 

- Please proofread the paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Summary:

The authors provide a deep learning based partial discharge (PD) model for detecting IOC faults.

General concept comments:

This is an extremely well written paper that is scientifically sound with appropriate study design and validation. The authors propose a novel noise reduction technique for PD based on wavelet transform. Based on the literature survey, the application of attention mechanisms for PD detection is new. The authors have compared the method with other algorithms in the field (Table 3). However, it is not clear if those algorithms are state-of-the-art in the field; the authors should clarify that.

Minor:

• “the evaluation index F1 is compared with the methods proposed in other papers” → Please provide references

• Typo: Change ICO to IOC in lines 12 and 202

I recommend accepting the article after the comments are satisfactorily addressed.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Review the Manuscript ID: electronics-2410869

Title: Bi-LSTM with Attention Mechanism for Partial Discharge Detection and Recognition in Insulated Overhead Conductor

In this work the authors proposed an AM-Bi-LSTM network to analyze and detect Insulated Overhead Conductor faults based on partial discharge signals.

The topic covered in the article is interesting and current.

The article is well structured, with an adequate bibliographic review.

However, authors must improve the paper.

Review the text and formatting of the document.

In my opinion the title of the article does not reflect the work presented

On line 12 and line 202 instead of  ICO, it should be IOC.

Section 3 - Simulation and analysis, could be improved, especially the sub-section 3.2 - PD signals de-noising and the sub-section 3.3 Feature extraction.

Review the text and formatting of the document.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf