A DC Arc Fault Detection Method Based on AR Model for Photovoltaic Systems
, Xiang Li 1,2, Yunsheng Ban 4,*, Xiaochen Ma 1,2, Chenguang Hao 1,2, Jiawang Zhou 1,2 and Huimao Cai 5
Round 1
Reviewer 1 Report
This paper presents an arc fault detection method based on autoregressive (AR) model. The Burg algorithm is used to determine model parameters and the Akaike Information Criterion (AIC) is used to determine the AR model order.
Some clarification of the presented method should be given. Do you use a set of historical normal operation data to estimate the AR model and use this AR model for fault diagnosis? Or did you continuously updating the AR model with new operation data? If the model is developed from a set of historical normal operation data, then the size of the data set should be given.
How is the fault detection threshold determined? Some clarification should be given.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
One of the most intriguing issues in PV systems is DC arc faults, which are discussed in this study. The study's findings are intriguing and warrant publishing.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The authors use the difference in the arc current's non-stationary characteristics and the PWM's noise (pulse width modulation) to identify arc faults. The authors have presented a good overview of the literature (comments on improving it are discussed later), a reasonable argument for using ML and PSD from the dataset, and a threshold parameter to indicate faults. There are a few typos and some of the discussions need clarity. But overall, it is a well-written paper.
The typos, suggestions, and clarification points are listed below:
[1] Eqn. 10: The formula for the correlation coefficient is incorrect, it should be SD for each variable.
[2] Line 228: Does the size of the data buffer become the limiting factor for the analysis?
[3] In what scenario would you expect the noise from PWM to override the method and result in a false positive (for instance the observation window is too or too large) since for the selected window it seems to perform well? (The reviewer agrees that such a study could be a part of future work but it would be nice for the readers to understand possible scenarios where this may fail but would need to be verified by further testing)
[4] It would be nice if the authors could improve the quality of the figures.
[5] The reviewer thinks that it would be good if the algorithm is compared with the current literature as compared to a commercial device and the report on the performance of commercial and current SOA methods.
[6] Line 204: It is unclear what is meant by exploring the best order.
[7] Line 167: What is the value of p used for this study?
[8] Line 50: It is unclear what is meant by three windows (it should be more specific such as the time window of observation).
[9] Line 39: The full form of the acronym TZ1 should be specified?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors have adequately addressed my comments and the revised manuscript can be accepted.
