Review Reports

Round 1

Reviewer 1 Report

 In the presented work, the authors present a novel arc fault detection method for residential low power grids based on the use of random fore  st, improved multi-scale permutation entropy and wavelet packet transform.
 
  The paper addresses an important and hot topic of arc faults detection in residential power networks. The results obtained and presented in t  his work are relevant and are of significant academic contribution. However, the work presented in this paper needs some major improvements.
 
  The style of the paper and the language should be considerably improved. There are lot of typos, mispelled words, misused verbs. A carreful p  roofreading of the paper by a natived skilled author is more than needed. Couple of examples of the style:
  * figure -> Figure
  * section -> Section (without the)
  * ac -> AC
>>* line 38: ... combustible(close to the ... -> ... combustible (close to the ...
  * line 74: identification[19] -> identification [19] (this misuse of brackets can be found throughout the whole paper!)
  * Figure 1: Control plotform -> Control platform
 
  Some bibliographic references are outdated. For instance, the reference [3] is a 1982 reference showing some statistical data! I'm pretty sur  e that there are more recent data on this subject. The same reference [3] is repeated twice [12]. The authors present an arc fault detection   method in residential low-power grids without citing the most recent works in the field. For instance, see the latest research papers publish  ed in the Electric Power Systems Research.
 
>>The writing style is from here to there confusing. For instance, in line 98, "This paper introduces the improved multi-scale permutation entr  opy (IMPE) [31] to avoid this problem". Here, it is a bit confusing if the IMPE method has been presented in this work or in the reference [3  1]. There are several sentences of this type in the paper.
 
  lines 96-97: "But the multi-scale analysis dose not make fulle use the sample point"; This should be reformulated
>>line 139: Then, the electrode(s) ...

The paragraph from line 117 to 126 is completely confusing and does not need to be in the introduction part. The authors present their method  , some details even some partial results in this paragraph. This should be moved in the next section.
  In the experimental data section, more details are needed about the used samples for training and test. The type of arc series faults? Carbon  ized path? Or other type? What is the duration of the recoded samples? Does the steady state is only observed or even the transient states?
 
  The authors "filter" the currents of all loads with the SVD method. How does this filtering influence the detection performances?
  Line 188: For any matrix Hm, which size -> whose size
  Line 191: some special character before lambda
  Line 198: singular value, however -> singular value. However
  Line 201: This paper, n=2 -> this should be reformulated and justified. Why only two components?
  Line 208: Firstly, the The matrix
  Line 209: Matrix A used without being introduced
  Line 233: This sentence should be reformulated. Too long
  Equation 7: Hp(m) is on the right and left hand side
  Lines 278-9 and 286-7 are repeating
  Line 300: dust collector ? why not be coherent and use throughout the whole paper vacuum cleaner?
  Line 308: are impossible to make distinguish -> to distinguish or to make difference
 
  The conclusions presented between the lines 341 and 356 can also be applied to all arc fault detection methods using energy as a discriminate   metrics between arc free and faulty signals. It is not only specific to this method and the permutation entropy and wpt. If instead of calcu  lating PE and WPT  to create feature matrix, we use simple frequency component of the load currents, would the obtained results be similar?
 
  Line 387: liner -> linear
 
  In Figure 10, y axis is represented in seconds. How this time is obtained?
 
  The authors use confusingly identification with detection. The more appropriate term is detection, because the identification is only related   to the features which are only specific to their work.
 
  Arc fault detection methods in residential power networks should be embedded eventually. How the proposed method is prone to embedded systems  ? Could you evaluate the cost and real time performances of the proposed method?
 
  Testing conditions should be more explained. How many times a recorded sample of one load is presented to the proposed method? Does the same   sample repeated several times give the same result in terms of detection?

Line 423: Which type of NN is implemented? MLP or other?
  The authors chose 190 neurons in the hidden layer. How they obtained this value? The mention a "cross-validation" method? More details about   this method should be provided
 
  Table 1: The number of correctly identify -> identified

Author Response

attached

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments are attache.

Comments for author File: Comments.pdf

Author Response

attached

Author Response File: Author Response.pdf

Reviewer 3 Report

The following comments will improve the paper quality;

In the introduction please make a discussion on different fault types and please also make a discussion on previous researches on fault location. The following papers are the best you can refer to in this regard;

a) Precise fault location on transmission lines using ensemble Kalman filter

b) Current-only directional overcurrent protection using postfault current

c) Ground fault location using the low voltage side record data in distribution systems

Please provide more explanation on fig. 5

Please present equation (2) with more explanations and details

Please represent the information in table 2 and table 3 in form of curve so that their comparison would be easier to the readers

The whole paper must be proofread carefully.

Please give more explanation on fig. 11.

Please also discuss the effect of other transient events like the capacitor switching, power swing, inrush current, power electronic loads, and furnaces in dection of arc faults. In the following are some of good references to refer to in this regard;

a) High-Impedance Fault Detection in the Distribution Network Using the Time-Frequency-Based Algorithm

b) The comparison of two contol methods of power swing reduction in power system with UPFC compensator

c) Detection of High Impedance Faults Using Wavelet Transform

d) Reachability analysis for a grid-connected voltage-sourced converter (VSC)

Author Response

Attached

Author Response File: Author Response.pdf