Deep Neural Network with Hilbert–Huang Transform for Smart Fault Detection in Microgrid

Round 1

Reviewer 1 Report

This work presents a hybrid framework based on deep learning for detecting, identifying, and locating SC faults. The proposed framework is supported by a signal processing module using EMD and HHT. The attained results show the effectiveness of the proposed method. The paper is well-written and well-organized, and I have the following minor comments on this work:

- Literature review should include a discussion on time-frequency-domain signal processing techniques such as VMD, PMD, etc.

- The term fault classification (or identification) could be used instead of fault type categorization, which is a more general term used in the literature. 

- In contrast to the statement given in the Abstract, fault location has also been well-studied in power/smart grid systems, where data-driven models should also be reviewed in the Introduction. See 10.1109/JSYST.2020.3001932 and references therein. 

- The authors make use of time-series simulations, so why not make use of LSTM for the sake of FD/fault classification? 

- The study in https://doi.org/10.1016/j.engappai.2020.104150 shows that both the current and voltage measurements could be of superior performance for the sake of fault classification in distributed power systems. The selection of only current measurements for fault detection/classification in this work could be discussed.

- Does the proposed framework locate those faults that the network is not trained on? For instance, what would be the fault estimation error for the case, in which the fault is located on 5% of the faulty line?

Author Response

Reviewer # 1:

This work presents a hybrid framework based on deep learning for detecting, identifying, and locating SC faults. The proposed framework is supported by a signal processing module using EMD and HHT. The attained results show the effectiveness of the proposed method. The paper is well-written and well-organized, and I have the following minor comments on this work:

Authors’ Response: Firstly, we thank you for your accurate attention and valuable recommendations. We are glad to hear your positive feedback and would like to thank you for giving us constructive suggestions to improve the quality of the paper.

Comment # 1: Literature review should include a discussion on time-frequency-domain signal processing techniques such as VMD, PMD, etc.

Authors’ Response: Thank you very much for your good comment. A discussion is added in the introduction section based on your comments as below. Please see the paper.

According to refs [10,11], HHT can be used to reduce the dimensionality of a signal and decompose it. In general, HHT involves 2 stages: Firstly, applying an information-driven decomposition method called Empirical Mode Decomposition (EMD), and secondly, implementing Hilbert spectral analysis (HSA) on the decomposed signals, called Intrinsic Mode Functions (IMF's). In addition to all of these features, EMD is able to offer numerous enhancements over Fourier analysis, including the ability to automatically and adaptively extract oscillations from each signal; furthermore, this is straightforward to execute; and in addition, it is robust enough to cope with all non-linear and non-stationary signals. Additionally, EMD effectively captures non-linear features of amplitude and frequency modulation at local time scales. During IMFs, Hilbert spectral assessments give frequency information that is crucial to assessing non-stationary signals.

Comment # 2: The term fault classification (or identification) could be used instead of fault type categorization, which is a more general term used in the literature.

Authors’ Response: Thank you very much for your good comment. The term fault identification is used instead of fault type categorization. Please see the paper.

Comment # 3: In contrast to the statement given in the Abstract, fault location has also been well-studied in power/smart grid systems, where data-driven models should also be reviewed in the Introduction. See 10.1109/JSYST.2020.3001932 and references therein.

Authors’ Response: Thank you very much for your good comment and suggestion. This paper has been used to improve the quality of literature review as below. Please see the paper.

According to ref [15], just scheme components of 3-phase voltage values are used to locate faults in smart grids. A graph analytic-driven algorithm identifies the faulty lines by contributing system topology and attributing affinity values to the quest space to identify the affected areas by the fault. Using a wavelet multiresolution evaluation of the associated scheme components, heuristic indexes have been used to identify the faulty lines within the faulty areas.

Comment # 4: The authors make use of time-series simulations, so why not make use of LSTM for the sake of FD/fault classification? 

Authors’ Response: Thank you very much for your good comment. This section is added to conclusion as future work as below. Please see the paper.

LSTM models may be investigated in future research in order to achieve a prediction outcome capable of being used to diagnose faults by electrical utilities. It is promising that HHT can be applied in the future to improve LSTM scheme efficiency using the HHT and LSTM scheme combination.

Comment # 5: The study in https://doi.org/10.1016/j.engappai.2020.104150 shows that both the current and voltage measurements could be of superior performance for the sake of fault classification in distributed power systems. The selection of only current measurements for fault detection/classification in this work could be discussed.

Authors’ Response: Thank you very much for your good comment. Thank you very much for your good comment and suggestion. This paper has been used to improve the quality of introduction section as below. Please see the paper.

In ref [23], a new concrete characteristic choosing approach is proposed to improve fault detection and cyber-attacks in power systems using mutual data.

Comment # 6: Does the proposed framework locate those faults that the network is not trained on? For instance, what would be the fault estimation error for the case, in which the fault is located on 5% of the faulty line?

Authors’ Response: Thank you very much for your good comment. Several useful conclusions were drawn based on the expected FL errors. The majority of the relays were able to predict FLs fairly accurately, resulting in less than  for the training scenarios and  for the test scenarios.  and , on the other hand, have a fairly poor forecasting accuracy. It is due to the fact that Line 23 has a smaller length compared to the other lines. Therefore, for various fault lines on Line 23, the power dynamics would be the same as long as there is an adequate fault resistance. Using this suggested method, the fault location could be within an error of approximately . MG systems are relatively usual to involve underground fault recovery, so it makes sense.

Reviewer 2 Report

 

Review

 

The paper presented a novel HHT and DNN to construct an AC MG-FD model.

 

The paper is well written, and the text is fluent.

 

 

Small corrections:

 

Row 43:

 

DG hasn’t been defined before. Please define it.

 

Row 71:

 

adaptive network-based fuzzy inference system (ANFIS) hasn’t been defined before. Please define it.

Author Response

Reviewer # 2:

The paper presented a novel HHT and DNN to construct an AC MG-FD model.

The paper is well written, and the text is fluent.

Authors’ Response: Firstly, we thank you for your accurate attention and valuable recommendations. We are glad to hear your positive feedback and would like to thank you for giving us constructive suggestions to improve the quality of the paper.

Small corrections:

Comment # 1: Row 43: DG hasn’t been defined before. Please define it.

Authors’ Response: Thank you very much for your good attention. It is defined in the paper. Please see the paper.

Comment # 2: Row 71:adaptive network-based fuzzy inference system (ANFIS) hasn’t been defined before. Please define it.

Authors’ Response: Thank you very much for your good comment. . It is defined in the paper. Please see the paper.

Reviewer 3 Report

The paper presents implementation oh Hilbert transforms for fault detection in smart grid. Following are my observations.

a.     Authors have not employed some data pre-processing techniques for smoothing data? If yes kindly demonstrate the applicability.

b.     Representation of the results are poor, the voltages and other parameters shall be demonstrated properly. For example authors can see following paper.

1.       Saxena, A. (2022). An efficient harmonic estimator design based on Augmented Crow Search Algorithm in noisy environment. Expert Systems with Applications, 194, 116470.

c.      A supervised architecture may be evaluated on the basis of confusion values, Hence, it is required. Also what will be the implication of applying PCA or IDA in this case. How you will deal large data. Authors are encouraged to read the paper for confusion diagrams and scalability analysis.

2.       Saxena, A., Alshamrani, A. M., Alrasheedi, A. F., Alnowibet, K. A., & Mohamed, A. W. (2022). A Hybrid Approach Based on Principal Component Analysis for Power Quality Event Classification Using Support Vector Machines. Mathematics, 10(15), 2780.

d.     Table 7 should be explained in more lucid manner. Also the paper require lot of visualization of the data

Author Response

Reviewer # 3:

The paper presents implementation oh Hilbert transforms for fault detection in smart grid. Following are my observations.

Authors’ Response: Firstly, I would like to thank for your accurate attention and valuable recommendations. All your comments and suggestions have been thoroughly addressed to improve the quality of the paper; so the paper has been revised accordingly.

Comment # 1: Authors have not employed some data pre-processing techniques for smoothing data? If yes kindly demonstrate the applicability.

Authors’ Response: Thank you very much for your good comment. The singular value decomposition method is used to extract singular values to compute entropy of the IMF as feature. This section is added to the paper with description. Please see the paper.

Comment # 2: Representation of the results are poor, the voltages and other parameters shall be demonstrated properly. For example authors can see following paper.

1. Saxena, A. (2022). An efficient harmonic estimator design based on Augmented Crow Search Algorithm in noisy environment. Expert Systems with Applications, 194, 116470.

Authors’ Response: Thank you very much for your good comment and suggestion to improve the quality of the paper. The suggested work is useful and we use and cite it in our paper (Reference 20). Please see the paper.

Comment # 3: A supervised architecture may be evaluated on the basis of confusion values, Hence, it is required. Also what will be the implication of applying PCA or IDA in this case. How you will deal large data. Authors are encouraged to read the paper for confusion diagrams and scalability analysis.

2. Saxena, A., Alshamrani, A. M., Alrasheedi, A. F., Alnowibet, K. A., & Mohamed, A. W. (2022). A Hybrid Approach Based on Principal Component Analysis for Power Quality Event Classification Using Support Vector Machines. Mathematics, 10(15), 2780.

Authors’ Response: Thank you very much for your good comment. Firstly, Intrinsic Mode Functions (IMF's) is extracted, and then signal spectral energy has been obtained as descried in the paper, also singular value decomposition is used to extract singular values of IMF to compute the entropy of the IMF, then these features are used as input of DNNs. These parts added to the paper and described.

The suggested paper is useful and also used and cited in the paper (Reference 29). Please see the paper.

Comment # 4: Table 7 should be explained in more lucid manner. Also the paper require lot of visualization of the data

Authors’ Response: Thank you very much for your good comment. table 7 explained in more details as below, please see the paper.

As can be seen in this table, the impact and influence of noise in current measurements are very low, and it can be ignored. 30 db SNR is the worst scenario and the precision is decreased by less than 0.3 % in comparison of the perfect scenario. Hence, the measurement uncertainties have little impact on the FD efficiency, and it can be concluded that the efficiency of the suggested layout is approximately similar in different uncertainties.

Reviewer 4 Report

Dear Authors,

1) The paper could be improved by some extra information and comments about the current status of technical condition of grid, from the producer-customer point of view. In simple words, many areas experience PV shut downs in the summertime because electricity delivery system does not withstand it. This means sth. completely different do citizens of e.g. Spain, California or Russia. My comment aims in encouraging you to do small quantitative research on the condition of grid in selected region (like Europe) to place merits of your work in the concept of IIOT/Industry 4.0. Especially, since you mention BESS, large-scale influence of which is hard to predict. This is to be added as txt/table to Intro part.

2) I'm not sure if Figure 1 is correct. For instance, it shows that output of "Feature extraction" is one of inputs to "DNN for fault kind classification" as well as raw data. Why raw data and Features enter this block together? After all, features are extracted in order not to work on raw data. Here the word "attributes" in line 108 is not clear.

3) Describe all elements in Figure 2.

4) Differentiate names 3.2 and 4.1

5) No funding listed.

Author Response

Reviewer # 4:

Comment # 1:  The paper could be improved by some extra information and comments about the current status of technical condition of grid, from the producer-customer point of view. In simple words, many areas experience PV shut downs in the summertime because electricity delivery system does not withstand it. This means sth. completely different do citizens of e.g. Spain, California or Russia. My comment aims in encouraging you to do small quantitative research on the condition of grid in selected region (like Europe) to place merits of your work in the concept of IIOT/Industry 4.0. Especially, since you mention BESS, large-scale influence of which is hard to predict. This is to be added as txt/table to Intro part.

Authors’ Response: Firstly, I would like to thank for your accurate attention and valuable recommendations. All your comments and suggestions have been thoroughly addressed to improve the quality of the paper; so the paper has been revised accordingly.

Thanks for your valuable comment, some description are presented in the introduction section, also a paragraph added to the conclusion section as future work which is as below:

The concept of IIOT/Industry 4.0 with PV shut downs in the summertime for fault detection and cyber-attack detection can be considered in future works to separate fault and cyber-attack detection from each other.

In terms of operational concerns, protecting MGs is a significant issue [4]. As renewable energy resources become more prevalent in current energy grids, MGs tend to be connected to inverter-interfaced distributed generations (IIDG), including battery energy storage systems (BESS) and photovoltaic (PV)-distributed generation (DG). Fault currents can overwhelm conventional protection relays for FDM in distribution systems. Nevertheless, IIDGs produce small fault currents that prevent the activation of protective mechanisms [5]. Therefore, the MGs are not protected by such relays. In [6], the current and voltage dynamics of the MGs were thoroughly examined.

Comment # 2:  I'm not sure if Figure 1 is correct. For instance, it shows that output of "Feature extraction" is one of inputs to "DNN for fault kind classification" as well as raw data. Why raw data and Features enter this block together? After all, features are extracted in order not to work on raw data. Here the word "attributes" in line 108 is not clear.

Authors’ Response: Thank you very much for your good comment. Figure 1 is edited.  More description added to the paper to make it clear as below; please see the paper.

SVD is used to extract some features from IMFs and signal spectral energy that obtained from signal after is processed by HHT to use as input of DNNs.

It uses the 3-phases current magnitudes of a branch in a cycle gathered through a protection relay as the input information. Then, feature extraction was performed using HHT, and SVD. Subsequently, the attributes and features (signal spectral energy and entropy of IMFs) were fed into 3 DNNs for fault-kind FC, fault-phase detection, and FLD.

Comment # 3:  Describe all elements in Figure 2.

Authors’ Response: Thank you very much for your good comment. the elements are descried in the paper as below, please see the paper.

The elements in Figure 2 are described as follow: CB and R define the circuit breaker and relay, respectively. L and T define the load, and transformer, respectively. Line defines the transmission line.

Comment # 4: Differentiate names 3.2 and 4.1

Authors’ Response: Thank you very much for your good attention. Section 4.1 is “condensed layer and GRU” and this typo edited. Please see the paper.

Comment # 5: No funding listed.

Authors’ Response: Thank you very much for your comment. This research received no external funding, and this expressed in paper after conclusion section, please see the paper.

Round 2

Reviewer 3 Report

Improved.