Closure on “Multiresolution S-Transform-Based Fuzzy Recognition System for Power Quality Events”

Abstract

Had the authors actually omitted the negative values of m from their algorithm, they could not have obtained the results shown in Figs. 1 and 2 (which appear to be correct).3) In the right-hand side (RHS) of (12), it appears that a has been set equal to 0 (i.e., that the original S-transform as defined in [2] is being used). I suggest that the RHS of (12) be changed to e 0[2 m =(aNT +bjnj) ] for consistency with the scaling function used in (4). 4) In step 3 of the list following (13), G[n; m] should be replaced with G[m; n] for consistency with the earlier notation. Similarly, S[n; j] should read S[j; n] in step 6; the indices of S are also reversed in (13). 5)Step 5 of the same list should read "Determine B[n; m] = H[m+n]1G[m; n]," instead of "Determine B(n; m) = H(m+ n) 1 G(m + n)." The quantity G defined in (12) has two arguments. 6) The text following (14) contains Matlab code that should be explained. Here, s is a matrix whose columns denote times and whose rows denote frequencies. Max(s) is a row vector that contains the maxima of the columns of s; max(s 0 ) is a row vector that contains the maxima of the rows of s, and in the definitions of A 2 and A 4 ; min(max(abs(s 0 ))) denotes the smallest number obtained when the maximum absolute value of the S-transform is found at each frequency. 7) The scaling function of (4) is similar to a scaling function defined by [3], viz. (f) = 1=(a + bf). Equation (4) is preferable to the scaling function of [3], because the use of jfj in place of f prevents (f) from becoming infinite at f = 0a=b. It also ensures that the same window is used at f = f 0 and f = 0f 0 (here, f0 is an arbitrary frequency), which is necessary if the S-transform is to be used for time-frequency filtering. REFERENCES[1] M. V. Chilukuri and P. K. Dash, "Multiresolution S-transform-based fuzzy recognition system for power quality events,"The authors would like to thank the discusser for his interest in the paper and valuable comments and suggestions.1) As the discusser correctly pointed out, (7) should be 1 01 S(; f) = H(f):2) The limits of (11) do not need to be changed [2], [3]. However, Fig. 2(c) should be replaced with the following figure, which is obtained by the plotting magnitude of the S-transform matrix. 3) Suggestion (3) can be considered for maintaining consistency of the equations. 4) Suggestion (4) can be considered for maintaining consistency of the equations. 5) The authors apologize for the error in the text after (14). It should be read as A 1 = max(max(abs(s))) with PQ disturbance; A 2 = min(max(abs(s))) with PQ disturbance; B1 = max(max(abs(s))) without PQ disturbance; B 2 = min(max(abs(s))) without PQ disturbance. where max(abs(s)) is a row vector, plotting upon which gives the amplitude versus time plot as shown in Fig. 3(c). Hence, A1 is maxima and A 2 is minima of the amplitude versus time plot with PQ disturbance. Similarly, B1 is maxima and B2 is minima of the amplitude versus time plot without PQ disturbance. 6) As suggested by the discusser, step (5) on page 2 should be...

Power quality disturbance signal segmentation and classification based on modified BI-LSTM with double attention mechanism

“…It is found from the results that the classification accuracy of the training confusion matrix is 99.73% for nine types of PQD signals with SNR of 20 dB. The classification accuracy for the proposed model is also presented for nine different basic PQD signals [15] with SNR of 0 dB (noiseless), 20dB and 40dB of noise. Table 3 shows the classification accuracy of nine…”

“…In [14] nine types of PQDs are classified using a modification of S transform method (Double-Resolution S-transform (DRST)) for essential feature extraction and Directed Acyclic Graph Support Vector Machines (DAG-SVMs) a SVM based method for classification of the disturbance with an average accuracy of 97%. In [15] a combination of S transform and Fuzzy logic is applied for PQD classification of fourteen types of PQDs obtaining an accuracy of 98% with SNR ratio of 0dB. In [7] a classification accuracy of 99% is obtained for 16 types of PQD using HHT as feature extraction method and Weighted bidirectional extreme learning machine (WBELM) as signal classifier.…”

Power quality disturbance signal segmentation and classification based on modified BI‐LSTM with double attention mechanism

“…It is found from the results that the classification accuracy of the training confusion matrix is 99.73% for nine types of PQD signals with noise level of X 1 . The classification accuracy for the proposed model is also presented for nine different basic PQD signals [15] without noise, X 1 level and X 2 levels of noise.…”

“…[14], nine types of PQDs has been classified using a modified version of the S‐transform method called Double‐Resolution S‐transform (DRST) for essential feature extraction and Directed Acyclic Graph Support Ve(DAG‐SVMs), a SVM‐based method for classification of the disturbance with an average accuracy of 97%. A combination of S transform and FL has been applied for PQD classification of fourteen types of PQDs obtaining an accuracy of 98% without a noise [15]. In ref.…”

Automatic power quality recognition and analysis system using wavelet-based neural network

“…Up until now, considerable efforts have been focused on this area, for instance, assessing impacts brought about by deterioration of power quality, monitoring variant disturbances occurring in transmission and distribution networks, and seeking measures for power service improvement [1,2] . Hence, to improve power quality, it is required to know the sources of power system disturbances and find ways to mitigate them [3][4][5] . To monitor electrical power quality disturbance, short time discrete Fourier transform (STFT) is most often used.…”