A Detection Method for On-Load Tap Changer Faults Based on Singular Value Entropy

Abstract

The present application relates to the technical field of transformer equipment fault detection, and discloses a detection method for on-load tap-changer faults based on singular value entropy. The current signal during the switching of the on-load tap-changer is obtained through a current sensor, and the vibration signal of the on-load tap-changer is obtained through an acceleration sensor; according to the current signal, the main vibration waveform during the switching of the switch is obtained from the vibration signal; the main vibration waveform is processed to determine the singular value entropy of the intrinsic mode components; the singular value entropy of the intrinsic mode components is reduced in dimension by principal component analysis, and the K-Means unsupervised clustering algorithm is used to separate different types of fault samples from non-fault samples to complete fault detection and classification. The present application can effectively and timely determine the type of fault that occurs after the on-load tap-changer fails.

Description

Technical Field

[0001] The present application relates to the technical field of transformer equipment fault detection, and particularly to a detection method for on-load tap-changer faults based on singular value entropy. Background Art

[0002] On-load tap-changers are widely used in various power transformers and play a key role in on-load voltage regulation. Faults often cause serious consequences. The on-load tap-changer is the only moving mechanism in the transformer, with complex mechanical structures and electrical characteristics. Its faults can be divided into mechanical faults and electrical faults. According to statistics, mechanical faults of on-load tap-changers account for more than 80%, and electrical faults are usually evolved from mechanical faults.

[0003] On-load tap-changers are usually placed inside the transformer tank and immersed in insulating oil. Non-intrusive detection methods such as vibration signals and sound signals are the most commonly used detection means for on-load tap-changers. The switching operation of the on-load tap-changer has a fixed time sequence, and the vibration signal can reflect the mechanical state during the switching of the on-load tap-changer to a certain extent. As a typical non-stationary signal, traditional time-domain and frequency-domain analysis methods are difficult to comprehensively extract the characteristics of vibration signals and cannot timely determine the type of fault after a fault occurs. Summary of the Invention

[0004] The present application discloses a detection method for on-load tap-changer faults based on singular value entropy to solve the technical problem that in the prior art, traditional time-domain and frequency-domain analysis methods are difficult to comprehensively extract the characteristics of vibration signals and cannot timely determine the type of fault after a fault occurs.

[0005] The present application discloses a detection method for on-load tap-changer faults based on singular value entropy, including:

[0006] Obtaining the current signal during the switching of the on-load tap-changer through a current sensor, and obtaining the vibration signal of the on-load tap-changer through an acceleration sensor;

[0007] Obtaining the main vibration waveform during switch switching in the vibration signal according to the current signal;

[0008] Processing the main vibration waveform to determine the singular value entropy of the intrinsic mode component;

[0009] Performing principal component analysis and dimensionality reduction on the singular value entropy of the intrinsic mode component, and using the K-Means unsupervised clustering algorithm to separate different types of fault samples from non-fault samples to complete fault detection and classification.

[0010] Optionally, the obtaining the main vibration waveform during switch switching in the vibration signal according to the current signal includes:

[0011] Determine the start and end of the vibration signal according to the current signal, and align the vibration signal on the time axis;

[0012] Perform principal component truncation on the vibration signal aligned on the time axis;

[0013] Extract time series features of the signal according to the truncated vibration signal, and obtain the main vibration waveform when the switch is switched.

[0014] Optionally, the performing principal component truncation on the vibration signal aligned on the time axis includes:

[0015] Based on the peak value of the vibration signal at each switch, perform principal component truncation on the vibration signal according to the preset vibration time of the switching switch, and the vibration time of the switching switch is 250 ms.

[0016] Optionally, the processing the main vibration waveform to determine the singular value entropy of the intrinsic mode component includes:

[0017] Perform empirical mode decomposition on the main vibration waveform to obtain multiple intrinsic mode components;

[0018] Reconstruct the multiple intrinsic mode components to obtain a Hankel matrix;

[0019] Perform singular value decomposition on the Hankel matrix to obtain the singular values of the Hankel matrix, and perform normalization processing;

[0020] Substitute the result of the normalization processing into the singular value entropy calculation formula to determine the singular value entropy of multiple intrinsic mode components.

[0021] Optionally, the performing empirical mode decomposition on the main vibration waveform to obtain multiple intrinsic mode components includes:

[0022] The basis function of the empirical mode decomposition is automatically generated according to the characteristics of the main vibration waveform.

[0023] Optionally, the reconstructing the multiple intrinsic mode components to obtain a Hankel matrix includes:

[0024] Determine the Hankel matrix through the following formula:

[0025]

[0026] Among them, A represents the Hankel matrix, x(t) represents the vibration signal, N represents the preset vibration signal length, and K represents the preset phase space length.

[0027] Optionally, performing singular value decomposition on the Hankel matrix to obtain the singular values of the Hankel matrix and performing normalization processing includes:

[0028] Performing singular value decomposition through the following formula:

[0029] H = USV T ;

[0030] S = diag(σ1, σ2, σ3,...., σ K );

[0031] where S = diag(σ1, σ2, σ3,...., σ K ) represents the singular value matrix of matrix H, and U and V represent orthogonal matrices;

[0032] Squaring the singular value matrix S to obtain and performing normalization processing, we have and we can get ∑E i = 1, meeting the normalization condition of entropy.

[0033] Optionally, substituting the result of normalization processing into the singular value entropy calculation formula to determine the singular value entropy of multiple intrinsic mode components includes:

[0034] The singular value entropy calculation formula is as follows:

[0035] P = -∑E i lnE i ;

[0036] where P represents the singular value entropy of the intrinsic mode component. F

[0037] Optionally, performing principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode component includes:

[0038] Performing principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode component, reducing 9-dimensional features to 2-dimensional features to reduce the risk of overfitting.

[0039] Optionally, the length of the vibration signal is 12800, and the length of the phase space is 1024.

[0040] This application relates to the technical field of transformer equipment fault detection, and discloses a detection method for on-load tap-changer faults based on singular value entropy. The current signal during the switching of the on-load tap-changer is obtained through a current sensor, and the vibration signal of the on-load tap-changer is obtained through an acceleration sensor; according to the current signal, the main vibration waveform during the switching of the switch is obtained from the vibration signal; the main vibration waveform is processed to determine the singular value entropy of the intrinsic mode components; the singular value entropy of the intrinsic mode components is reduced in dimension by principal component analysis (PCA), and the K-Means unsupervised clustering algorithm is used to separate different types of fault samples from non-fault samples to complete fault detection and classification. This application can effectively and timely determine the type of fault that occurs after the on-load tap-changer fails. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of this application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 It is a schematic diagram of the working process of a detection method for on-load tap-changer faults based on singular value entropy provided by an embodiment of this application;

[0043] Figure 2 It is a waveform diagram of the vibration signal during the switching of the on-load tap-changer under normal conditions provided by an embodiment of this application;

[0044] Figure 3 It is a waveform diagram of the main vibration signal during the switching of the on-load tap-changer under normal conditions provided by an embodiment of this application;

[0045] Figure 4 It is a waveform diagram of the main vibration signal during the switching of the on-load tap-changer under the fault of the vacuum tube provided by an embodiment of this application;

[0046] Figure 5 It is a schematic diagram of each intrinsic mode component after empirical mode decomposition under the fault state provided by an embodiment of this application;

[0047] Figure 6 It is a schematic diagram of the result after K-Means unsupervised clustering provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] In order to solve the technical problem in the prior art that traditional time-domain and frequency-domain analysis methods are difficult to comprehensively extract the characteristics of vibration signals and cannot timely determine the type of fault that occurs after a fault occurs, this application discloses a detection method for on-load tap-changer faults based on singular value entropy through the following embodiments.

[0049] The present application discloses a detection method for on-load tap-changer faults based on singular value entropy. Refer to Figure 1 the schematic diagram of the workflow shown in

[0050] Step S101: Obtain the current signal during the switching of the on-load tap-changer through a current sensor, and obtain the vibration signal of the on-load tap-changer through an acceleration sensor.

[0051] Among them, the acceleration sensor is placed near the top cover of the on-load tap-changer.

[0052] Step S102: According to the current signal, obtain the main vibration waveform during the switching of the switch in the vibration signal.

[0053] In some embodiments of the present application, the obtaining the main vibration waveform during the switching of the switch in the vibration signal according to the current signal includes:

[0054] According to the current signal, determine the start and end of the vibration signal, and align the vibration signal on the time axis.

[0055] Perform principal component truncation on the vibration signal aligned on the time axis.

[0056] Extract time series features of the signal according to the truncated vibration signal to obtain the main vibration waveform during the switching of the switch in the vibration signal.

[0057] Further, the performing principal component truncation on the vibration signal aligned on the time axis includes:

[0058] Based on the peak value of the vibration signal during each switch switching, perform principal component truncation on the vibration signal according to the preset vibration time of the switching switch. The vibration time of the switching switch is 250 ms, reducing unnecessary vibration information during analysis.

[0059] Step S103: Process the main vibration waveform to determine the singular value entropy of the intrinsic mode component.

[0060] In some embodiments of the present application, the processing the main vibration waveform to determine the singular value entropy of the intrinsic mode component includes:

[0061] Perform empirical mode decomposition (EMD) on the main vibration waveform to obtain multiple intrinsic mode functions (IMFs).

[0062] Reconstruct the multiple intrinsic mode components to obtain a Hankel matrix. Specifically, use the delay embedding technique to perform phase space reconstruction on the IMF components to construct a Hankel matrix

[0063] Perform singular value decomposition on the Hankel matrix to obtain the singular values of the Hankel matrix and perform normalization processing.

[0064] Substitute the result of the normalization process into the singular value entropy calculation formula to determine the singular value entropy of multiple intrinsic mode components.

[0065] In some embodiments of the present application, the empirical mode decomposition of the main vibration waveform to obtain multiple intrinsic mode components includes:

[0066] The basis function of the empirical mode decomposition is automatically generated according to the characteristics of the main vibration waveform, avoiding the problem of difficulty in pre-selecting the optimal basis function similar to wavelet (packet) decomposition.

[0067] In some embodiments of the present application, the reconstruction of the multiple intrinsic mode components to obtain a Hankel matrix includes:

[0068] Determine the Hankel matrix through the following formula:

[0069]

[0070] where A represents the Hankel matrix, x(t) represents the vibration signal, N represents the preset length of the vibration signal, and K represents the preset length of the phase space.

[0071] In some embodiments of the present application, the length of the vibration signal is 12,800 and the length of the phase space is 1,024.

[0072] Specifically, the vibration signal is intercepted for 0.25 s and the sampling rate is 51.2 kHz. Therefore, the length N of the vibration signal is 12,800. Assuming that the length of the phase space (number of columns) K is as large as possible at 1,024, and the number of rows is (N - K + 1) = 11,776, finally a Hankel matrix A of order K*(N - K + 1) is obtained. Therefore, the Hankel matrix A can be determined as follows:

[0073]

[0074] In some embodiments of the present application, the singular value decomposition of the Hankel matrix to obtain the singular values S = diag(σ1, σ2, σ3,...., σ K ), and perform normalization processing, including:

[0075] Perform singular value decomposition through the following formula:

[0076] H = USV T ;

[0077] S = diag(σ1, σ2, σ3,...., σ K );

[0078] Wherein, S = diag(σ1, σ2, σ3,...., σ K ) represents the singular value matrix of matrix H, and U and V represent orthogonal matrices.

[0079] Substitute the obtained Hankel matrix into the formula H = USV T to obtain the singular value matrix S = diag(σ1, σ2, σ3,...., σ K ), and then square the singular value matrix S to obtain and perform normalization processing, resulting in and it can be obtained that ∑E i = 1, meeting the normalization condition of entropy. Then substitute it into the formula for calculating entropy to calculate the IMF singular value entropy of the vibration signal x(t).

[0080] In some embodiments of the present application, substituting the result of the normalization processing into the singular value entropy calculation formula to determine the singular value entropy of multiple intrinsic mode components includes:

[0081] The singular value entropy calculation formula is as follows:

[0082] P = -∑Ei l nE i ;

[0083] Wherein, P represents the singular value entropy of the intrinsic mode component.

[0084] Specifically, by comparing the entropies of the singular values of the IMF components in different states, the fault type can be accurately identified. Compared with only calculating the singular values of the matrix, the singular value entropy can better reflect the differences in the complexity of the signals in each frequency band, thereby more accurately identifying the fault type.

[0085] Step S104, perform principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode component, and use the K-Means unsupervised clustering algorithm to separate different types of fault samples from non-fault samples to complete fault detection and classification.

[0086] In some embodiments of the present application, performing principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode component includes:

[0087] Perform principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode component, reducing the 9-dimensional features to 2-dimensional features, thereby reducing the complexity of the K-Means unsupervised clustering algorithm (the feature matrix changes from R m×n (m is the number of samples, n is the dimension before dimensionality reduction) to Rm×2 ) to reduce the risk of overfitting.

[0088] Specifically, the singular value eigenvectors after dimensionality reduction are brought into the K-Means unsupervised clustering algorithm to accurately separate different types of fault samples from non-fault samples, achieving the effect of fault classification and diagnosis.

[0089] A method for detecting on-load tap-changer faults based on singular value entropy disclosed in the above embodiments of the present application obtains the current signal during the switching of the on-load tap-changer through a current sensor and the vibration signal of the on-load tap-changer through an acceleration sensor; obtains the main vibration waveform during the switching in the vibration signal according to the current signal; processes the main vibration waveform to determine the singular value entropy of the intrinsic mode components; performs principal component analysis (PCA) dimensionality reduction on the singular value entropy of the intrinsic mode components, and uses the K-Means unsupervised clustering algorithm to separate different types of fault samples from non-fault samples to complete fault detection and classification. The present application can effectively and timely determine the type of fault that occurs after a fault occurs in the on-load tap-changer.

[0090] Exemplarily, the acceleration sensor is placed on the top cover of the on-load tap-changer to collect the vibration signal during the switching of the on-load tap-changer, and the sampling frequency is 51.2 kHz. The vibration signals during switching in the normal and faulty states are collected, and the original signals are as Figure 2 shown.

[0091] A total of 250 ms of the main vibration signal is intercepted based on the highest amplitude, and the intercepted signal already contains all the information of the tap-changer operation. The main vibration signals of the normal and vacuum tube faults unified on the time axis are as Figure 3 and Figure 4 shown.

[0092] See Figure 5 , which is a schematic diagram of each intrinsic mode component after empirical mode decomposition in the fault state provided by the embodiment of the present application.

[0093] As shown in Table 1, it can be seen that the entropy values of the singular values of each intrinsic mode component are relatively large when a fault occurs. The entropy values of the singular values of the components extracted by this method can provide reliable feature components for the subsequent application of the K-Means unsupervised clustering algorithm and provide key information for fault identification and classification.

[0094] Table 1

[0095]

[0096] The total number of samples in this embodiment is 45, and 20 of them are faulty samples. Therefore, the entropy of the singular values obtained above is further reduced in dimension by PCA, reducing the 9-dimensional features to 2-dimensional features, thereby reducing the complexity of the classification algorithm (the feature matrix changes from R 45×9 to R 45×2 ), reducing the risk of overfitting.

[0097] Furthermore, the dimension-reduced singular value feature vectors obtained above are brought into the K-Means unsupervised clustering algorithm to accurately separate different types of faulty samples from non-faulty samples, achieving the effect of fault classification and diagnosis. See Figure 6 , which is a schematic diagram of the results after K-Means unsupervised clustering provided by the embodiment of the present application.

[0098] The present application has been described in detail above in combination with specific embodiments and exemplary examples, but these descriptions should not be construed as limiting the present application. Those skilled in the art understand that without departing from the spirit and scope of the present application, various equivalent replacements, modifications or improvements can be made to the technical solutions and their implementation manners of the present application, and these all fall within the scope of the present application. The protection scope of the present application is subject to the appended claims.

Claims

1. A detection method for on-load tap-changer faults based on singular value entropy, characterized in that Including: Obtaining a current signal when the on-load tap-changer switches through a current sensor, and obtaining a vibration signal of the on-load tap-changer through an acceleration sensor; According to the current signal, obtaining the main vibration waveform during switch switching in the vibration signal; Processing the main vibration waveform to determine the singular value entropy of the intrinsic mode components; Performing principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode components, and using the K-Means unsupervised clustering algorithm to separate different types of fault samples from non-fault samples to complete fault detection and classification; The processing of the main vibration waveform to determine the singular value entropy of the intrinsic mode components includes: Performing empirical mode decomposition on the main vibration waveform to obtain multiple intrinsic mode components; Reconstructing the multiple intrinsic mode components to obtain a Hankel matrix; Performing singular value decomposition on the Hankel matrix to obtain the singular values of the Hankel matrix and performing normalization processing; Substituting the result of the normalization processing into the singular value entropy calculation formula to determine the singular value entropy of multiple intrinsic mode components; The obtaining of the main vibration waveform during switch switching in the vibration signal according to the current signal includes: According to the current signal, determining the start and end of the vibration signal and aligning the vibration signal on the time axis; Performing principal component truncation on the vibration signal aligned on the time axis; Performing time series feature extraction on the signal according to the truncated vibration signal to obtain the main vibration waveform during switch switching in the vibration signal; The reconstructing of the multiple intrinsic mode components to obtain a Hankel matrix includes: Determining the Hankel matrix through the following formula: ； where A represents a Hankel matrix, represents the vibration signal, N represents the preset length of the vibration signal, and K represents the preset length of the phase space; The performing of singular value decomposition on the Hankel matrix to obtain the singular values of the Hankel matrix and performing normalization processing includes: Performing singular value decomposition through the following formula: ； ； Among them, represents the singular value matrix of matrix H, and U and V represent orthogonal matrices; Square the singular value matrix S to obtain and perform normalization, resulting in , and , from which we can obtain , meeting the normalization condition of entropy; The substituting of the result of the normalization processing into the singular value entropy calculation formula to determine the singular value entropy of multiple intrinsic mode components includes: The singular value entropy calculation formula is as follows: ； Where, P represents the singular value entropy of the intrinsic mode components.

2. The on-load tap-changer fault detection method according to claim 1, characterized in that The performing of principal component truncation on the vibration signal aligned on the time axis includes: Based on the peak value of the vibration signal during each switch switching, performing principal component truncation on the vibration signal according to the preset vibration time of the switching switch, and the vibration time of the switching switch is 250 ms.

3. The on-load tap-changer fault detection method according to claim 1, characterized in that The performing of empirical mode decomposition on the main vibration waveform to obtain multiple intrinsic mode components includes: The basis function of the empirical mode decomposition is automatically generated according to the characteristics of the main vibration waveform.

4. The on-load tap-changer fault detection method according to claim 1, characterized in that, The performing of principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode components includes: Performing principal component analysis dimensionality reduction on the singular value entropy of the intrinsic mode components, reducing the 9-dimensional features to 2-dimensional features to reduce the risk of overfitting.

5. The on-load tap-changer fault detection method according to claim 1, characterized in that, The length of the vibration signal is 12800, and the length of the phase space is 1024.

Images