Circuit breaker fault diagnosis method and device based on multi-source signal fusion

Abstract

The invention belongs to the technical field of circuit breaker fault diagnosis, and particularly relates to a circuit breaker fault diagnosis method and device based on multi-source signal fusion. The method comprises the steps that a current signal and a vibration signal of the circuit breaker are collected, the current signal is an opening and closing coil current signal, and the vibration signal comprises a closing coil vibration signal and an operating mechanism vibration signal; the collected current signals and vibration signals are preprocessed, corresponding characteristic parameters are extracted, and a characteristic vector set containing characteristic parameters of opening and closing coil current signals, closing coil vibration signals and operating mechanism vibration signals is constructed; and inputting the feature vector set into a circuit breaker fault diagnosis model obtained by training a machine learning classifier based on the feature vector sets of the circuit breaker in different operation states, thereby obtaining a circuit breaker fault diagnosis result, improving the generalization ability and accuracy of the fault diagnosis model, and further improving the accuracy of circuit breaker fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of circuit breaker fault diagnosis technology, specifically relating to a circuit breaker fault diagnosis method and device based on multi-source signal fusion. Background Technology

[0002] With the continuous construction of my country's new power system, the power grid operation is becoming increasingly complex. The ever-increasing grid load and the integration of more and more new energy power are placing higher demands on the stability and reliability of the power system. Circuit breakers play a control and protection role in the power system, and their stable and reliable operation has a direct impact on the safety and stability of the power grid. Therefore, research on circuit breaker fault diagnosis is of great significance for ensuring the safety of the power system and improving socio-economic benefits.

[0003] To address this issue, Chinese invention patent application CN112345213A discloses a method for diagnosing mechanical faults in low-voltage DC circuit breakers. This method involves acquiring a current signal from one of the opening / closing coils and a mechanical vibration signal from the low-voltage DC circuit breaker. The acquired signals are preprocessed to extract feature parameters. A support vector machine (SVM) algorithm is used to model the circuit and find optimal parameters. The feature parameters from the current signal and the short-time energy characteristics and wavelet packet frequency band energy characteristics of the vibration signal are then combined. The K-Fold algorithm is used to determine the feature parameter combination with the highest accuracy. Finally, the feature parameter combination is dimensionality-reduced, and a new diagnostic model for the low-voltage DC circuit breaker is constructed using the dimensionality-reduced features. However, this method uses only one current signal and one vibration signal for fault feature extraction and requires finding optimal parameters to optimize the classification algorithm, resulting in low accuracy in circuit breaker fault diagnosis. Summary of the Invention

[0004] The purpose of this invention is to provide a circuit breaker fault diagnosis method and device based on multi-source signal fusion, so as to solve the problem of low diagnostic accuracy of existing circuit breaker fault diagnosis methods.

[0005] This invention provides a circuit breaker fault diagnosis method based on multi-source signal fusion to solve the above-mentioned technical problems. The method involves collecting current and vibration signals from the circuit breaker, where the current signal is the opening and closing coil current signal, and the vibration signal includes the closing coil vibration signal and the operating mechanism vibration signal. The collected current and vibration signals are preprocessed, and corresponding feature parameters are extracted to construct a feature vector set containing feature parameters of the opening and closing coil current signal, the closing coil vibration signal, and the operating mechanism vibration signal. This feature vector set is then input into a circuit breaker fault diagnosis model to obtain the circuit breaker fault diagnosis result. The circuit breaker fault diagnosis model is obtained by training a machine learning classifier using a feature vector set formed from the current and vibration signals of the circuit breaker under different operating conditions.

[0006] Furthermore, the characteristic parameters of the opening and closing coil current signal include the maximum value of the current waveform, the time corresponding to the maximum value, the current value corresponding to the first peak before the maximum value, the time corresponding to the first peak before the maximum value, the current value corresponding to the first trough before the maximum value, and the time corresponding to the first trough before the maximum value.

[0007] Furthermore, the characteristic parameters of the opening and closing coil current signal include a first current value, a first time, the current value corresponding to the first peak before the maximum value of the current waveform, the time corresponding to the first peak before the maximum value of the current waveform, the current value corresponding to the first trough before the maximum value of the current waveform, and the time corresponding to the first trough before the maximum value of the current waveform. The method for obtaining the first current value and the first time is as follows: after determining the maximum value of the current waveform, select all extreme points in the current waveform that are greater than a set multiple of the maximum value of the current waveform, and take the average value of the current values ​​corresponding to these extreme points, which is the first current value; take the average value of the time corresponding to these extreme points, which is the first time; the set multiple is greater than 0.8 and less than 1.

[0008] Furthermore, the current waveform is the waveform of the current signal obtained by performing mode decomposition on the preprocessed opening and closing coil current signal and reconstructing it using the mode components of a set order obtained after mode decomposition.

[0009] Furthermore, the characteristic parameters of the vibration signal include short-time energy and wavelet packet band energy; the short-time energy includes the time point of the short-time energy extreme value and the corresponding energy, and the wavelet packet band energy includes the band energy of a predetermined number obtained by 6-layer wavelet packet decomposition.

[0010] Furthermore, the machine learning classifier is a support vector machine with linear kernel function parameters.

[0011] Furthermore, the feature vector set refers to the feature vector set after feature dimensionality reduction.

[0012] Furthermore, the dimensionality reduction process includes: using linear discriminant analysis to perform linear dimensionality reduction on the feature vector set, determining the maximum dimension after feature dimensionality reduction to be K-1 based on the number of fault categories K, then constructing feature vector sets from 1 to K-1 dimensions, and using cross-validation algorithm to cross-validate the machine learning classifier based on the feature vector sets of each dimension after dimensionality reduction and the corresponding fault types, training to obtain the dimension of the feature vector set with the highest accuracy of the circuit breaker fault diagnosis model.

[0013] Furthermore, the preprocessing includes: removing zero drift of the current signal using the averaging method and smoothing the current signal using a sliding window filter; and removing the trend term of the vibration signal using the least squares method.

[0014] The beneficial effects of the above technical solution are as follows: This invention is an improved invention. It integrates the feature parameters of the opening and closing coil current signal, the closing coil vibration signal, and the operating mechanism vibration signal to construct a feature vector set. Based on the feature vector set of the circuit breaker under different operating states, a machine learning classifier is trained to obtain a circuit breaker fault diagnosis model. During use, the opening and closing coil current signal and vibration signals at two locations of the circuit breaker are collected, and a feature vector set after the fusion of the three signals is constructed. This feature vector set is then input into the trained circuit breaker fault diagnosis model, thus obtaining an accurate circuit breaker fault diagnosis result. Compared to using only one current signal and one vibration signal for fault diagnosis, this invention integrates the feature vector set after the fusion of three multi-source signals for fault diagnosis, improving the generalization ability and accuracy of the fault diagnosis model, thereby improving the accuracy of circuit breaker fault diagnosis.

[0015] To address the aforementioned technical problems, the present invention also provides a circuit breaker fault diagnosis device based on multi-source signal fusion, comprising a processor, wherein the processor is used for computer program instructions to implement the circuit breaker fault diagnosis method based on multi-source signal fusion described above.

[0016] The beneficial effects of the above technical solution are as follows: This invention is an improved invention. It integrates the feature parameters of the opening and closing coil current signal, the closing coil vibration signal, and the operating mechanism vibration signal to construct a feature vector set. Based on the feature vector set constructed under different operating states of the circuit breaker, a machine learning classifier is trained to obtain a circuit breaker fault diagnosis model. During use, the opening and closing coil current signal and vibration signals at two locations of the circuit breaker are collected, and a feature vector set after the fusion of the three signals is constructed. This feature vector set is then input into the trained circuit breaker fault diagnosis model, thus obtaining an accurate circuit breaker fault diagnosis result. Compared to using only one current signal and one vibration signal for fault diagnosis, this invention integrates the feature vector set after the fusion of three multi-source signals for fault diagnosis, improving the generalization ability and accuracy of the fault diagnosis model, thereby improving the accuracy of circuit breaker fault diagnosis. Attached Figure Description

[0017] Figure 1 This is a flowchart of a circuit breaker fault diagnosis method based on multi-source signal fusion according to an embodiment of the present invention.

[0018] Figure 2 This is a flowchart of the circuit breaker fault diagnosis model establishment process according to an embodiment of the method of the present invention;

[0019] Figure 3 This is a flowchart of the current signal feature parameter extraction process according to an embodiment of the method of the present invention;

[0020] Figure 4 This is a flowchart of linear discriminant analysis dimensionality reduction and five-fold cross-validation diagnostics in an embodiment of the method of the present invention;

[0021] Figure 5-a This is a comparison chart of the single-source signal accuracy of embodiments of the method of the present invention;

[0022] Figure 5-b This is a comparison chart of the accuracy of multi-source signals in an embodiment of the method of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0024] This invention is a pioneering innovation. It integrates feature parameters from the opening and closing coil current signal, the closing coil vibration signal, and the operating mechanism vibration signal to construct a feature vector set. Based on this feature vector set constructed under different operating conditions of the circuit breaker, a machine learning classifier is trained to obtain a circuit breaker fault diagnosis model. During use, the circuit breaker's opening and closing coil current signal and vibration signals at two locations are collected, and a fused feature vector set of the three signals is constructed. This fused feature vector set is then input into the trained circuit breaker fault diagnosis model, resulting in an accurate circuit breaker fault diagnosis. Compared to using only a current signal and a vibration signal for fault diagnosis, this invention integrates a feature vector set from three multi-source signals, improving the generalization ability and accuracy of the fault diagnosis model, thereby enhancing the accuracy of circuit breaker fault diagnosis.

[0025] Method Implementation Examples

[0026] This invention provides a circuit breaker fault diagnosis method based on multi-source signal fusion, such as... Figure 1 As shown, the diagnostic process is as follows:

[0027] 1. Based on the feature vector set formed by the current and vibration signals of the circuit breaker under different operating conditions, a machine learning classifier is trained to obtain a circuit breaker fault diagnosis model. Specifically, such as... Figure 2 As shown, it includes the following steps:

[0028] 1) The current and vibration signals of the circuit breaker under different operating conditions were collected through fault simulation tests. The current signal is the current signal of the opening and closing coils, and the vibration signal includes the vibration signal of the closing coil and the vibration signal of the operating mechanism.

[0029] This invention conducts circuit breaker fault simulation tests based on mechanical and electrical circuit fault types. Fault states include: poor coil circuit contact during opening / closing, abnormal coil control voltage during opening / closing, jamming of the opening electromagnet core during opening, abnormal gap of the opening electromagnet during opening, failure of the mechanism to move during opening / closing, spring fatigue during opening / closing, loose foundation bolts during opening / closing, and jamming of the transmission mechanism during opening / closing, including the normal state during opening / closing. Fault simulations are performed according to the simulation methods and simulation times in Table 1, resulting in 28 fault categories and 560 samples. Current and vibration signals of the circuit breaker under each fault category are collected.

[0030] Table 1

[0031]

[0032]

[0033] 2) Preprocess the collected current and vibration signals.

[0034] Specifically, the preprocessing of the current signal includes: using the averaging method to remove the zero drift of the current signal, and using sliding window filtering to smooth and denoise the current signal; the preprocessing of the vibration signal includes: using the least squares method to remove the trend term of the vibration signal.

[0035] 3) Extract the feature parameters of the current signal and the two vibration signals, and construct a higher-dimensional feature vector set based on the extracted feature parameters, which includes the feature parameters of the opening and closing coil current signal, the closing coil vibration signal and the operating mechanism vibration signal.

[0036] In one embodiment, to improve the accuracy of the feature parameters, the current signal feature parameter extraction method of the present invention is as follows: Figure 3As shown, the process includes: after preprocessing the current signal, performing mode decomposition on the preprocessed opening and closing coil current signal using the ensemble empirical mode decomposition method (EEMD) to further remove high-frequency noise from the opening and closing coil. The current signal is then reconstructed using mode components of a set order obtained after mode decomposition. In this embodiment, the set order is selected according to requirements; in this embodiment, mode components of orders 6 to 12 are selected to reconstruct the current signal. Then, all the maximum and minimum points of the current waveform are determined, and the maximum value Imax of the current waveform and the corresponding time tmax are found. Traversing the extreme points, find the current value I2 corresponding to the first trough before the maximum value and the time t2 corresponding to the first trough before the maximum value; traversing the extreme points, find the current value I1 corresponding to the first peak before the maximum value and the time t1 corresponding to the first peak before the maximum value based on I2; traversing the extreme points, select all extreme points in the current waveform that are greater than a set multiple of the maximum value of the current waveform, average the current values ​​corresponding to these extreme points (this average value is the first current value), and average the times corresponding to these extreme points (this average value is the first time). The characteristic parameters of the opening and closing coil current signal are a six-dimensional characteristic parameter constructed from I1, t1, I2, t2, the first current value, and the first time. The set multiple is greater than 0.8 and less than 1, and can be set according to requirements during use. In this embodiment, the preferred set multiple is 0.9.

[0037] In another embodiment, the maximum value Imax of the initially found current waveform and the time tmax corresponding to the maximum value are directly used as feature parameters. At this time, the feature parameters of the opening and closing coil current signal are I1, t1, I2, t2, Imax and tmax.

[0038] The characteristic parameters of the vibration signal include short-time energy and wavelet packet frequency band energy. Short-time energy includes the time points of short-time energy extrema and their corresponding energies. Wavelet packet frequency band energy includes the energies of the first 32 frequency bands obtained from 6-level wavelet packet decomposition. The method for extracting the short-time energy characteristic parameters is as follows: a rectangular window of length 50 is used as a moving window function to perform short-time energy analysis on the preprocessed vibration signal, extracting the time points of short-time energy extrema and their energies as feature vectors. The method for extracting the wavelet packet frequency band energy characteristic parameters is as follows: the db10 wavelet is used as the mother wavelet function, and the preprocessed vibration signal is decomposed into 6 levels of wavelet packets using Shannon entropy as the standard. The energy of the first set number of frequency bands is obtained as characteristic parameters. The set number is selected according to requirements; in this embodiment, the first 32 frequency band energies are selected as characteristic parameters.

[0039] 4) Based on the feature vector sets of the circuit breaker under different operating states, a fault diagnosis model is trained using a machine learning classifier. In this embodiment, the K-fold (cross-validation) algorithm is used to divide the training and test sets. The divided data is then normalized to the [0, 1] interval using a maximum-minimum normalization method. Five machine learning classifiers are constructed using a third-party library and subjected to five rounds of 5-fold cross-validation and training to obtain the diagnostic accuracy of the five machine learning classifiers. The machine learning classifier with the highest accuracy is selected as the circuit breaker fault diagnosis model. The machine learning classifiers are Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Bayesian, LightGBM, and Random Forest classifiers. All machine learning classifiers use default parameters, eliminating the need for algorithms to find optimal parameters. The SVM uses a linear kernel function, and the K value for the K-Nearest Neighbor algorithm is set to 3. Preferably, the circuit breaker fault diagnosis model of this invention is trained using a Support Vector Machine with default linear kernel function parameters.

[0040] 2. Collect real-time current and vibration signals from the circuit breaker, preprocess the collected current and vibration signals and extract the corresponding feature parameters, and construct a feature vector set containing feature parameters of the opening and closing coil current signal, closing coil vibration signal and operating mechanism vibration signal; input the feature vector set into the circuit breaker fault diagnosis model to obtain the circuit breaker fault diagnosis result.

[0041] As a preferred implementation, in order to avoid dimensionality explosion and further improve algorithm efficiency and model generalization ability, the feature vector set used during training and input into the circuit breaker fault diagnosis model is the feature vector set after feature dimensionality reduction.

[0042] Specifically, such as Figure 4 As shown, linear discriminant analysis (LDA) is used to linearly reduce the dimensionality of the feature vector set. Based on the number of fault categories K, the maximum dimension after dimensionality reduction is determined to be K-1. Feature vector sets from 1 to K-1 dimensions are constructed. Based on the feature vector sets of each dimension after dimensionality reduction and their corresponding fault types, a cross-validation algorithm is used to cross-validate the machine learning classifier, obtaining the dimension of the feature vector set that achieves the highest accuracy for the circuit breaker fault diagnosis model. The machine classification model is then trained using this dimensional feature vector set, resulting in the dimensionality-reduced circuit breaker fault diagnosis model.

[0043] Specifically, taking K=28 as an example, the maximum dimension of the feature vector set after dimensionality reduction is determined to be 27. The K-fold algorithm is used to divide the feature vector set into training and testing sets. The divided training and testing sets are normalized to the interval [0, 1]. Linear discriminant analysis is used to perform linear dimensionality reduction on the feature vector dataset, constructing training and testing sets from 1 to 27 dimensions. Based on the training and testing sets of each dimension, five-fold cross-validation is performed in the machine learning classifier to obtain the prediction accuracy of each machine learning classifier under different dimensions. The feature dimension of the feature vector set with the highest accuracy is selected for fault diagnosis. In this embodiment, the optimal dimension obtained after dimensionality reduction is 17, and this 17-dimensional feature vector set is obtained by dimensionality reduction transformation of all feature vectors.

[0044] The accuracy of multi-source signal fusion, which combines the eigenvectors of a single-source signal, a coil signal, and two vibration signals into four different combinations, is compared to that of other signals. Figure 5-a and Figure 5-b As shown, the fault diagnosis accuracy of single-source signal circuit breakers is the highest at 91.67%, while the fault diagnosis accuracy of multi-source signal circuit breakers using feature vector sets containing characteristic parameters of opening and closing coil current signals, closing coil vibration signals, and operating mechanism vibration signals can reach 99.73%, greatly improving the accuracy of diagnosis.

[0045] Device Examples

[0046] The present invention provides a circuit breaker fault diagnosis device based on multi-source signal fusion, comprising a processor for executing computer program instructions to implement a circuit breaker fault diagnosis method based on multi-source signal fusion as described in the method embodiments of the present invention.

Claims

1. A circuit breaker fault diagnosis method based on multi-source signal fusion, characterized in that, Includes the following steps: The circuit breaker's current and vibration signals are collected. The current signal is the opening and closing coil current signal, and the vibration signal includes the closing coil vibration signal and the operating mechanism vibration signal. The collected current and vibration signals are preprocessed and the corresponding feature parameters are extracted to construct a feature vector set containing feature parameters of the opening and closing coil current signal, the closing coil vibration signal and the operating mechanism vibration signal; The feature vector set is input into the circuit breaker fault diagnosis model to obtain the circuit breaker fault diagnosis result; the circuit breaker fault diagnosis model is obtained by training a machine learning classifier based on the feature vector set formed by the current signal and vibration signal of the circuit breaker under different operating conditions.

2. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 1, characterized in that, The characteristic parameters of the opening and closing coil current signal include the maximum value of the current waveform, the time corresponding to the maximum value, the current value corresponding to the first peak before the maximum value, the time corresponding to the first peak before the maximum value, the current value corresponding to the first trough before the maximum value, and the time corresponding to the first trough before the maximum value.

3. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 1, characterized in that, The characteristic parameters of the opening and closing coil current signal include a first current value, a first time, the current value corresponding to the first peak before the maximum value of the current waveform, the time corresponding to the first peak before the maximum value of the current waveform, the current value corresponding to the first trough before the maximum value of the current waveform, and the time corresponding to the first trough before the maximum value of the current waveform. The method for obtaining the first current value and the first time is as follows: after determining the maximum value of the current waveform, select all extreme points in the current waveform that are greater than a set multiple of the maximum value of the current waveform, and take the average value of the current values ​​corresponding to these extreme points. This average value is the first current value. Take the average value of the time corresponding to these extreme points. This average value is the first time. The set multiple is greater than 0.8 and less than 1.

4. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 2 or 3, characterized in that, The current waveform is the waveform of the current signal obtained by performing mode decomposition on the preprocessed opening and closing coil current signal and reconstructing it using the mode components of a set order obtained after mode decomposition.

5. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 1, characterized in that, The characteristic parameters of the vibration signal include short-time energy and wavelet packet band energy; short-time energy includes the time point of the short-time energy extreme value and the corresponding energy, and wavelet packet band energy includes the band energy of a predetermined number obtained by 6-layer wavelet packet decomposition.

6. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 1, characterized in that, The machine learning classifier is a support vector machine with linear kernel function parameters.

7. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 1, characterized in that, The feature vector set refers to the feature vector set after feature dimensionality reduction.

8. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 7, characterized in that, The dimensionality reduction process includes: using linear discriminant analysis to perform linear dimensionality reduction on the feature vector set; determining the maximum dimension of the feature vector set after dimensionality reduction to be K-1 based on the number of fault categories K; then constructing a feature vector set from 1 to K-1 dimensions; and using the feature vector sets of each dimension after dimensionality reduction and the corresponding fault types to perform cross-validation on the machine learning classifier using a cross-validation algorithm, thereby training to obtain the dimension of the feature vector set with the highest accuracy of the circuit breaker fault diagnosis model.

9. The circuit breaker fault diagnosis method based on multi-source signal fusion according to claim 1, characterized in that, The preprocessing includes: removing zero drift of the current signal using the averaging method and smoothing the current signal using a sliding window filter; and removing the trend term of the vibration signal using the least squares method.

10. A circuit breaker fault diagnosis device based on multi-source signal fusion, comprising a processor, characterized in that, The processor is used to execute computer program instructions to implement the circuit breaker fault diagnosis method based on multi-source signal fusion as described in any one of claims 1-9.

Images