Circuit breaker fault identification method and device, storage medium and computer equipment

Abstract

The invention discloses a circuit breaker fault identification method and device, a storage medium and computer equipment, and the method comprises the steps: synchronously collecting a multi-mode signal of a target circuit breaker in a switching-on and switching-off process, determining a time-frequency diagram corresponding to the multi-mode signal, and carrying out the isomorphic processing of the time-frequency diagram of the multi-mode signal, and obtaining an isomorphic time-frequency diagram; determining an isomorphic time-frequency matrix corresponding to the isomorphic time-frequency graph, expanding the isomorphic time-frequency matrix into an isomorphic time-frequency vector according to a matrix column sequence, dividing the isomorphic time-frequency vector into a plurality of vector blocks, and determining a local mask matrix corresponding to each vector block; determining a local precision matrix corresponding to each local mask matrix, and determining a local energy contribution diagram of the multi-modal signal based on the local precision matrixes; and performing deep joint feature extraction on the local energy contribution diagram of the multi-modal signal to obtain a fault sensitive feature vector of the target circuit breaker, and determining a fault identification result of the target circuit breaker based on the fault sensitive feature vector.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment fault diagnosis technology, and in particular to a circuit breaker fault identification method, device, storage medium, and computer equipment. Background Technology

[0002] Circuit breakers are crucial switching devices in power systems, responsible for current interruption, fault isolation, and safe system operation. Their operational status directly impacts the safety, stability, and reliability of the power system. With the continuous expansion of power grids and the sustained increase in transmission voltage and capacity, circuit breakers are susceptible to mechanical shocks, electrothermal stress, and environmental factors during long-term operation. This can lead to problems such as energy storage mechanism failures, loosening of transmission components, spring fatigue, and insulation aging, resulting in potential hazards such as slow opening and closing, failure to operate, false tripping, or arc extinguishing failure. If a faulty circuit breaker fails to promptly disconnect the fault current, it can trigger equipment damage, busbar outages, regional power outages, or even system-wide cascading failures, causing significant economic losses and safety risks. Therefore, the rapid and accurate identification and diagnosis of potential circuit breaker faults is of paramount importance for ensuring the safe and stable operation of the power system.

[0003] Currently, circuit breaker faults are usually identified manually. However, manual identification is time-consuming and labor-intensive, and due to varying skill levels or negligence among staff, errors in circuit breaker fault identification can occur. Summary of the Invention

[0004] This invention provides a circuit breaker fault identification method, device, storage medium, and computer equipment, which mainly improves the identification efficiency and accuracy of circuit breaker faults.

[0005] According to a first aspect of the present invention, a circuit breaker fault identification method is provided, comprising: In response to the fault identification signal of the target circuit breaker, the multi-mode signals of the target circuit breaker during the opening and closing process are synchronously acquired, wherein the multi-mode signals include acoustic signals and vibration signals; The time-frequency diagram corresponding to the multimodal signal is determined, and isomorphic processing is performed on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal. Determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency map, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block. Determine the local precision matrix corresponding to each local mask matrix, and based on the local precision matrix, determine the local energy contribution map of the multimodal signal; Deep joint feature extraction is performed on the local energy contribution map of the multimodal signal to obtain the fault-sensitive feature vector of the target circuit breaker. Based on the fault-sensitive feature vector, the fault identification result of the target circuit breaker is determined.

[0006] Optionally, before determining the time-frequency diagram corresponding to the multimodal signal, the method further includes: Obtain the sample circuit breaker at different times t during the opening and closing process. i Multiple sets of sample sound signals and multiple sets of sample vibration signals were used to determine multiple time delay variables. ; For each time delay variable , each t i The sample sound signal at time and the corresponding Multiply the sample vibration signals at time t, and then multiply each t... i The results of multiplying the time points are summed to obtain each time delay variable. Evaluation parameters ,in, ; In each evaluation parameter, a maximum evaluation parameter is determined, and the time delay variable corresponding to the maximum evaluation parameter is taken as the optimal time delay between the sound signal and the vibration signal. Based on the optimal time delay, the sound signal and the vibration signal in the multimodal signal are time-aligned to obtain the aligned sound signal and the aligned vibration signal.

[0007] Optionally, before determining the time-frequency diagram corresponding to the multimodal signal, the method further includes: The multimodal signal is subjected to multi-scale wavelet decomposition to obtain different levels. Approximation coefficients and detail coefficients at different levels of j ; Determine the detail coefficients Each coefficient value the median of Based on each coefficient value and the median Determine the noise standard deviation of the multimodal signal. ,in, , Noise figure; Determine the signal-to-noise ratio of the multimodal signal. Based on the signal-to-noise ratio Determine the signal-to-noise ratio factor of the multimodal signal. ,in, ; Based on the noise standard deviation and the signal-to-noise ratio factor Determine the detail coefficients contraction threshold ,in, N is the length of the multimodal signal; Based on the shrinkage threshold For the detail coefficients Perform shrinkage processing to obtain the detail coefficients after shrinkage. ,in, , It is a contraction function; For the approximation coefficients and the detail coefficient after shrinkage Perform inverse wavelet transform to obtain the denoised multimodal signal.

[0008] Optionally, determining the time-frequency diagram corresponding to the multimodal signal includes: The multimodal signals are unified to the same sampling rate, and the window length, frame shift, and FFT points for constructing the time-frequency graph are determined. Based on the window length, frame shift, and FFT points, the multimodal signals are segmented to determine the spectral representation of each segment. The spectral representations of each segment are then spliced ​​together according to the time dimension to obtain the time-frequency graph of the multimodal signals. Optionally, the isomorphic processing of the time-frequency graph of the multimodal signal to obtain the corresponding isomorphic time-frequency graph of the multimodal signal includes: The energy from the lowest frequency to the current frequency in the time-frequency graph is accumulated, and the ratio of the accumulated energy to the total energy corresponding to the time-frequency graph is used as the cumulative energy percentage of the current frequency. When the accumulated energy percentage is greater than a preset percentage threshold, the current frequency is used as the online frequency index. The lowest frequency is used as the lower frequency index. Based on the online frequency index and the offline frequency index Determine the characteristic frequency band of the time-frequency diagram. ; Based on the characteristic frequency band The time-frequency graph is cropped, and the cropped time-frequency graph is used as an isomorphic time-frequency graph.

[0009] Optionally, determining the local mask matrix corresponding to each vector block includes: Determine the eight neighboring nodes corresponding to each node in each vector block. Based on each node and its corresponding eight neighboring nodes, construct a local mask matrix corresponding to each vector block. In the local mask matrix, the matrix elements at the corresponding positions of the nodes in the vector block and their corresponding eight neighboring nodes are 1. The remaining matrix elements after removing the positions with matrix elements of 1 in the local mask matrix are 0. Each node corresponds to a time-frequency unit in the isomorphic time-frequency graph.

[0010] Optionally, determining the local precision matrix corresponding to each local mask matrix includes: Determine the mean and covariance of the block-level vector elements within each vector block; Diagonally shrink the block-level vector element covariance to obtain the shrunken block-level vector element covariance. Using the covariance of the elements of the shrinking block-level vector as a statistic, and the local mask matrix as a constraint, the local precision matrix corresponding to each local mask matrix is ​​determined. Determining the local energy contribution map of the multimodal signal based on the local precision matrix includes: Determine the channel map vector of the random field channel map with the same size as the isomorphic time-frequency map. And based on the local precision matrix Vector of vector block i Determine the local energy contribution map of the multimodal signal. ,in, .

[0011] Optionally, the step of performing deep joint feature extraction on the local energy contribution map of the multimodal signal to obtain the fault-sensitive feature vector of the target circuit breaker includes: The local energy contribution maps of the multimodal signals are superimposed to obtain superimposed features. The superimposed features are then input into a preset feature extraction model for feature extraction to obtain the fault-sensitive feature vector of the target circuit breaker. The step of determining the fault identification result of the target circuit breaker based on the fault-sensitive feature vector includes: Identify the different clusters corresponding to different fault types, and determine the centroid vectors corresponding to different clusters; Determine the distance between the fault-sensitive feature vector and the centroid vectors corresponding to different clusters, and based on the distance, determine the target cluster to which the fault-sensitive feature vector belongs, and take the fault type corresponding to the target cluster as the fault type of the target circuit breaker.

[0012] According to a second aspect of the present invention, a circuit breaker fault identification device is provided, comprising: The acquisition unit is used to synchronously acquire the multi-mode signals of the target circuit breaker during the opening and closing process in response to the fault identification signal of the target circuit breaker, wherein the multi-mode signals include acoustic signals and vibration signals; The processing unit is configured to determine the time-frequency diagram corresponding to the multimodal signal, and perform isomorphic processing on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal. The first determining unit is used to determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency map, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in matrix column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block. The second determining unit is used to determine the local precision matrix corresponding to each local mask matrix, and to determine the local energy contribution map of the multimodal signal based on the local precision matrix. The fault identification unit is used to perform deep joint feature extraction on the local energy contribution map of the multimodal signal to obtain the fault sensitive feature vector of the target circuit breaker, and to determine the fault identification result of the target circuit breaker based on the fault sensitive feature vector.

[0013] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described circuit breaker fault identification device.

[0014] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned circuit breaker fault identification device.

[0015] According to the present invention, a circuit breaker fault identification method, apparatus, storage medium, and computer equipment, compared with the current method of identifying circuit breaker faults manually, the present invention simultaneously acquires multi-modal signals such as acoustic signals and vibration signals during the opening and closing process of the target circuit breaker. Acoustic signals reflect the sound characteristics generated by the movement and friction of internal mechanical components of the circuit breaker, while vibration signals reflect the vibration state of the mechanical structure. The combination of these two signals provides more comprehensive information on the circuit breaker's operating status, and compared to single-modal signals, it can more accurately capture fault characteristics, thereby improving the accuracy of circuit breaker fault identification. The time-frequency diagrams of the multi-modal signals are subjected to isomorphic processing to obtain isomorphic time-frequency diagrams, allowing different modal signals to be processed simultaneously. The invention features a unified representation in the time-frequency domain, facilitating the use of the same analysis methods and models, simplifying subsequent analysis processes, and improving processing efficiency. Expanding the isomorphic time-frequency matrix into vectors and dividing it into vector blocks, and determining the local mask matrix and local precision matrix for each vector block, allows focusing on local regions of the time-frequency graph. This avoids the time wasted analyzing redundant regions unrelated to fault identification and also prevents interference from redundant regions, thus improving the efficiency and accuracy of circuit breaker fault identification. Furthermore, deep joint feature extraction of the local energy contribution map of multimodal signals uncovers hidden features within the data, thereby improving the accuracy of circuit breaker identification. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart of a circuit breaker fault identification method provided by an embodiment of the present invention is shown; Figure 2 An embodiment of the present invention provides a time-frequency diagram; Figure 3 This invention provides a local energy contribution map according to an embodiment of the invention. Figure 4 A flowchart of another circuit breaker fault identification method provided by an embodiment of the present invention is shown; Figure 5 This diagram illustrates the structure of a circuit breaker fault identification device according to an embodiment of the present invention. Figure 6 This invention provides a schematic diagram of the structure of another circuit breaker fault identification device according to an embodiment of the invention. Figure 7 A schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0017] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0018] Currently, manually identifying circuit breaker faults is time-consuming and labor-intensive, and due to varying skill levels or negligence among staff, errors in fault identification can occur.

[0019] To address the aforementioned problems, embodiments of the present invention provide a circuit breaker fault identification method, such as... Figure 1 As shown, the method includes: 101. In response to the fault identification signal of the target circuit breaker, the multi-mode signals of the target circuit breaker during the opening and closing process are collected synchronously. The multi-mode signals include acoustic signals and vibration signals.

[0020] In this embodiment of the invention, during the circuit breaker's opening and closing operation, acoustic and vibration signals of the circuit breaker are acquired using an acoustic sensor and a vibration acceleration sensor, respectively. This embodiment of the invention simultaneously acquires multi-modal signals, including acoustic and vibration signals, during the target circuit breaker's opening and closing process. The acoustic signals reflect the sound characteristics generated by the movement and friction of internal mechanical components within the circuit breaker, while the vibration signals reflect the vibration state of the mechanical structure. Combining these two signals provides more comprehensive information on the circuit breaker's operating status. Compared to single-modal signals, this approach can more accurately capture fault characteristics, thereby improving the accuracy of circuit breaker fault identification. 102. Determine the time-frequency diagram corresponding to the multimodal signal, and perform isomorphic processing on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal.

[0021] In this embodiment of the invention, to ensure signal quality, preprocessing of the multimode signal is required after acquisition. Therefore, the method includes: acquiring the sample circuit breaker at different times t during the opening and closing process. i Multiple sets of sample sound signals and multiple sets of sample vibration signals were used to determine multiple time delay variables. For each time delay variable , each t i The sample sound signal at time and the corresponding Multiply the sample vibration signals at time t, and then multiply each t... i The results of multiplying the time points are summed to obtain each time delay variable. Evaluation parameters ,in, In each evaluation parameter, a maximum evaluation parameter is determined, and the time delay variable corresponding to the maximum evaluation parameter is used as the optimal time delay between the sound signal and the vibration signal. Based on the optimal time delay, the sound signal and the vibration signal in the multimodal signal are time-aligned to obtain the aligned sound signal and the aligned vibration signal.

[0022] Specifically, multiple time delay variables are determined, which should cover a reasonable time range, such as from -T to +T, with a step size of [value missing]. , where T and Based on the actual situation, for each time delay variable, the sample sound signal of each ti is compared with the corresponding... Sample vibration signal at time 10:00 Multiplication: Sum the results of multiplication at each time step ti to obtain each time delay variable. Evaluation parameters In all the evaluation parameters obtained from the calculation In this process, the maximum evaluation parameter is determined. The time delay variable corresponding to this maximum evaluation parameter is taken as the optimal time delay between the sound signal and the vibration signal. Based on the optimal time delay obtained above, the sound signal and vibration signal in the multimodal signal are time-aligned. Specifically, according to the value of the optimal time delay, the sound signal or vibration signal is time-shifted accordingly to align them on the time axis. For example, if the optimal time delay is positive, it means that the vibration signal lags behind the sound signal, and the vibration signal can be shifted forward by the optimal time delay; if the optimal time delay is negative, it means that the sound signal lags behind the vibration signal, and the sound signal is shifted forward by the corresponding absolute value of time. Through the above specific implementation steps, the optimal time delay between the sound signal and the vibration signal during the opening and closing process of the sample circuit breaker can be effectively determined, and their time alignment can be achieved, providing an accurate data basis for subsequent analysis and diagnosis of the circuit breaker status.

[0023] Furthermore, to further improve signal quality, denoising processing can be performed on the multimodal signal. Based on this, the method includes: performing multi-scale wavelet decomposition on the multimodal signal to obtain different levels of noise reduction. Approximation coefficients and detail coefficients at different levels of j Determine the detail coefficients Each coefficient value the median of Based on each coefficient value and the median Determine the noise standard deviation of the multimodal signal. ,in, , The noise figure is used to determine the signal-to-noise ratio of the multimodal signal. Based on the signal-to-noise ratio Determine the signal-to-noise ratio factor of the multimodal signal. ,in, Based on the noise standard deviation and the signal-to-noise ratio factor Determine the detail coefficients contraction threshold ,in, N is the length of the multimodal signal; based on the shrinkage threshold For the detail coefficients Perform shrinkage processing to obtain the detail coefficients after shrinkage. ,in, , For the contraction function; for the approximation coefficients and the detail coefficient after shrinkage Perform inverse wavelet transform to obtain the denoised multimodal signal.

[0024] Specifically, when mechanical equipment experiences an early failure, the vibration and sound signals it generates are very weak and easily drowned out by the powerful vibrations and noise of normal operation. To obtain a more accurate signal with a high signal-to-noise ratio, denoising processing of the original signal is necessary. First, the multimodal signal (sound or vibration signal) is decomposed into multi-scale wavelet decompositions using the following formula:

[0025] in, This represents any modal signal within a multimodal signal. The noise standard deviation is then determined. To further mitigate the impact of noise, the universal threshold is adaptively modified by incorporating signal-to-noise ratio (SNR): a signal-to-noise ratio factor is constructed. Then based on the noise standard deviation and signal-to-noise ratio factor Determine the detail coefficients contraction threshold Based on this shrinkage threshold For detail coefficients After shrinkage processing, the approximation coefficients are finally adjusted according to the following formula. and the detail coefficient after shrinkage Perform inverse wavelet transform to obtain the denoised multimodal signal. :

[0026] in, It is the inverse wavelet transform function.

[0027] Furthermore, in order to identify circuit breaker faults, it is first necessary to determine the time-frequency diagram of the time-aligned and denoised multimodal signal. Based on this, step 102 specifically includes: unifying the multimodal signal to the same sampling rate, and determining the window length, frame shift, and FFT points for constructing the time-frequency diagram; segmenting the multimodal signal based on the window length, frame shift, and FFT points; determining the spectral representation of each signal segment; and splicing the spectral representations of each signal segment according to the time dimension to obtain the time-frequency diagram of the multimodal signal.

[0028] Specifically, the time-frequency diagram of the multimodal signal is determined according to the following formula:

[0029] Specifically, sound signals Set window function The length L, such as L=1024, the frame shift H, such as H=256, and the number of FFT points N. FFT , such as N FFT =1024, for audio signals Starting from n=0, the window is moved step by step according to the frame shift H. When m=0, the signal is taken. The data segment from n=0 to n=L-1, i.e. , and window function Multiply to obtain the windowed signal. Then perform N on this windowed signal. FFT =1024-point FFT transformation, multiplied by an exponential term ,get The value of , where k is the frequency index, when m=1, the window moves H=256 sampling points, at which time the signal is taken. The data from n=256 to n=256+L-1=1279, that is... Then with window function Multiply, perform FFT transformation and other operations to obtain By continuously moving the window and calculating the FFT for each frame, a two-dimensional time-frequency graph is eventually obtained. Where m represents the frame index and k is the index frequency, the time-frequency diagrams of the sound and vibration signals can be plotted. For example... Figure 2 The image shows an example of a time-frequency diagram.

[0030] Furthermore, after determining the time-frequency graph of each modal signal, isomorphic processing needs to be performed on each time-frequency graph. Based on this, the method includes: accumulating the energy from the lowest frequency to the current frequency in the time-frequency graph, and using the ratio of the accumulated energy to the total energy corresponding to the time-frequency graph as the accumulated energy percentage of the current frequency; when the accumulated energy percentage is greater than a preset percentage threshold, using the current frequency as the online frequency index. The lowest frequency is used as the lower frequency index. Based on the online frequency index and the offline frequency index Determine the characteristic frequency band of the time-frequency diagram. Based on the characteristic frequency band The time-frequency graph is cropped, and the cropped time-frequency graph is used as an isomorphic time-frequency graph.

[0031] The preset percentage threshold is set according to actual needs. Specifically, the time-frequency graph is converted into an energy representation as follows:

[0032] Where L is the length of the window function. The time-frequency energy of the time-frequency diagram, where frequency index k corresponds to the actual physical frequency f. k As shown below:

[0033] Where NFFT is the number of points in the Fourier transform, f s These are the frequency conversion coefficients.

[0034] Time-frequency energy Summing along the time dimension yields the total energy distribution of each frequency throughout the entire signal:

[0035] Where m is time. This is for accumulating energy. The total energy of the signal: Where f is the frequency. The location of the main peak in the frequency energy: Where k0 is the maximum energy, the main peak position is the center of the frequency band, and the left boundary is taken as... (Offline frequency index), right boundary take (Upper frequency index), compare the energy levels of adjacent frequency points to the left and right of the main peak, and extend the boundary towards higher energy frequencies. The resulting frequency region's energy: ,in, The energy in the frequency range. When the energy proportion of the frequency band... : ,in, This represents the total energy of the time-frequency plot. The final adaptive characteristic frequency band is obtained. f 1, f 2] is: The time-frequency diagrams of multimodal signals are uniformly cropped to the characteristic frequency band. To suppress dynamic range and improve robustness, the amplitude or power is logarithmically compressed, and in-band normalization is performed as follows:

[0036] in, Let be the mean of the k-th frequency band in the time dimension. Let be the standard deviation of the k-th frequency band in the time dimension. The scaling factor. The standardized time-frequency diagram is obtained by taking the same time window (opening / closing trigger ±250ms) from both time-frequency diagrams and unifying them to the same dimension F×T: Where A is the isomorphic time-frequency diagram corresponding to the sound signal, and V is the isomorphic time-frequency diagram corresponding to the vibration signal. This embodiment of the invention obtains isomorphic time-frequency diagrams by isomorphically processing the time-frequency diagrams of multimodal signals, thus giving different modal signals a unified representation in the time-frequency domain. This facilitates processing using the same analysis methods and models, simplifies subsequent analysis procedures, and improves processing efficiency.

[0037] 103. Determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency diagram, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block.

[0038] In this embodiment of the invention, the isomorphic time-frequency map is transformed into a Gaussian Markov random field representation that can characterize its internal spatial dependency structure, that is, each time-frequency unit (i.e., pixel) of the isomorphic time-frequency map is transformed into a Gaussian Markov random field representation that can characterize its internal spatial dependency structure. Considering a node as a node in a random field, the entire isomorphic time-frequency diagram constitutes a two-dimensional random field (isomorphic time-frequency matrix) containing N=F×T nodes. An eight-neighborhood system is established for each node, i.e., for each node located at... Location node Its eight-neighborhood contains eight neighboring nodes surrounding the node. The isomorphic time-frequency matrix is ​​then expanded into isomorphic time-frequency vectors in column order. These vectors are divided into multiple vector blocks. For each vector block, a local mask matrix is ​​introduced to mathematically describe the neighborhood relationships. The embodiments of the present invention expand the isomorphic time-frequency matrix into vectors and divide it into vector blocks, and determine the local mask matrix and local precision matrix corresponding to each vector block. This allows the focus to be placed on local regions of the time-frequency graph, avoiding the time spent analyzing redundant regions that are not relevant to fault identification, and also avoiding interference from redundant regions to fault identification. Thus, the present invention can improve the identification efficiency and accuracy of circuit breaker faults.

[0039] 104. Determine the local precision matrix corresponding to each local mask matrix, and based on the local precision matrix, determine the local energy contribution map of the multimodal signal.

[0040] In this embodiment of the invention, the local precision matrix is ​​related to the local mask matrix, reflecting the correlation strength between nodes within the local region. Assuming the nodes within the local region follow a multivariate Gaussian distribution, its precision matrix (i.e., the inverse covariance matrix) can characterize the conditional dependencies between nodes. For a given local mask matrix... The local accuracy matrix can be estimated using maximum likelihood estimation or other parameter estimation methods. Specifically, firstly, time-frequency cell data corresponding to nodes within the local region are collected to form a data vector. Based on the probability density function of the multivariate Gaussian distribution We use this to solve for the local precision matrix, where Q is the local precision matrix. Let be the probability density function. For example, let the likelihood function be... We calculate the gradient of Q with respect to it and set the gradient to zero. We then use an iterative optimization algorithm to gradually adjust the value of Q until the convergence condition is met, thereby obtaining the value of the local precision matrix Q.

[0041] Furthermore, for each node within a local region, its contribution to the local energy is calculated. First, the local precision matrix is ​​decomposed into eigenvector matrices and eigenvalue matrices. The magnitude of the eigenvalues ​​reflects the energy distribution along the corresponding eigenvector direction. For each node, its energy contribution can be calculated based on its correlation with each eigenvector. For example, the data vector of node k within the local region is projected onto each eigenvector to obtain projection coefficients. Then, the sum of squares of these projection coefficients is used to measure node k's contribution to the local energy. Arranging the energy contribution values ​​of all nodes within each local region according to their positions in the time-frequency graph yields the local energy contribution map of the multimodal signal. By observing the local energy contribution map, one can intuitively understand the contribution of each time-frequency unit to the signal energy in different local regions, thus aiding in the analysis of the characteristics and properties of the multimodal signal. Figure 3 The image illustrates one type of local energy contribution map. 105. Perform deep joint feature extraction on the local energy contribution map of the multimodal signal to obtain the fault-sensitive feature vector of the target circuit breaker. Based on the fault-sensitive feature vector, determine the fault identification result of the target circuit breaker.

[0042] In this embodiment of the invention, deep joint feature extraction is performed on the local energy contribution map of multimodal signals to obtain the fault-sensitive feature vector of the target circuit breaker. Finally, the fault type and other identification results of the target circuit breaker are determined based on the fault-sensitive feature vector. This embodiment of the invention performs deep joint feature extraction on the local energy contribution map of multimodal signals, which can uncover deep features hidden in the data, thereby improving the identification accuracy of circuit breakers.

[0043] According to the circuit breaker fault identification method provided by this invention, compared with the current method of identifying circuit breaker faults manually, this invention simultaneously collects multi-modal signals such as acoustic signals and vibration signals during the opening and closing process of the target circuit breaker. Acoustic signals reflect the sound characteristics generated by the movement and friction of internal mechanical components of the circuit breaker, while vibration signals reflect the vibration state of the mechanical structure. Combining these two signals provides more comprehensive information on the circuit breaker's operating status, and compared to single-modal signals, it can more accurately capture fault characteristics, thereby improving the accuracy of circuit breaker fault identification. Furthermore, isomorphic processing is performed on the time-frequency diagrams of the multi-modal signals to obtain isomorphic time-frequency diagrams, ensuring that different modal signals have uniformity in the time-frequency domain. The unified representation format facilitates the use of the same analysis methods and models, simplifying subsequent analysis processes and improving processing efficiency. Expanding the isomorphic time-frequency matrix into vectors and dividing it into vector blocks, and determining the local mask matrix and local precision matrix corresponding to each vector block, allows focusing on local regions of the time-frequency graph. This avoids the time wasted analyzing redundant regions unrelated to fault identification and also avoids interference from redundant regions in fault identification. Thus, this invention can improve the identification efficiency and accuracy of circuit breaker faults. Deep joint feature extraction of the local energy contribution map of multi-mode signals can uncover deep features hidden in the data, thereby improving the identification accuracy of circuit breakers.

[0044] Furthermore, to better illustrate the process of identifying circuit breaker faults described above, and as a refinement and extension of the above embodiments, this invention provides another method for identifying circuit breaker faults, such as... Figure 4 As shown, the method includes: 201. In response to the fault identification signal of the target circuit breaker, the multi-mode signals of the target circuit breaker during the opening and closing process are collected synchronously. The multi-mode signals include acoustic signals and vibration signals.

[0045] Specifically, the sound and vibration signals of the target circuit breaker during the opening and closing process are collected synchronously through sensors and other devices.

[0046] 202. Determine the time-frequency diagram corresponding to the multimodal signal, and perform isomorphic processing on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal.

[0047] Specifically, multimodal signals typically contain multiple different types of signals, such as sound signals and vibration signals. For each mode of signal, methods such as Short Time Fourier Transform (STFT) are used to transform it into the time-frequency domain, thereby obtaining the corresponding time-frequency plot. Then, isomorphic processing is performed on the time-frequency plots of the multimodal signals to obtain the corresponding isomorphic time-frequency plots of the multimodal signals.

[0048] 203. Determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency diagram, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block.

[0049] In this embodiment of the invention, in order to identify faults in circuit breakers, it is first necessary to determine the local mask matrix corresponding to each vector block. Based on this, step 203 specifically includes: determining the eight neighboring nodes corresponding to each node in each vector block, and constructing the local mask matrix corresponding to each vector block based on each node and its corresponding eight neighboring nodes. In this local mask matrix, the matrix elements at the corresponding positions of the nodes in the vector block and their corresponding eight neighboring nodes are 1, and the remaining matrix elements after removing the positions with matrix elements of 1 are 0. The node corresponds to a time-frequency unit in the isomorphic time-frequency diagram.

[0050] Specifically, the isomorphic time-frequency graph is transformed into a Gaussian Markov random field representation that can characterize its internal spatial dependency structure. The isomorphic time-frequency graph S∈R F×T Each time-frequency unit (i.e., pixel S) in i,j Consider each node as a node in a random field. The entire time-frequency diagram constitutes a two-dimensional random field (isomorphic time-frequency matrix) containing N = F × T nodes. An eight-neighborhood system is established for each node. The adjacency mask is obtained. When (i,j) is a neighbor or i=j =1, otherwise =0. Expanding using column-major order is shown below:

[0051] in, It is an isomorphic time-frequency matrix. For the m-th vector block, This is the function for partitioning vector blocks, where N is the matrix dimension. That is, for... Divide into h×w small blocks, with a step size of For block index b, retrieve the vector of all signals within that block. Simultaneously extract the local mask matrix. In this context, the nodes in the vector block and their corresponding eight neighboring nodes have matrix elements of 1 at their corresponding positions in the local mask matrix. The remaining matrix elements after removing the positions with matrix elements of 1 in the local mask matrix are 0. Each node corresponds to a time-frequency unit in the isomorphic time-frequency diagram.

[0052] 204. Determine the local precision matrix corresponding to each local mask matrix, and based on the local precision matrix, determine the local energy contribution map of the multimodal signal.

[0053] For embodiments of the present invention, it is also necessary to determine the local precision matrix corresponding to the local mask matrix, and to determine the local energy contribution map based on the local precision matrix. Therefore, step 204 specifically includes: determining the mean and covariance of the block-level vector elements within each vector block; diagonally shrinking the block-level vector element covariance to obtain the shrunken block-level vector element covariance; using the shrunken block-level vector element covariance as a statistic and the local mask matrix as a constraint to determine the local precision matrix corresponding to each local mask matrix; and determining the channel map vector of the random field channel map of the same size as the isomorphic time-frequency map. And based on the local precision matrix Vector of vector block i Determine the local energy contribution map of the multimodal signal. ,in, .

[0054] Specifically, the mean of the block-level vector elements is determined according to the following formula. and block-level vector element covariance :

[0055]

[0056] Where M is the total number of vector blocks, and m is the vector block identifier. For the channel graph vector, the covariance of the block-level vector elements is calculated according to the following formula. Diagonal contraction is performed to improve positive definiteness and robustness, resulting in the covariance of the contracted block-level vector elements. :

[0057] in, The contraction coefficient. Defined under the Gaussian function:

[0058] in, Let Q be the data vector formed by each video unit in the isomorphic time-frequency graph, and let Q be the initial local precision matrix. Let P(x) be a Gaussian function. Then, using the covariance of the elements of the contraction block-level vector as a statistic, and the local mask matrix as a constraint, the local precision matrix corresponding to each local mask matrix is ​​determined. :

[0059] in, For position The local mask matrix at that location, For position The initial local precision matrix is ​​determined. Finally, based on the local precision matrix, the local energy contribution map of the multimodal signal is determined.

[0060] 205. The local energy contribution maps of the multimodal signals are superimposed to obtain superimposed features. The superimposed features are then input into a preset feature extraction model for feature extraction to obtain the fault-sensitive feature vector of the target circuit breaker.

[0061] For the implementation of this invention, in order to improve the preset, it is first necessary to train and construct the preset. Based on this, the method includes: constructing the preset initial; obtaining a sample dataset, wherein the sample dataset includes the superimposed features of the local energy contribution maps of multimodal signals of sample circuit breakers with fault-sensitive feature labels; dividing the sample dataset into a training set and a test set, using the training set to train the preset initial, and using the test set to test the trained preset initial, and finally using the trained preset initial that meets the test conditions as the preset. Specifically, in the model training process, firstly, a preset initial score prediction model is constructed, and secondly, the sample dataset is obtained. Ensure that the dataset contains all necessary files. Convert the data into a format that the preset initial score prediction model can understand, and finally train and test the model. Specifically, the dataset can be divided first: the sample dataset can be divided into a training set and a test set using random or specific strategies (such as stratified sampling). Then, the model is trained using the training set, and the trained model is tested using the test set to evaluate its performance on unseen data. Calculate and record metrics such as precision and recall on the test set. If the model performance does not meet the requirements, it can return to the training phase for more iterations or adjustments. This yields a preset feature extraction model that meets the requirements.

[0062] Specifically, z-score compression is applied to the local energy contribution map to suppress extrema, resulting in a set of channels for both acoustic and vibration paths. These channels collectively constitute a local energy contribution map obtained through a Gaussian Markov random field, reflecting local dependencies and anomalous clustering characteristics in the time-frequency structure. This map is then used for feature extraction in subsequent multi-kernel convolutional neural networks. Multi-kernel convolutional feature extraction: The obtained local energy contribution maps are superimposed along the channel dimension to form an input tensor of uniform size. To avoid the amplitude of individual channels dominating network learning, z-score normalization is first performed at the channel level. The normalization formula is shown below:

[0063] in, This is a standardized local energy contribution map. For local energy contribution map, The mean of the local energy contribution map. The standard deviation of the local energy contribution map. Enter multi-core parallel convolution: two parallel layers: Conv3×3 and Conv5×5, each connected to ReLU; the two local energy contribution maps are concatenated in the channel dimension, and the concatenated channels are merged into a fixed width (64) by 1×1 convolution, and then 2×2 max pooling is performed to halve the size. Global average pooling is performed on the spatial dimension to scale the vector length to 1, eliminating the overall amplitude difference, thereby obtaining the fault-sensitive feature vector.

[0064] In another embodiment of the present invention, the method for superimposing the local energy contribution map of the multimodal signal includes: determining the energy feature vector corresponding to the local energy contribution map of the multimodal signal; performing feature-level superposition processing on each energy feature vector to obtain a feature superposition vector; performing element-level superposition processing on each energy feature vector to obtain an element superposition vector; performing low-order superposition processing on each energy feature vector to obtain a low-order superposition vector; and performing transformation processing on the feature superposition vector, element superposition vector, and low-order superposition vector to obtain superimposed features. The superimposed features are then input into a preset feature extraction model, and the fault-sensitive feature vector of the target circuit breaker is directly output through the preset feature extraction model. For example, a specific superposition processing method is as follows: if the first energy feature vector is (a1, a2), the second energy feature vector is (b1, b2), the third energy feature vector is (c1, c2), and the fourth energy feature vector is (d1, d2). The specific combination processing method includes: combining different feature vectors by feature dimension, that is, performing a Hadamard product on all elements of the vectors, followed by a convolution transformation with a certain weight w1, to obtain a feature superimposed vector f(w1×(a1×b1×c1×d1,a2×b2×c2×d2)); simultaneously, combining all feature vector data at the point level, that is, performing a Hadamard product on each element (component) of the vectors, assigning different weight values ​​to each product result, and then performing a linear transformation, to obtain an element superimposed vector f(w2×a1×b1×c1×d1,w3×a1×b1×c1×d1,w2×b1×c1×d2)). (a2×b2×c2×d2); In addition, all feature vectors are subjected to basic combination processing, and then weight coefficients are assigned to the combined result. Then, a linear transformation is performed to obtain the low-order superimposed vector f(w4(a1,a2,b1,b2,c1,c2,d1,d2)); Finally, the above feature superimposed vector, component superimposed vector, and basic superimposed vector are combined using a preset transformation function, such as horizontal splicing, to obtain superimposed features. It should be noted that the above examples are only illustrative and do not limit the embodiments of this application. Thus, by combining the first energy feature vector, the second energy feature vector, the third energy feature vector, and the fourth energy feature vector, different features can be automatically or explicitly combined to generate new feature combinations. These combined features may contain complex nonlinear relationships between the original features, enabling the model to capture more refined and richer information in the data. That is, it can make full use of the relationships between various data, extract more latent features, and take into account both high-order and low-order processing, making data utilization more efficient and the subsequent model prediction results more accurate, meeting the needs of practical application scenarios.

[0065] 206. Determine the different clusters corresponding to different fault types, and determine the centroid vectors corresponding to different clusters.

[0066] 207. Determine the distance between the fault-sensitive feature vector and the centroid vectors corresponding to different clusters, and based on the distance, determine the target cluster to which the fault-sensitive feature vector belongs, and take the fault type corresponding to the target cluster as the fault type of the target circuit breaker.

[0067] Specifically, the centroid vectors corresponding to different clusters are determined by joint feature extraction of the local energy contribution maps of the multimodal signals (sound signals and vibration signals) of the sample circuit breaker, resulting in fault-sensitive feature vectors. For example, during the opening and closing operations of the sample circuit breaker, sound signals under normal operating conditions and mechanical fault conditions (such as loose base, stuck core, etc.) are acquired by sound sensors and vibration acceleration sensors, respectively. and vibration signals Fifty groups of each signal were used. For each group of sound and vibration signals, time alignment was achieved through cross-correlation calibration, and denoising based on the signal-to-noise ratio factor was performed to obtain high signal-to-noise ratio time-domain data. Short-time Fourier transforms were performed on the sound and vibration signals respectively to generate isomorphic time-frequency maps with uniform window length, overlap rate, and frequency band range, providing consistent input for subsequent joint analysis. Using the time-frequency map as a two-dimensional random field, local conditional dependencies were established based on the neighborhood structure. The accuracy matrix was calculated using maximum likelihood, and a local energy contribution map was constructed. The sound-vibration local energy contribution maps were then superimposed by channel and input into a multi-kernel convolutional neural network. Deep joint features were extracted through multi-scale convolution, pooling, and activation operations, outputting a sample fault-sensitive feature vector. Then, the sample fault-sensitive feature vectors of each group of acoustic-vibration signals are clustered. The specific clustering method is as follows: initialize the sample centroid vectors corresponding to different clusters; calculate the distance between the sample fault-sensitive feature vectors and the sample centroid vectors corresponding to the different clusters, and based on the distance, divide each group of sample fault-sensitive feature vectors into different clusters; based on the sample fault-sensitive feature vectors in different clusters, obtain the updated sample centroid vectors corresponding to different clusters; based on the updated sample centroid vectors, redivide the sample fault-sensitive feature vectors into different clusters until the updated sample centroid vectors do not change. The sample fault-sensitive feature vectors finally divided into different clusters are determined as sample fault-sensitive feature vectors under different fault types.

[0068] Furthermore, the mean of the sample fault-sensitive feature vectors in different fault types can be used as the centroid vectors corresponding to different clusters. Then, the Euclidean distance between the fault-sensitive feature vectors and the centroid vectors corresponding to different clusters is calculated, and the fault type corresponding to the cluster with the smallest Euclidean distance is taken as the fault type of the target circuit breaker.

[0069] According to another circuit breaker fault identification method provided by the present invention, compared with the current method of identifying circuit breaker faults manually, the present invention simultaneously collects multi-modal signals such as acoustic signals and vibration signals during the opening and closing process of the target circuit breaker. Acoustic signals reflect the sound characteristics generated by the movement and friction of internal mechanical components of the circuit breaker, while vibration signals reflect the vibration state of the mechanical structure. The combination of these two signals provides more comprehensive information on the circuit breaker's operating status, and compared to single-modal signals, it can more accurately capture fault characteristics, thereby improving the accuracy of circuit breaker fault identification. Furthermore, isomorphic processing is performed on the time-frequency diagrams of the multi-modal signals to obtain isomorphic time-frequency diagrams, ensuring that different modal signals have [a certain characteristic] in the time-frequency domain. A unified representation format facilitates the use of the same analysis methods and models, simplifying subsequent analysis processes and improving processing efficiency. Expanding the isomorphic time-frequency matrix into vectors and dividing it into vector blocks, and determining the local mask matrix and local precision matrix corresponding to each vector block, allows focusing on local regions of the time-frequency graph. This avoids the time wasted analyzing redundant regions unrelated to fault identification and also avoids interference from redundant regions in fault identification. Thus, this invention can improve the identification efficiency and accuracy of circuit breaker faults. Deep joint feature extraction of the local energy contribution map of multi-mode signals can uncover deep features hidden in the data, thereby improving the identification accuracy of circuit breakers.

[0070] Furthermore, as Figure 1 In specific implementation, embodiments of the present invention provide a circuit breaker fault identification device, such as... Figure 5 As shown, the device includes: a data acquisition unit 31, a processing unit 32, a first determination unit 33, a second determination unit 34, and a fault identification unit 35.

[0071] The acquisition unit 31 can be used to synchronously acquire the multi-mode signals of the target circuit breaker during the opening and closing process in response to the fault identification signal of the target circuit breaker, wherein the multi-mode signals include acoustic signals and vibration signals.

[0072] The processing unit 32 can be used to determine the time-frequency diagram corresponding to the multimodal signal, and perform isomorphic processing on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal.

[0073] The first determining unit 33 can be used to determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency map, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in matrix column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block.

[0074] The second determining unit 34 can be used to determine the local precision matrix corresponding to each local mask matrix, and based on the local precision matrix, determine the local energy contribution map of the multimodal signal.

[0075] The fault identification unit 35 can be used to perform deep joint feature extraction on the local energy contribution map of the multimodal signal to obtain the fault sensitive feature vector of the target circuit breaker, and determine the fault identification result of the target circuit breaker based on the fault sensitive feature vector.

[0076] In specific application scenarios, in order to perform time alignment of multimodal data, such as Figure 6 As shown, the device also includes an alignment unit 36.

[0077] The alignment unit 36 ​​can be used to acquire multiple sets of sample sound signals and multiple sets of sample vibration signals at different times ti during the opening and closing process of the sample circuit breaker, and determine multiple time delay variables. For each time delay variable The sample sound signal at each time step ti is compared with the corresponding The sample vibration signals at time ti are multiplied together, and the multiplication results at each time ti are summed to obtain each time delay variable. Evaluation parameters ,in, In each evaluation parameter, a maximum evaluation parameter is determined, and the time delay variable corresponding to the maximum evaluation parameter is used as the optimal time delay between the sound signal and the vibration signal. Based on the optimal time delay, the sound signal and the vibration signal in the multimodal signal are time-aligned to obtain the aligned sound signal and the aligned vibration signal.

[0078] In specific application scenarios, in order to perform noise reduction processing on multimodal data, the device further includes a noise reduction unit 37.

[0079] The denoising unit 37 can be used to perform multi-scale wavelet decomposition on the multimodal signal to obtain different levels of noise reduction. Approximation coefficients and detail coefficients at different levels of j Determine the detail coefficients Each coefficient value the median of Based on each coefficient value and the median Determine the noise standard deviation of the multimodal signal. ,in, , The noise figure is used to determine the signal-to-noise ratio of the multimodal signal. Based on the signal-to-noise ratio Determine the signal-to-noise ratio factor of the multimodal signal. ,in, Based on the noise standard deviation and the signal-to-noise ratio factor Determine the detail coefficients contraction threshold ,in, N is the length of the multimodal signal; based on the shrinkage threshold For the detail coefficients Perform shrinkage processing to obtain the detail coefficients after shrinkage. ,in, , For the contraction function; for the approximation coefficients and the detail coefficient after shrinkage Perform inverse wavelet transform to obtain the denoised multimodal signal.

[0080] In specific application scenarios, in order to determine the isomorphic time-frequency diagram corresponding to the multimodal signal, the processing unit 32 includes a segmentation module 321, a superposition module 322, a determination module 323, and a trimming module 324.

[0081] The segmentation module 321 can be used to unify the multimodal signals to the same sampling rate, determine the window length, frame shift, and FFT points for constructing the time-frequency graph, and perform segmentation processing on the multimodal signals based on the window length, the frame shift, and the FFT points to determine the spectral representation of each signal segment. The spectral representations of each signal segment are then spliced ​​together according to the time dimension to obtain the time-frequency graph of the multimodal signals.

[0082] The superposition module 322 can be used to accumulate the energy from the lowest frequency to the current frequency of the time-frequency graph, and use the ratio of the accumulated energy to the total energy corresponding to the time-frequency graph as the cumulative energy percentage of the current frequency.

[0083] The determining module 323 can be used to use the current frequency as the online frequency index when the accumulated energy percentage is greater than a preset percentage threshold. The lowest frequency is used as the lower frequency index. Based on the online frequency index and the offline frequency index Determine the characteristic frequency band of the time-frequency diagram. .

[0084] The cropping module 324 can be used based on the characteristic frequency band. The time-frequency graph is cropped, and the cropped time-frequency graph is used as an isomorphic time-frequency graph.

[0085] In specific application scenarios, in order to determine the local mask matrix corresponding to each vector block, the first determining unit 33 can be used to determine the eight neighboring nodes corresponding to each node in each vector block. Based on each node and its corresponding eight neighboring nodes, a local mask matrix corresponding to each vector block is constructed. In this local mask matrix, the matrix elements at the corresponding positions of the nodes in the vector block and their corresponding eight neighboring nodes are 1, and the remaining matrix elements after removing the positions with matrix elements of 1 in the local mask matrix are 0. The node corresponds to a time-frequency unit in the isomorphic time-frequency graph.

[0086] In specific application scenarios, in order to determine the local precision matrix corresponding to each local mask matrix, the second determining unit 34 can be used to determine the mean and covariance of the block-level vector elements within each vector block; diagonally shrink the block-level vector element covariance to obtain the shrunken block-level vector element covariance; using the shrunken block-level vector element covariance as a statistic and the local mask matrix as a constraint, determine the local precision matrix corresponding to each local mask matrix.

[0087] In specific application scenarios, in order to determine the local energy contribution map of a multimodal signal, the second determining unit 34 can be used to determine the channel map vector of a random field channel map of the same size as the isomorphic time-frequency map. And based on the local precision matrix Vector of vector block i Determine the local energy contribution map of the multimodal signal. ,in, .

[0088] In specific application scenarios, in order to determine the fault-sensitive feature vector of the target circuit breaker, the fault identification unit 35 can be used to superimpose the local energy contribution map of the multi-mode signal to obtain superimposed features, and input the superimposed features into a preset feature extraction model for feature extraction to obtain the fault-sensitive feature vector of the target circuit breaker.

[0089] In specific application scenarios, in order to determine the fault identification result of the target circuit breaker, the fault identification unit 35 can be used to determine different clusters corresponding to different fault types, determine the centroid vectors corresponding to different clusters; determine the distance between the fault-sensitive feature vector and the centroid vectors corresponding to different clusters, and based on the distance, determine the target cluster to which the fault-sensitive feature vector belongs, and take the fault type corresponding to the target cluster as the fault type of the target circuit breaker.

[0090] It should be noted that other corresponding descriptions of the functional modules involved in the circuit breaker fault identification device provided in this embodiment of the invention can be found in [reference]. Figure 1 The corresponding description of the method shown will not be repeated here.

[0091] Based on the above, Figure 1 Accordingly, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: In response to a fault identification signal of a target circuit breaker, synchronously acquire multi-mode signals of the target circuit breaker during the opening and closing process, wherein the multi-mode signals include acoustic signals and vibration signals; determine the time-frequency diagram corresponding to the multi-mode signals, and perform isomorphic processing on the time-frequency diagram of the multi-mode signals to obtain a corresponding isomorphic time-frequency diagram of the multi-mode signals; determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency diagram, and expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block; determine the local precision matrix corresponding to each local mask matrix, and determine the local energy contribution map of the multi-mode signals based on the local precision matrix; perform deep joint feature extraction on the local energy contribution map of the multi-mode signals to obtain the fault-sensitive feature vector of the target circuit breaker, and determine the fault identification result of the target circuit breaker based on the fault-sensitive feature vector.

[0092] Based on the above, Figure 1 The method shown and as Figure 5 The embodiment of the device shown in the invention also provides a physical structure diagram of a computer device, such as... Figure 7As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: in response to a fault identification signal from the target circuit breaker, it synchronously acquires multi-mode signals of the target circuit breaker during the opening and closing process, wherein the multi-mode signals include acoustic signals and vibration signals; it determines the time-frequency diagram corresponding to the multi-mode signals and performs isomorphic processing on the time-frequency diagram of the multi-mode signals to obtain the isomorphic time-frequency diagram of the multi-mode signals. The process involves: determining the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency map; expanding the isomorphic time-frequency matrix into isomorphic time-frequency vectors in column order; dividing the isomorphic time-frequency vectors into multiple vector blocks; determining the local mask matrix corresponding to each vector block; determining the local precision matrix corresponding to each local mask matrix; determining the local energy contribution map of the multimodal signal based on the local precision matrix; performing deep joint feature extraction on the local energy contribution map of the multimodal signal to obtain the fault-sensitive feature vector of the target circuit breaker; and determining the fault identification result of the target circuit breaker based on the fault-sensitive feature vector.

[0093] Through the technical solution of this invention, multi-modal signals such as acoustic and vibration signals are simultaneously acquired during the opening and closing process of the target circuit breaker. Acoustic signals reflect the sound characteristics generated by the movement and friction of internal mechanical components of the circuit breaker, while vibration signals reflect the vibration state of the mechanical structure. Combining these two signals provides more comprehensive information on the circuit breaker's operating status, and compared to single-modal signals, it can more accurately capture fault characteristics, thereby improving the accuracy of circuit breaker fault identification. Furthermore, isomorphic processing is applied to the time-frequency diagrams of the multi-modal signals to obtain isomorphic time-frequency diagrams, ensuring that different modal signals have a unified representation in the time-frequency domain, facilitating the use of the same analysis methods. The method and model simplify the subsequent analysis process and improve processing efficiency. Expanding the isomorphic time-frequency matrix into vectors and dividing it into vector blocks, and determining the local mask matrix and local precision matrix corresponding to each vector block, allows focusing on local regions of the time-frequency graph. This avoids the time wasted analyzing redundant regions unrelated to fault identification and also avoids interference from redundant regions in fault identification. Thus, this invention can improve the identification efficiency and accuracy of circuit breaker faults. Deep joint feature extraction of the local energy contribution map of multi-mode signals can uncover deep features hidden in the data, thereby improving the identification accuracy of circuit breakers.

[0094] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0095] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying circuit breaker faults, characterized in that, include: In response to the fault identification signal of the target circuit breaker, the multi-mode signals of the target circuit breaker during the opening and closing process are synchronously acquired, wherein the multi-mode signals include acoustic signals and vibration signals; The time-frequency diagram corresponding to the multimodal signal is determined, and isomorphic processing is performed on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal. Determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency map, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block. Determine the local precision matrix corresponding to each local mask matrix, and based on the local precision matrix, determine the local energy contribution map of the multimodal signal; Deep joint feature extraction is performed on the local energy contribution map of the multimodal signal to obtain the fault-sensitive feature vector of the target circuit breaker. Based on the fault-sensitive feature vector, the fault identification result of the target circuit breaker is determined.

2. The method according to claim 1, characterized in that, Before determining the time-frequency diagram corresponding to the multimodal signal, the method further includes: Obtain the sample circuit breaker at different times t during the opening and closing process. i Multiple sets of sample sound signals and multiple sets of sample vibration signals were used to determine multiple time delay variables. ; For each time delay variable , each t i The sample sound signal at time and the corresponding Multiply the sample vibration signals at time t, and then multiply each t... i The results of multiplying the time points are summed to obtain each time delay variable. Evaluation parameters ,in, ; In each evaluation parameter, a maximum evaluation parameter is determined, and the time delay variable corresponding to the maximum evaluation parameter is taken as the optimal time delay between the sound signal and the vibration signal. Based on the optimal time delay, the sound signal and the vibration signal in the multimodal signal are time-aligned to obtain the aligned sound signal and the aligned vibration signal.

3. The method according to claim 1, characterized in that, Before determining the time-frequency diagram corresponding to the multimodal signal, the method further includes: The multimodal signal is subjected to multi-scale wavelet decomposition to obtain different levels. Approximation coefficients and detail coefficients at different levels of j ; Determine the detail coefficients Each coefficient value the median of Based on each coefficient value and the median Determine the noise standard deviation of the multimodal signal. ,in, , Noise figure; Determine the signal-to-noise ratio of the multimodal signal. Based on the signal-to-noise ratio Determine the signal-to-noise ratio factor of the multimodal signal. ,in, ; Based on the noise standard deviation and the signal-to-noise ratio factor Determine the detail coefficients contraction threshold ,in, N is the length of the multimodal signal; Based on the shrinkage threshold For the detail coefficients Perform shrinkage processing to obtain the detail coefficients after shrinkage. ,in, , It is a contraction function; For the approximation coefficients and the detail coefficient after shrinkage Perform inverse wavelet transform to obtain the denoised multimodal signal.

4. The method according to claim 1, characterized in that, Determining the time-frequency diagram corresponding to the multimodal signal includes: The multimodal signals are unified to the same sampling rate, and the window length, frame shift, and FFT points for constructing the time-frequency graph are determined. Based on the window length, frame shift, and FFT points, the multimodal signals are segmented to determine the spectral representation of each segment. The spectral representations of each segment are then spliced ​​together according to the time dimension to obtain the time-frequency graph of the multimodal signals. The isomorphic processing of the time-frequency graph of the multimodal signal to obtain the corresponding isomorphic time-frequency graph of the multimodal signal includes... The energy from the lowest frequency to the current frequency in the time-frequency graph is accumulated, and the ratio of the accumulated energy to the total energy corresponding to the time-frequency graph is used as the cumulative energy percentage of the current frequency. When the accumulated energy percentage is greater than a preset percentage threshold, the current frequency is used as the online frequency index. The lowest frequency is used as the lower frequency index. Based on the online frequency index and the offline frequency index Determine the characteristic frequency band of the time-frequency diagram. ; Based on the characteristic frequency band The time-frequency graph is cropped, and the cropped time-frequency graph is used as an isomorphic time-frequency graph.

5. The method according to claim 1, characterized in that, Determining the local mask matrix corresponding to each vector block includes: Determine the eight neighboring nodes corresponding to each node in each vector block. Based on each node and its corresponding eight neighboring nodes, construct a local mask matrix corresponding to each vector block. In the local mask matrix, the matrix elements at the corresponding positions of the nodes in the vector block and their corresponding eight neighboring nodes are 1. The remaining matrix elements after removing the positions with matrix elements of 1 in the local mask matrix are 0. Each node corresponds to a time-frequency unit in the isomorphic time-frequency graph.

6. The method according to claim 1, characterized in that, Determining the local precision matrix corresponding to each local mask matrix includes: Determine the mean and covariance of the block-level vector elements within each vector block; Diagonally shrink the block-level vector element covariance to obtain the shrunken block-level vector element covariance. Using the covariance of the elements of the shrinking block-level vector as a statistic, and the local mask matrix as a constraint, the local precision matrix corresponding to each local mask matrix is ​​determined. Determining the local energy contribution map of the multimodal signal based on the local precision matrix includes: Determine the channel map vector of the random field channel map with the same size as the isomorphic time-frequency map. And based on the local precision matrix Vector of vector block i Determine the local energy contribution map of the multimodal signal. ,in, .

7. The method according to claim 1, characterized in that, The deep joint feature extraction of the local energy contribution map of the multimodal signal to obtain the fault-sensitive feature vector of the target circuit breaker includes: The local energy contribution maps of the multimodal signals are superimposed to obtain superimposed features. The superimposed features are then input into a preset feature extraction model for feature extraction to obtain the fault-sensitive feature vector of the target circuit breaker. The step of determining the fault identification result of the target circuit breaker based on the fault-sensitive feature vector includes: Identify the different clusters corresponding to different fault types, and determine the centroid vectors corresponding to different clusters; Determine the distance between the fault-sensitive feature vector and the centroid vectors corresponding to different clusters, and based on the distance, determine the target cluster to which the fault-sensitive feature vector belongs, and take the fault type corresponding to the target cluster as the fault type of the target circuit breaker.

8. A circuit breaker fault identification device, characterized in that, include: The acquisition unit is used to synchronously acquire the multi-mode signals of the target circuit breaker during the opening and closing process in response to the fault identification signal of the target circuit breaker, wherein the multi-mode signals include acoustic signals and vibration signals; The processing unit is configured to determine the time-frequency diagram corresponding to the multimodal signal, and perform isomorphic processing on the time-frequency diagram of the multimodal signal to obtain the isomorphic time-frequency diagram of the multimodal signal. The first determining unit is used to determine the isomorphic time-frequency matrix corresponding to the isomorphic time-frequency map, expand the isomorphic time-frequency matrix into isomorphic time-frequency vectors in matrix column order, divide the isomorphic time-frequency vectors into multiple vector blocks, and determine the local mask matrix corresponding to each vector block. The second determining unit is used to determine the local precision matrix corresponding to each local mask matrix, and to determine the local energy contribution map of the multimodal signal based on the local precision matrix. The fault identification unit is used to perform deep joint feature extraction on the local energy contribution map of the multimodal signal to obtain the fault sensitive feature vector of the target circuit breaker, and to determine the fault identification result of the target circuit breaker based on the fault sensitive feature vector.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images