Intelligent diagnosis method and system for circuit breaker fault identification

Abstract

The invention discloses an intelligent diagnosis method and system for circuit breaker fault identification, and relates to the related technical field of circuit breaker fault diagnosis, and the method comprises the steps: synchronously collecting the multi-source heterogeneous operation data of a target circuit breaker through a high-precision sensor network; a multi-scale spatio-temporal feature extraction network is used for processing, and circuit breaker joint features are obtained; the combined characteristics of the circuit breaker are input into the bidirectional LSTM-GAN hybrid model; based on the fault sample, driving the target circuit breaker digital twin to obtain target circuit breaker fault data; and executing edge-cloud collaborative diagnosis of the fault data of the target circuit breaker, and generating a three-dimensional diagnosis report of the circuit breaker. The technical problems of low fault diagnosis efficiency and insufficient diagnosis result accuracy of the circuit breaker caused by single circuit breaker fault diagnosis means and low detection sensitivity to complex fault modes in the prior art are solved, and the technical effects of improving the accuracy of the diagnosis result and the fault diagnosis efficiency of the circuit breaker are achieved.

Description

Technical Field

[0001] The present application relates to the technical field related to circuit breaker fault diagnosis, and specifically to an intelligent diagnosis method and system for circuit breaker fault identification. Background Art

[0002] As power systems continue to expand and become increasingly complex, ensuring their safe and stable operation is crucial. As key protection and control devices in power systems, the reliable operation of circuit breakers is directly related to the system's power supply quality and stability. Traditional circuit breaker fault identification methods rely heavily on manual experience and simple monitoring, making them difficult to adapt to the modern power system's demand for fast and accurate equipment diagnosis. Fault diagnosis uses simple electrical parameter monitoring methods, such as monitoring the circuit breaker's opening and closing times and operating current. By comparing these parameters with the equipment's normal operating parameters, a fault can be determined. However, these parameters reflect limited device information and cannot provide a comprehensive and accurate diagnostic basis for complex, multi-factor faults. When a circuit breaker experiences hidden faults such as internal contact wear and insulation aging, these monitoring methods often fail to detect the problem early, leading to further development of the fault and potentially serious power accidents.

[0003] Therefore, in the current related technologies, there are technical problems such as single circuit breaker fault diagnosis means and low sensitivity to complex fault mode detection, which leads to low circuit breaker fault diagnosis efficiency and insufficient accuracy of diagnosis results. Summary of the Invention

[0004] This application provides an intelligent diagnostic method and system for circuit breaker fault identification, which solves the technical problems in the prior art, such as the single means of circuit breaker fault diagnosis and low sensitivity to complex fault mode detection, which in turn leads to low circuit breaker fault diagnosis efficiency and insufficient accuracy of diagnostic results. It achieves the technical effect of improving the accuracy of circuit breaker diagnostic results and fault diagnosis efficiency.

[0005] The present application provides an intelligent diagnostic method for circuit breaker fault identification, which includes: synchronously collecting multi-source heterogeneous operating data of a target circuit breaker through a high-precision sensor network; processing the multi-source heterogeneous operating data using a multi-scale spatiotemporal feature extraction network to obtain circuit breaker joint features; inputting the circuit breaker joint features into a bidirectional LSTM-GAN hybrid model, wherein the bidirectional LSTM network is used to assign feature importance levels and the generative adversarial network (GAN) is used to dynamically generate fault samples; based on the fault samples, driving a target circuit breaker digital twin to obtain target circuit breaker fault data, wherein the target circuit breaker digital twin is configured with a preset fault mode; performing edge-cloud collaborative diagnosis of the target circuit breaker fault data to generate a three-dimensional diagnostic report for the circuit breaker.

[0006] In a possible implementation, the intelligent diagnosis method for circuit breaker fault identification further performs the following processing: performing noise reduction preprocessing on the multi-source heterogeneous operation data to establish circuit breaker noise reduction data; performing multi-branch parallel feature extraction on the circuit breaker noise reduction data, wherein the circuit breaker noise reduction data includes high-frequency current signals, three-dimensional vibration signals, and voiceprint signals; fusing the time series features and frequency domain features output by each branch to obtain high-dimensional fusion features; extracting multi-scale information of the high-dimensional fusion features to generate the circuit breaker joint features, wherein the circuit breaker joint features include circuit breaker time domain features and circuit breaker frequency domain features.

[0007] In a possible implementation, the intelligent diagnostic method for circuit breaker fault identification further performs the following processing: using a deep separable convolutional layer to perform preliminary feature extraction on the high-frequency current signal and output the circuit breaker timing features; extracting the spatial-temporal joint features of the three-dimensional vibration signal through the Inception module, calculating the energy proportion to generate frequency domain features; converting the voiceprint signal into a Mel-spectrogram, extracting the time-frequency-space three-dimensional features, and then performing image feature conversion.

[0008] In a possible implementation, the intelligent diagnosis method for circuit breaker fault identification further performs the following processing: training a generative adversarial network, including training a generator and a discriminator, wherein the generator is used to generate fault samples, and the discriminator is used to determine whether the input sample is a generated fault sample to update the generator parameters; establishing a bidirectional LSTM network to calculate the contribution of each feature and determine the feature importance level.

[0009] In a possible implementation, the intelligent diagnosis method for circuit breaker fault identification further performs the following processing: deploying a lightweight knowledge distillation model on the edge node, obtaining key diagnostic logic based on the feature importance level, and generating circuit breaker edge diagnosis results; constructing a cloud-based dynamic fault knowledge graph, mining associated fault patterns based on a graph attention network, and generating circuit breaker cloud-based diagnosis results; and fusing the circuit breaker edge diagnosis results and the circuit breaker cloud-based diagnosis results to generate a three-dimensional diagnosis report for the circuit breaker.

[0010] In a possible implementation, the intelligent diagnosis method for circuit breaker fault identification further performs the following processing: the circuit breaker three-dimensional diagnosis report includes fault confidence information, failed component stress distribution prediction information and remaining life assessment information.

[0011] In a possible implementation, the intelligent diagnosis method for circuit breaker fault identification further performs the following processing: the multi-source heterogeneous operation data includes current waveform high-frequency sampling data, three-dimensional vibration acceleration signals, and voiceprint feature spectra.

[0012] The present application also provides an intelligent diagnosis system for circuit breaker fault identification, including: a multi-source heterogeneous operation data acquisition module, which is used to synchronously collect multi-source heterogeneous operation data of a target circuit breaker through a high-precision sensor network; a circuit breaker joint feature acquisition module, which is used to process the multi-source heterogeneous operation data using a multi-scale spatiotemporal feature extraction network to obtain circuit breaker joint features; a fault sample generation module, which is used to input the circuit breaker joint features into a bidirectional LSTM-GAN hybrid model, wherein the bidirectional LSTM network is used to assign feature importance levels and the generative adversarial network GAN is used to dynamically generate fault samples; a target circuit breaker fault data acquisition module, which is used to drive a target circuit breaker digital twin based on the fault sample to obtain target circuit breaker fault data, wherein the target circuit breaker digital twin is configured with a preset fault mode; a circuit breaker three-dimensional diagnostic report generation module, which is used to perform edge-cloud collaborative diagnosis of the target circuit breaker fault data and generate a three-dimensional diagnostic report for the circuit breaker.

[0013] The intelligent diagnosis method and system for circuit breaker fault identification proposed in this application aims to synchronously collect multi-source heterogeneous operating data of the target circuit breaker through a high-precision sensor network; process the data using a multi-scale spatiotemporal feature extraction network to obtain the circuit breaker joint features; input the circuit breaker joint features into a bidirectional LSTM-GAN hybrid model; drive the target circuit breaker digital twin based on the fault samples to obtain the target circuit breaker fault data; perform edge-cloud collaborative diagnosis of the target circuit breaker fault data and generate a three-dimensional circuit breaker diagnosis report. This solves the technical problems existing in the prior art of single circuit breaker fault diagnosis means and low sensitivity to complex fault mode detection, which in turn leads to low circuit breaker fault diagnosis efficiency and insufficient diagnostic accuracy, and achieves the technical effect of improving the accuracy of circuit breaker diagnostic results and fault diagnosis efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] Figure 1 A flow chart of an intelligent diagnostic method for circuit breaker fault identification provided in an embodiment of the present application.

[0016] Figure 2 A schematic diagram of the structure of an intelligent diagnosis system for circuit breaker fault identification provided in an embodiment of the present application.

[0017] Description of the accompanying drawings: multi-source heterogeneous operation data acquisition module 10, circuit breaker joint feature acquisition module 20, fault sample generation module 30, target circuit breaker fault data acquisition module 40, circuit breaker three-dimensional diagnosis report generation module 50. DETAILED DESCRIPTION

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0019] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0021] The present application provides an intelligent diagnostic method for circuit breaker fault identification, such as Figure 1 As shown, the method includes:

[0022] Step S100: synchronously collecting multi-source heterogeneous operating data of a target circuit breaker through a high-precision sensor network.

[0023] Step S100 further includes step S110, wherein the multi-source heterogeneous operation data includes current waveform high-frequency sampling data, three-dimensional vibration acceleration signals and voiceprint characteristic spectra.

[0024] Preferably, a high-precision sensor network (high-precision sensors and network communication) is used to obtain multi-source data from multiple dimensions and different properties during the operation of the target circuit breaker to comprehensively and real-time reflect the operating status of the target circuit breaker. The multi-source heterogeneous operating data of the target circuit breaker specifically includes high-frequency sampling data of the current waveform, three-dimensional vibration acceleration signals, and soundprint characteristic spectra. Specifically, current is a key electrical quantity reflecting the operating status of the circuit breaker. During normal operation, the current passing through the circuit breaker has specific waveform and amplitude characteristics. When a circuit breaker fails, such as poor contact or short circuit, the current will change, and its waveform will be distorted and the amplitude will abnormally increase or decrease. High-frequency sampling is to more accurately capture the detailed changes in the current waveform. Generally speaking, the current frequency in the power system is 50Hz or 60Hz, but the transient current generated during the fault may contain high-frequency components. Through high-frequency sampling, these subtle change information can be obtained, which is helpful for analyzing the type and severity of the fault. For example, at the moment of a short circuit fault, the current waveform will rise rapidly, and high-frequency sampling data can accurately record the rapid change process.

[0025] Preferably, the circuit breaker generates mechanical vibrations during opening and closing operations and normal operation. These vibration signals contain rich information about the device status. For example, normal opening and closing operations generate vibrations of specific frequencies and amplitudes. When mechanical components inside the circuit breaker become loose, worn, or stuck, the vibration intensity, frequency distribution, and other characteristics will change. A three-dimensional vibration acceleration signal refers to vibration acceleration data collected in three mutually perpendicular directions (usually the x, y, and z axes). By analyzing the vibration signals in these three directions, we can fully understand the vibration conditions of the circuit breaker and more accurately locate the fault location. For example, if the vibration acceleration in a certain direction increases abnormally, it may indicate that there is a problem with the mechanical components in that direction. The circuit breaker will make sounds during operation, and these sounds can also be used as a basis for judging its working status. During normal operation, the sounds made by the circuit breaker are relatively stable and regular. When a fault occurs, such as arc discharge or partial discharge, abnormal sounds will be generated, and their spectral characteristics will change. The voiceprint characteristic spectrum is the result of spectral analysis of the sound emitted by the circuit breaker. By collecting and processing the sound signal, the sound is decomposed into components of different frequencies and displayed in the form of a spectrum diagram. By analyzing the voiceprint characteristic spectrum, abnormal frequency components in the sound can be identified, and then the presence and type of fault can be determined. For example, partial discharge will generate electromagnetic waves in a specific frequency range. These electromagnetic waves will cause vibrations in the surrounding air and produce characteristic sounds. These characteristics can be captured through the voiceprint characteristic spectrum to detect partial discharge faults.

[0026] Step S200: Process the multi-source heterogeneous operation data using a multi-scale spatiotemporal feature extraction network to obtain circuit breaker joint features.

[0027] Preferably, circuit breaker operating data exhibits different characteristics over different time spans. For example, short-term current fluctuations may reflect momentary load changes or electromagnetic interference, while long-term temperature changes may indicate the long-term operating status and aging trends of the device. By designing multi-scale time windows, such as milliseconds, seconds, minutes, or even hours, data characteristics at different time scales can be captured. For example, at the millisecond scale, the details of the current surge at the moment of a short-circuit fault can be analyzed, while at the minute scale, the cyclical load variations can be observed. Data collected by sensors at different locations on the circuit breaker may have different spatial characteristics. For example, the signals detected by temperature and vibration sensors near the contacts may differ from those detected by sensors located elsewhere on the circuit breaker housing. The multi-scale spatiotemporal feature extraction network processes these data at different spatial locations in a hierarchical manner, analyzing data characteristics from a local to a global perspective. For example, it first analyzes the local thermal image features near the contacts and then combines them with the temperature distribution of the entire circuit breaker to provide a more comprehensive understanding of the device's thermal state.

[0028] Preferably, for example, the signals detected by temperature and vibration sensors near the contacts may differ from sensor data at other locations on the circuit breaker housing. The multi-scale spatiotemporal feature extraction network processes these data at different spatial locations in a hierarchical manner, analyzing data features from a local to global perspective. For example, the local thermal image features near the contacts are first analyzed, and then combined with the temperature distribution of the entire circuit breaker to gain a more comprehensive understanding of the thermal state of the device. The spatial distribution features of the data collected by sensors at different locations are extracted. For three-dimensional vibration acceleration signals, the network analyzes information such as the vibration energy distribution, frequency characteristics, and vibration propagation path along different coordinate axes. For example, by comparing the signal strength and phase difference of vibration sensors at different locations, the source and propagation direction of mechanical vibration can be determined, and the location of potentially faulty components can be inferred. For thermal image data, spatial feature extraction can include information such as temperature distribution unevenness and the location and size of hot spots.

[0029] Preferably, spatial distribution features are extracted from data collected by sensors at different locations. For three-dimensional vibration acceleration signals, network analysis analyzes information such as the vibration energy distribution, frequency characteristics, and vibration propagation path along different coordinate axes. For example, by comparing the signal strength and phase difference of vibration sensors at different locations, the source and propagation direction of mechanical vibration can be determined, and the location of potentially faulty components can be inferred. For thermal image data, spatial feature extraction can include information such as temperature distribution unevenness and the location and size of hot spots.

[0030] Furthermore, step S200 also includes step S210, performing noise reduction preprocessing on the multi-source heterogeneous operation data to establish circuit breaker noise reduction data; step S220, performing multi-branch parallel feature extraction on the circuit breaker noise reduction data, wherein the circuit breaker noise reduction data includes high-frequency current signals, three-dimensional vibration signals, and voiceprint signals; step S230, fusing the time series features and frequency domain features output by each branch to obtain high-dimensional fusion features; step S240, extracting multi-scale information of the high-dimensional fusion features to generate the circuit breaker joint features, wherein the circuit breaker joint features include circuit breaker time domain features and circuit breaker frequency domain features.

[0031] Preferably, the multi-source heterogeneous operation data is pre-processed for noise reduction, that is, various signal processing techniques such as filtering, wavelet denoising and other methods are used to remove or weaken the noise components in the data, improve the quality and reliability of the data, and obtain circuit breaker noise reduction data. For example, high-frequency current signals may contain noise generated by electromagnetic interference. Appropriate filters can effectively filter out this noise, making the signal purer and better reflecting the actual current changes in the circuit breaker. Different feature extraction methods are used to process different types of signals in the circuit breaker noise reduction data, such as high-frequency current signals, three-dimensional vibration signals, and voiceprint signals. This is called multi-branch parallel feature extraction. Specifically, for high-frequency current signals, attention may be paid to features such as current amplitude, frequency, and phase. The time domain signal is converted to the frequency domain through methods such as fast Fourier transform to extract frequency domain features. At the same time, its changes at different times, namely time domain features, are also analyzed. Three-dimensional vibration signals need to consider features such as vibration amplitude, frequency, and acceleration in the directions of the three coordinate axes. By performing spectrum analysis and time-frequency analysis on the vibration signal, feature information that can reflect the mechanical state of the circuit breaker is extracted. Voiceprint signals mainly analyze their audio features, such as the frequency distribution and energy spectrum of the sound. Different circuit breaker states may produce different sound characteristics. By extracting features from the voiceprint signal, these subtle changes can be captured and used to judge the working state of the circuit breaker.

[0032] Preferably, the frequency domain features of the high-frequency current signal (such as harmonic content) are fused with the time domain features of the three-dimensional vibration signal (such as the changing trend of the vibration amplitude) and the frequency domain features of the voiceprint signal (such as the energy distribution in a specific frequency band) to form a high-dimensional fusion feature vector containing multiple dimensional features, which more comprehensively reflects the circuit breaker status information; finally, the high-dimensional fusion features are analyzed and processed from different time scales and spatial scales to generate the circuit breaker joint features, which include the circuit breaker time domain features and the circuit breaker frequency domain features. The time domain features mainly reflect the changes in the circuit breaker operating status over time, such as the amplitude and change rate of signals such as current and vibration at different times; the frequency domain features focus on describing the distribution of signals at different frequency components, such as the harmonic components of current, the frequency characteristics of vibration, the spectral distribution of voiceprints, etc.

[0033] Furthermore, step S220 also includes step S221, using a depthwise separable convolutional layer to perform preliminary feature extraction on the high-frequency current signal and output the circuit breaker timing features; step S222, extracting the spatial-temporal joint features of the three-dimensional vibration signal through the Inception module, calculating the energy proportion to generate frequency domain features; step S223, converting the voiceprint signal into a Mel-spectrogram, extracting the time-frequency-space three-dimensional features, and performing image feature conversion.

[0034] Preferably, a depthwise separable convolution layer is used to process the high-frequency current signal. The depthwise convolution part can capture the local features of the current signal at different time points, such as current mutations, peak values ​​and other information, while the pointwise convolution combines and fuses these local features, and finally outputs the circuit breaker timing characteristics that can reflect the time-varying law of the high-frequency current signal, including the current change trend, periodic characteristics, instantaneous abnormal fluctuations, etc., which helps to analyze the operating status of the circuit breaker from the perspective of time series. Among them, the depthwise separable convolution layer decomposes the traditional convolution operation into depthwise convolution and pointwise convolution. The depthwise convolution performs convolution operations on each input channel independently and only considers feature extraction in the spatial dimension, while the pointwise convolution is used to combine the features obtained by the depthwise convolution to realize information fusion between channels.

[0035] Preferably, the Inception module contains multiple convolution kernels and pooling operations of different sizes, which can simultaneously extract features of different scales. When processing three-dimensional vibration signals, the Inception module can analyze the signal from both spatial and temporal dimensions. For three-dimensional vibration signals, it is necessary not only to consider the distribution of the vibration amplitude at each moment in the three spatial dimensions, but also to consider the change pattern of the vibration signal over time. Through multiple convolution and pooling operations within it, the Inception module can capture the local characteristics of the vibration signal in space (such as the difference in vibration intensity at different locations) and the dynamic change characteristics in time (such as the frequency and phase change of vibration), thereby extracting the joint space-time features. The vibration signal is then subjected to frequency domain analysis, that is, by calculating the energy proportion of different frequency components, the frequency domain features can be obtained. For example, the vibration signal is converted from the time domain to the frequency domain through fast Fourier transform, and then the proportion of energy in each frequency interval to the total energy is calculated to reflect the relative importance of different frequency components in the vibration signal.

[0036] Preferably, the Mel-spectrogram is a method for converting a sound signal into a spectrum representation. Considering the characteristics of the human auditory system, it has different resolutions for sounds of different frequencies. Converting the voiceprint signal into a Mel-spectrogram can better highlight the key features in the sound signal and is closer to the way humans perceive sound. Specifically, the voiceprint signal is framed, and then each frame is fast Fourier transformed to obtain a spectrum, which is then transformed according to the Mel-frequency scale to finally obtain a Mel-spectrogram. Features are extracted from three dimensions: time, frequency, and space. In the time dimension, the change of the voiceprint signal over time is considered, such as the duration of the sound, the start and end time, and the feature changes in different time periods; in the frequency dimension, the features of different frequency components in the Mel-spectrogram are analyzed, such as the frequency distribution range, peak frequency, and frequency change trend; in the spatial dimension, the space is the different feature dimensions or different frequency channels of the sound signal, such as the energy distribution on different frequency channels, the shape characteristics of the spectrum, etc.; it is further converted into image features, for example, using convolution operations and pooling operations in convolutional neural networks to extract and transform the Mel-spectrogram into a more abstract and representative image feature vector.

[0037] In step S300, the circuit breaker joint features are input into a bidirectional LSTM-GAN hybrid model, wherein the bidirectional LSTM network is used to assign feature importance levels, and the generative adversarial network (GAN) is used to dynamically generate fault samples.

[0038] Step S300 further includes step S310, training a generative adversarial network, including training a generator and a discriminator, wherein the generator is used to generate fault samples, and the discriminator is used to determine whether the input sample is a generated fault sample to update the generator parameters; step S320, establishing a bidirectional LSTM network to calculate the contribution of each feature and determine the feature importance level.

[0039] Preferably, by combining bidirectional LSTM and GAN into a hybrid model, on the one hand, bidirectional LSTM is used to perform importance analysis on the joint features of the circuit breaker, and on the other hand, GAN is used to generate more fault samples, thereby improving the accuracy and reliability of circuit breaker fault diagnosis. The bidirectional LSTM is a special recurrent neural network (RNN) architecture composed of a forward LSTM and a backward LSTM. The forward LSTM processes the input sequence in sequence, while the backward LSTM processes the input sequence in reverse order, while using past and future information to better understand the sequence data. Specifically, when the joint features of the circuit breaker are input into the bidirectional LSTM, the features are learned and analyzed. The LSTM memory unit captures the changes in the features at different time steps and assigns importance levels based on the degree of influence of the features on the final result. For example, the bidirectional LSTM assigns a higher importance level to features that can more accurately reflect the fault status of the circuit breaker, such as changes in vibration amplitude at certain specific frequencies or abnormal fluctuations in the current signal; while those features with less impact on fault judgment are assigned a lower level. GAN consists of a generator and a discriminator. The generator's function is to generate fake samples based on random noise, while the discriminator is responsible for distinguishing between real samples and fake samples generated by the generator. The two are continuously optimized through adversarial training. The generator strives to generate more realistic fake samples to deceive the discriminator, while the discriminator strives to improve its recognition ability, eventually reaching a balance. Specifically, the joint features of the circuit breaker are input into the GAN as conditions. The generator dynamically generates new fault samples based on these features and random noise. That is, by learning existing fault features, many possible fault samples are generated to expand the fault sample set for training fault diagnosis models, improving the model's generalization ability and ability to recognize various fault conditions.

[0040] Preferably, the main task of the generator is to generate fault samples that are as realistic as possible based on the input random noise vector. The parameter update of the generator depends on the feedback of the discriminator, that is, the discriminator judges the samples generated by the generator. The goal of the generator is to make it difficult for the discriminator to distinguish between the samples it generates and the real fault samples. Through the back-propagation algorithm, the loss function is calculated according to the judgment result of the discriminator, and then the parameters of the generator are updated so that the generator can generate more realistic fault samples; the discriminator receives two input samples, one is a real fault sample, and the other is a fault sample generated by the generator. The task of the discriminator is to accurately judge whether the input sample is real or generated. It also uses the back-propagation algorithm to update its own parameters to maximize the accuracy of distinguishing between real samples and generated samples. By continuously adjusting the parameters, the discriminator can better identify the false samples generated by the generator, thereby prompting the generator to continuously improve the quality of the generated samples.

[0041] Preferably, the circuit breaker joint features are input as an input sequence into the bidirectional LSTM network. The bidirectional LSTM calculates the hidden state at each time step to understand the impact of each feature on the network output at different times. The contribution of each feature to the final output is calculated through the attention mechanism, that is, a weight is assigned to each feature. The larger the weight, the greater the contribution of the feature to the final result. The features are then sorted according to the contribution of each feature, with features with high contribution ranked first and features with low contribution ranked last. The features are divided into different importance levels, thereby providing richer sample data and more targeted feature information for circuit breaker fault diagnosis, thereby improving the performance of fault diagnosis.

[0042] Step S400: Based on the fault sample, drive the target circuit breaker digital twin to obtain target circuit breaker fault data, wherein the target circuit breaker digital twin is configured with a preset fault mode.

[0043] Preferably, the digital twin is a digital mapping of the target circuit breaker. By integrating multi-physics, multi-scale, and multi-probability simulation processes, it can reflect the state, behavior, and performance of the physical entity in real time. The target circuit breaker digital twin is configured with preset fault modes, where the preset fault modes are modeled and set based on various fault conditions that may occur in the actual operation of the circuit breaker, such as contact wear, insulation aging, and mechanism jamming. The fault sample is input into the target circuit breaker digital twin. The digital twin dynamically simulates the fault condition based on this input information and the simulated circuit breaker physical model and operation mechanism. For example, if the fault sample shows that the current has increased abnormally at a certain moment, the digital twin will simulate the state changes of various components inside the circuit breaker under this abnormal current condition, including the temperature increase of the contact and the change of the electromagnetic force, thereby generating simulated fault data, such as the change of electrical parameters (voltage, current, power, etc.), the abnormal performance of mechanical parameters (contact travel, opening and closing time, vibration acceleration, etc.), and the change of thermal parameters (temperature distribution, temperature rise rate, etc.).

[0044] Step S500 : performing edge-cloud collaborative diagnosis of the target circuit breaker fault data to generate a three-dimensional diagnosis report of the circuit breaker.

[0045] Step S500 further includes step S510, deploying a lightweight knowledge distillation model on the edge node, obtaining key diagnostic logic based on the feature importance level, and generating a circuit breaker edge diagnosis result; step S52, constructing a cloud-based dynamic fault knowledge graph, mining associated fault modes based on a graph attention network, and generating a circuit breaker cloud-based diagnosis result; step S530, fusing the circuit breaker edge diagnosis result and the circuit breaker cloud-based diagnosis result to generate a three-dimensional diagnosis report for the circuit breaker.

[0046] Preferably, edge computing processes data at the edge of the network, close to the data source. For target circuit breaker fault data, the edge device will first perform preliminary processing and analysis on the collected raw data, such as data cleaning and feature extraction, to reduce the amount of data uploaded to the cloud and reduce network bandwidth pressure. At the same time, it can also achieve rapid response to some emergency faults. For example, the edge device monitors the key operating parameters of the circuit breaker in real time and immediately issues an early warning signal when it finds that the parameters are out of the normal range. The cloud can integrate the fault data of multiple circuit breakers and perform big data analysis to explore potential fault modes and patterns. For example, through machine learning algorithms, a large amount of fault data is trained to establish a fault prediction model to predict possible future faults of the circuit breaker. The edge and cloud collaborate with each other, and the edge device can adjust the local processing strategy and parameter settings based on the analysis results and feedback information of the cloud to achieve more optimized fault diagnosis. For example, after the cloud discovers a specific fault mode, it will send relevant information to the edge device. The edge device will focus on the characteristics corresponding to the mode in subsequent data analysis to improve the accuracy of fault detection.

[0047] Preferably, after completing the edge-cloud collaborative diagnosis, it is necessary to integrate the various data and analysis results obtained during the diagnosis process, including the operating status data of the circuit breaker, fault characteristic data, output results of the diagnostic model, etc., and use three-dimensional visualization technology to combine the structural model of the circuit breaker with the diagnostic data to display the internal state of the circuit breaker in an intuitive way. For example, by marking the location of the faulty component on the three-dimensional model, different fault types and severity are represented by different colors and graphics. At the same time, the state changes of the circuit breaker in different operating stages and the process of fault development can be dynamically displayed; and users can observe the internal structure and fault conditions of the circuit breaker from different angles through operations such as rotation and zooming, and have a clearer understanding of the overall picture of the fault; and then generate a three-dimensional diagnostic report to provide operation and maintenance personnel with comprehensive and intuitive information, helping them to make decisions quickly and take effective repair and maintenance measures to ensure the safe and reliable operation of the circuit breaker.

[0048] Preferably, the lightweight model obtained through knowledge distillation can retain the key knowledge required for diagnosis and run efficiently on edge devices. For example, a complex circuit breaker fault diagnosis model based on deep learning is converted into a lightweight model with a small number of parameters and low computational complexity through knowledge distillation, and deployed in edge computing devices (such as smart sensor gateways). Knowledge distillation is a model compression technology that transfers the knowledge of a complex large model (teacher model) to a smaller model (student model). By using feature importance information, we focus on features that play a key role in fault diagnosis, thereby quickly extracting key diagnostic logic. For example, if feature importance analysis finds that current harmonic features and vibration features of specific frequencies are most critical for diagnosing a certain type of circuit breaker fault; finally, according to the key diagnostic logic, the locally collected and preliminarily processed circuit breaker operation data is diagnosed and analyzed, and the edge diagnosis results are output, which may include the preliminary judgment of the fault type and the possibility of the fault occurring.

[0049] Preferably, circuit breaker fault data and related equipment and operating environment information are collected from edge nodes. A dynamic fault knowledge graph is constructed, using various circuit breaker fault types, causes, phenomena, equipment components, and operating parameters as nodes, and causal and correlational relationships as edges. A graph attention network is then used to analyze the dynamic fault knowledge graph, mining correlated fault patterns hidden within large amounts of data and complex relationships. The graph attention network is designed to process graph-structured data. Based on the mined correlated fault patterns and in-depth analysis of a large amount of historical fault data, the cloud generates comprehensive and in-depth diagnostic results. These include not only accurate fault type determination, but also analysis of fault development trends, potential impact on other equipment components, and detailed maintenance recommendations. The edge diagnosis results are then integrated with the cloud-based diagnosis results to generate a more accurate and comprehensive comprehensive diagnostic conclusion, namely a three-dimensional circuit breaker diagnostic report. The fault location and range are annotated on the three-dimensional model, and the severity and type of the fault are indicated using different colors and special effects. This allows power operation and maintenance personnel to fully understand the circuit breaker fault situation and quickly formulate maintenance plans.

[0050] Furthermore, step S530 also includes that the circuit breaker three-dimensional diagnosis report includes fault confidence information, failed component stress distribution prediction information and remaining life assessment information.

[0051] Preferably, the 3D circuit breaker diagnostic report includes circuit breaker fault confidence information, failed component stress distribution prediction information, and remaining life assessment information. Specifically, fault confidence information is a quantitative representation of the degree of certainty regarding the diagnosed circuit breaker fault type and status, reflecting the reliability of the diagnostic result and helping maintenance personnel determine its trustworthiness. For example, a high fault confidence, such as over 90%, indicates that the diagnosis is likely accurate, allowing maintenance personnel to decisively implement appropriate maintenance measures. A lower confidence level, such as 60% to 70%, requires further observation or verification using other testing methods to avoid misdiagnosis, resulting in unnecessary repair costs or excessive equipment downtime. Failed component stress distribution prediction information reflects the magnitude and direction of the forces acting on various parts of the component, allowing them to determine which parts are more susceptible to damage or fatigue. This helps maintenance personnel gain a deeper understanding of the extent and specific location of the fault's impact on equipment components. For example, if stress distribution prediction reveals that the contact connection of a circuit breaker experienced significant stress during a fault, maintenance personnel may be advised to focus on inspecting this area for wear during maintenance. Remaining life assessment information is a prediction of how long a circuit breaker will be able to operate normally in the future. If the remaining life assessment shows that the circuit breaker can continue to operate for a long time and the fault will not deteriorate rapidly, then maintenance can be selected during an appropriate planned maintenance window. If the remaining life is short, it may be necessary to arrange for repair or replacement of the equipment immediately to avoid sudden failures during operation and cause greater losses.

[0052] In the above, refer to Figure 1 The intelligent diagnosis method for circuit breaker fault identification according to the embodiment of the present invention is described in detail. Figure 2 An intelligent diagnosis system for circuit breaker fault identification according to an embodiment of the present invention is described.

[0053] The intelligent diagnosis system for circuit breaker fault identification according to the embodiment of the present invention is used to solve the technical problems existing in the prior art, such as the single means of circuit breaker fault diagnosis and low sensitivity to complex fault mode detection, which leads to low efficiency of circuit breaker fault diagnosis and insufficient accuracy of diagnosis results, thereby achieving the technical effect of improving the accuracy of circuit breaker diagnosis results and fault diagnosis efficiency. Figure 2 As shown, the intelligent diagnosis system for circuit breaker fault identification includes: a multi-source heterogeneous operation data acquisition module 10, a circuit breaker joint feature acquisition module 20, a fault sample generation module 30, a target circuit breaker fault data acquisition module 40, and a circuit breaker three-dimensional diagnosis report generation module 50.

[0054] The multi-source heterogeneous operation data acquisition module 10 is used to synchronously collect multi-source heterogeneous operation data of the target circuit breaker through a high-precision sensor network; the circuit breaker joint feature acquisition module 20 is used to process the multi-source heterogeneous operation data using a multi-scale spatiotemporal feature extraction network to obtain circuit breaker joint features; the fault sample generation module 30 is used to input the circuit breaker joint features into a bidirectional LSTM-GAN hybrid model, wherein the bidirectional LSTM network is used to assign feature importance levels and the generative adversarial network GAN is used to dynamically generate fault samples; the target circuit breaker fault data acquisition module 40 is used to drive the target circuit breaker digital twin based on the fault samples to obtain target circuit breaker fault data, wherein the target circuit breaker digital twin is configured with a preset fault mode; the circuit breaker three-dimensional diagnosis report generation module 50 is used to perform edge-cloud collaborative diagnosis of the target circuit breaker fault data and generate a three-dimensional circuit breaker diagnosis report.

[0055] The specific configuration of the circuit breaker joint feature acquisition module 20 will be described in detail below. The circuit breaker joint feature acquisition module 20 further includes: performing noise reduction preprocessing on the multi-source heterogeneous operation data to establish circuit breaker noise reduction data; performing multi-branch parallel feature extraction on the circuit breaker noise reduction data, where the circuit breaker noise reduction data includes high-frequency current signals, three-dimensional vibration signals, and voiceprint signals; fusing the time series features and frequency domain features output by each branch to obtain high-dimensional fused features; and extracting multi-scale information from the high-dimensional fused features to generate the circuit breaker joint features, where the circuit breaker joint features include circuit breaker time domain features and circuit breaker frequency domain features.

[0056] The specific configuration of circuit breaker joint feature acquisition module 20 will be described in detail below. This module further includes: using a depthwise separable convolutional layer to perform preliminary feature extraction on the high-frequency current signal and output circuit breaker timing features; extracting the spatial-temporal joint features of the three-dimensional vibration signal through the Inception module, calculating the energy contribution to generate frequency domain features; and converting the voiceprint signal into a mel-spectrogram, extracting the time-frequency-space three-dimensional features, and then performing image feature conversion.

[0057] The specific configuration of the fault sample generation module 30 will be described in detail below. The fault sample generation module 30 further includes: training a generative adversarial network, including training a generator and a discriminator, wherein the generator is used to generate fault samples, and the discriminator is used to determine whether the input sample is a generated fault sample to update the generator parameters; establishing a bidirectional LSTM network to calculate the contribution of each feature and determine the feature importance level.

[0058] The specific configuration of the circuit breaker 3D diagnostic report generation module 50 is described in detail below. The module further includes: deploying a lightweight knowledge distillation model at the edge node to extract key diagnostic logic based on the feature importance levels and generate circuit breaker edge diagnostic results; constructing a cloud-based dynamic fault knowledge graph, mining associated fault patterns based on a graph attention network, and generating circuit breaker cloud-based diagnostic results; and fusing the circuit breaker edge diagnostic results with the circuit breaker cloud-based diagnostic results to generate the circuit breaker 3D diagnostic report.

[0059] The following will further describe the specific configuration of the circuit breaker 3D diagnostic report generation module 50. The circuit breaker 3D diagnostic report generation module 50 further includes: the circuit breaker 3D diagnostic report includes fault confidence information, failed component stress distribution prediction information and remaining life assessment information.

[0060] The following describes in detail the specific configuration of the multi-source heterogeneous operation data acquisition module 10. The multi-source heterogeneous operation data acquisition module 10 further includes: the multi-source heterogeneous operation data includes current waveform high-frequency sampling data, three-dimensional vibration acceleration signal and voiceprint feature spectrum.

[0061] The intelligent diagnosis system for circuit breaker fault identification provided by the embodiment of the present invention can execute the intelligent diagnosis method for circuit breaker fault identification provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0062] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0063] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. An intelligent diagnostic method for circuit breaker fault identification, characterized in that: The method comprises: Synchronously collect multi-source heterogeneous operating data of target circuit breakers through a high-precision sensor network; Processing the multi-source heterogeneous operation data using a multi-scale spatiotemporal feature extraction network to obtain circuit breaker joint features; Inputting the circuit breaker joint features into a bidirectional LSTM-GAN hybrid model, wherein the bidirectional LSTM network is used to assign feature importance levels and the generative adversarial network (GAN) is used to dynamically generate fault samples; Based on the fault sample, driving the target circuit breaker digital twin to obtain target circuit breaker fault data, wherein the target circuit breaker digital twin is configured with a preset fault mode; Perform edge-cloud collaborative diagnosis of the target circuit breaker fault data to generate a three-dimensional diagnosis report for the circuit breaker.

2. The intelligent diagnosis method for circuit breaker fault identification according to claim 1, characterized in that: The multi-source heterogeneous operation data is processed using a multi-scale spatiotemporal feature extraction network, including: Performing noise reduction preprocessing on the multi-source heterogeneous operation data to establish circuit breaker noise reduction data; performing multi-branch parallel feature extraction on the circuit breaker noise reduction data, wherein the circuit breaker noise reduction data includes a high-frequency current signal, a three-dimensional vibration signal, and a voiceprint signal; Fuse the time series features and frequency domain features output by each branch to obtain high-dimensional fusion features; Multi-scale information of the high-dimensional fusion feature is extracted to generate the circuit breaker joint feature, wherein the circuit breaker joint feature includes the circuit breaker time domain feature and the circuit breaker frequency domain feature.

3. The intelligent diagnosis method for circuit breaker fault identification according to claim 2, characterized in that: Performing multi-branch parallel feature extraction on the circuit breaker noise reduction data, including: A depthwise separable convolutional layer is used to perform preliminary feature extraction on the high-frequency current signal and output the circuit breaker timing features; The Inception module extracts the spatial-temporal joint features of the three-dimensional vibration signal and calculates the energy proportion to generate frequency domain features. The voiceprint signal is converted into a Mel-spectrogram, and the time-frequency-space three-dimensional features are extracted and then converted into image features.

4. The intelligent diagnosis method for circuit breaker fault identification according to claim 1, characterized in that: Build a bidirectional LSTM-GAN hybrid model, including: Training a generative adversarial network includes training a generator and a discriminator, wherein the generator is used to generate fault samples, and the discriminator is used to determine whether the input sample is a generated fault sample to update the generator parameters; A bidirectional LSTM network is established to calculate the contribution of each feature and determine the feature importance level.

5. The intelligent diagnosis method for circuit breaker fault identification according to claim 1, characterized in that: The edge-cloud collaborative diagnosis of target circuit breaker fault data is performed to generate a three-dimensional circuit breaker diagnosis report, including: Deploy a lightweight knowledge distillation model on the edge node, obtain key diagnostic logic based on the feature importance level, and generate circuit breaker edge diagnosis results; Build a cloud-based dynamic fault knowledge graph, mine related fault patterns based on a graph attention network, and generate cloud-based diagnosis results for circuit breakers; The circuit breaker edge diagnosis result and the circuit breaker cloud diagnosis result are integrated to generate the circuit breaker three-dimensional diagnosis report.

6. The intelligent diagnosis method for circuit breaker fault identification according to claim 5, characterized in that: The circuit breaker three-dimensional diagnosis report includes fault confidence information, failed component stress distribution prediction information and remaining life assessment information.

7. The intelligent diagnosis method for circuit breaker fault identification according to claim 1, characterized in that: The multi-source heterogeneous operation data includes current waveform high-frequency sampling data, three-dimensional vibration acceleration signals and voiceprint characteristic spectra.

8. Intelligent diagnostic system for circuit breaker fault identification, characterized by: The system is used to implement the intelligent diagnosis method for circuit breaker fault identification according to any one of claims 1 to 7, and the system includes: Multi-source heterogeneous operation data acquisition module, used to synchronously collect multi-source heterogeneous operation data of the target circuit breaker through a high-precision sensor network; a circuit breaker joint feature acquisition module, configured to process the multi-source heterogeneous operation data using a multi-scale spatiotemporal feature extraction network to obtain circuit breaker joint features; a fault sample generation module, configured to input the circuit breaker joint features into a bidirectional LSTM-GAN hybrid model, wherein the bidirectional LSTM network is used to assign feature importance levels, and the generative adversarial network (GAN) is used to dynamically generate fault samples; a target circuit breaker fault data acquisition module, configured to drive a target circuit breaker digital twin based on the fault sample to obtain target circuit breaker fault data, wherein the target circuit breaker digital twin is configured with a preset fault mode; The circuit breaker three-dimensional diagnostic report generation module is used to perform edge-cloud collaborative diagnosis of the target circuit breaker fault data and generate a circuit breaker three-dimensional diagnostic report.

Images