A method, medium, and system for reliability diagnosis of a circuit breaker device

Abstract

The application provides a reliability diagnosis method, medium and system of a circuit breaker device, belongs to the switch safety identification technical field of the circuit breaker device, and through continuous acquisition of audio signals and vibration signals in the operation process of the circuit breaker, clean audio and vibration signals are obtained through short-time Fourier transform, adaptive median filtering, wavelet denoising and envelope analysis and the like processing. Then, the two types of signals are time-aligned and fused to generate multiple groups of fusion signals. Feature extraction is performed on the fusion signals, and a genetic algorithm is used to optimize the feature weight, and the optimal fusion feature is calculated. Finally, the optimal fusion feature is input into a pre-trained fault diagnosis model, and a fault diagnosis vector of the circuit breaker is output. The application solves the technical problem that the single signal source diagnosis method in the prior art is difficult to comprehensively reflect the operation state of the circuit breaker, is easily affected by environmental noise and interference, and leads to unstable diagnosis results.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of switch safety identification of circuit breaker equipment, and particularly relates to a reliability diagnosis method, medium and system of circuit breaker equipment. BACKGROUND

[0002] The circuit breaker is a key device in the power system, and its reliability directly affects the safe and stable operation of the power grid. With the rapid development of smart grid, higher requirements are put forward for the reliability and maintenance efficiency of the circuit breaker. The traditional maintenance method of the circuit breaker mainly relies on periodic maintenance and manual inspection, and this method has problems such as high maintenance cost, low efficiency, and inability to timely find potential faults.

[0003] In recent years, with the progress of sensor technology, signal processing technology and artificial intelligence technology, the circuit breaker fault diagnosis method based on state monitoring has been widely studied and applied. The existing circuit breaker fault diagnosis methods mainly include single signal source diagnosis methods based on vibration signal analysis, sound signal analysis, current signal analysis, etc. These methods improve the accuracy and real-time performance of circuit breaker fault diagnosis to a certain extent, but the single signal source diagnosis method is difficult to comprehensively reflect the operating state of the circuit breaker, is easily affected by environmental noise and interference, and leads to unstable safety identification results. SUMMARY

[0004] Therefore, the present application provides a reliability diagnosis method, medium and system of circuit breaker equipment, which can solve the technical problems that the single signal source diagnosis method in the prior art is difficult to comprehensively reflect the operating state of the circuit breaker, is easily affected by environmental noise and interference, and leads to unstable safety identification results.

[0005] The present application is implemented as follows:

[0006] The first aspect of the present application provides a reliability diagnosis method of circuit breaker equipment, which comprises the following steps:

[0007] S10, continuously acquiring audio signals and vibration signals in the operation process of the circuit breaker;

[0008] S20, performing short-time Fourier transform and adaptive median filtering on the audio signals to obtain first audio signals, and performing wavelet denoising and envelope analysis on the vibration signals to obtain first vibration signals;

[0009] S30, fusing the audio signals and vibration signals aligned in time using a preset fusion equation set to obtain a fusion signal, the fusion equation set comprising a weighted average fusion equation, an adaptive filtering fusion equation, a wavelet coefficient fusion equation, a cross-correlation fusion equation, an energy ratio fusion equation, and a principal component analysis fusion equation;

[0010] S40, adjusting parameters of the fusion equation set to obtain a plurality of groups of fusion signals;

[0011] S50, respectively extracting features of the plurality of groups of fusion signals to obtain a plurality of groups of fusion features;

[0012] S60, optimizing weights of the plurality of groups of fusion features by using a genetic algorithm, and calculating optimal fusion features based on the plurality of groups of fusion features according to a minimum mean square error criterion;

[0013] S70, inputting the optimal fusion features into a pre-trained mechanical fault diagnosis model of the circuit breaker to obtain a fault diagnosis result of the circuit breaker;

[0014] S80, comparing the fault diagnosis result with a preset safety threshold, determining that the circuit breaker has a potential fault risk if the fault diagnosis result exceeds the safety threshold, determining that the circuit breaker is in a normal working state if the fault diagnosis result does not exceed the safety threshold, and outputting a reliability of the circuit breaker based on a comparison result of the fault diagnosis result and the safety threshold.

[0015] Each equation in the fusion equation set is specifically represented as follows:

[0016] 1. Weighted average fusion equation:

[0017] F(t) = w a A(t) + w v V(t) ;

[0018] In the formula, F(t) is a fused signal; A(t) is an audio signal; V(t) is a vibration signal; w a and w v are weights of the audio and vibration signals respectively, and satisfy w a +w v = 1; the weights can be determined by analyzing signal-to-noise ratios or energies of the signals, or can be optimized by cross-validation and the like.

[0019] 2. Adaptive filter fusion equation:

[0020]

[0021] In the formula, F(n) is a fused discrete-time signal; h ξ (n) is a coefficient of an adaptive filter; M is a filter order; w a and w v are weight coefficients; the filter coefficient can be updated by using an LMS or RLS adaptive algorithm:

[0022] h ξ (n+1) = h ξ(n) + μe(n)x(n - ξ);

[0023] where μ is the step parameter, e(n) is the error signal, and x(n) is the input signal.

[0024] 3. Wavelet coefficient fusion equation:

[0025] F = W -1 (φ(W A (A),W V (V)));

[0026] where W A and W V represent the wavelet transform of the audio and vibration signals, respectively; W -1 represents the inverse wavelet transform; and φ is the fusion function of the wavelet coefficients, which can be selected as the maximum value or weighted average:

[0027]

[0028] or

[0029] φ(c A ,c V ) = w A c A + w V c V ;

[0030] The wavelet basis function and the number of decomposition layers are adjustable parameters.

[0031] 4. Cross-correlation fusion equation:

[0032]

[0033] or in discrete form:

[0034]

[0035] where N is the cross-correlation window size, which is an adjustable parameter.

[0036] 5. Energy ratio fusion equation:

[0037]

[0038] where E A and E V are the energies of the audio and vibration signals, respectively:

[0039]

[0040] Different time windows can be selected to calculate the local energy ratio.

[0041] 6. Principal component analysis fusion equation:

[0042]

[0043] In the formula, [A,V] is a matrix combining audio and vibration signals; U k ,Σ k V k The matrix is ​​obtained by taking the first k principal components after the singular value decomposition of [A,V]; k is the number of principal components, which can be used as an adjustment parameter.

[0044] Furthermore, the method for adjusting the parameters of the fusion equation system specifically includes:

[0045] 1. Define the parameter vector:

[0046] Θ=[θ1,θ2,...,θ N ]

[0047] Where θ i This represents the i-th parameter that needs to be adjusted, and N is the total number of parameters.

[0048] 2. Define the objective function:

[0049]

[0050] Where L is the loss function, y i For real labels, The prediction result is obtained using parameter Θ, where M is the number of samples.

[0051] 3. Parameter optimization process:

[0052]

[0053] in,

[0054] Where Θ * For the optimal parameter combination, and These are the lower and upper bounds of the i-th parameter, respectively.

[0055] 4. Iterative optimization algorithm (taking gradient descent as an example):

[0056]

[0057] Where t is the number of iterations and α is the learning rate.

[0058] 5. Cross-validation:

[0059]

[0060] Where K is the number of folds in the cross-validation, J k(Θ) is the objective function value on the validation set of the k-th fold.

[0061] 6. Grid search:

[0062]

[0063] where is the set of candidate values for the i-th parameter.

[0064] 7. Optionally, random search:

[0065]

[0066] where denotes a uniform distribution over the interval

[0067] 8. Optionally, Bayesian optimization:

[0068]

[0069] where EI is the expected improvement function, is the observed data set after t iterations.

[0070] 9. Optionally, genetic algorithm:

[0071] Θ (t:1) = GA(Θ (t) , f, p c , p m )

[0072] where GA denotes the genetic algorithm operation, f is the fitness function, p c is the crossover probability, and p m is the mutation probability.

[0073] 10. Optionally, particle swarm optimization:

[0074]

[0075] where v i is the particle velocity, w is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers, p i is the particle's best position, and g is the global best position.

[0076] Different parameters of the fusion equation are finally obtained.

[0077] Further, the training step of the circuit breaker mechanical fault diagnosis model is specifically:

[0078] A training data set is established, specifically, a plurality of groups of optimal fusion features of circuit breakers with known mechanical faults are obtained according to the steps S10-S60.​

[0079] Model training, specifically, a multi-layer perceptron network is established, and the training data set is used for training, wherein the input of the training is the optimal fusion features of each group of known mechanical fault circuit breakers, and the output of the training is the corresponding known mechanical fault, and a circuit breaker mechanical fault diagnosis model is obtained.

[0080] The plurality of groups is generally more than 32 groups, and the known mechanical faults include various types of mechanical faults.

[0081] Specifically, the step S10 comprises:

[0082] Step 101, high-sensitivity audio sensors and vibration sensors are installed at key positions of the circuit breaker, the audio sensors are selected to be capacitive microphones with a wide frequency response range, and the vibration sensors are selected to be three-axis piezoelectric acceleration sensors;

[0083] Step 102, connect the signals of the audio sensor and the vibration sensor to the data acquisition card;

[0084] Step 103, real-time acquisition and storage of audio signals and vibration signals through customized data acquisition software;

[0085] Step 104, time stamping of the collected original audio signals and original vibration signals.

[0086] The step S20 specifically comprises:

[0087] Step 201, short-time Fourier transform is performed on the audio signal to obtain a time-frequency representation of the audio signal;

[0088] Step 202, adaptive median filtering is performed on the time-frequency representation of the audio signal to remove impulse noise;

[0089] Step 203, wavelet denoising method is used to perform noise reduction processing on the vibration signal;

[0090] Step 204, Hilbert transform is performed on the denoised vibration signal to obtain an analytic signal;

[0091] Step 205, the modulus of the analytic signal is calculated to obtain the envelope of the vibration signal;

[0092] Step 206, smoothing processing is performed on the envelope of the vibration signal using a moving average filter.

[0093] The step S30 specifically comprises:

[0094] Step 301, time alignment of the first audio signal and the first vibration signal is performed, and a cross-correlation method is used to determine the time delay of the two signals;

[0095] Step 302, weighted average fusion of time-aligned signals;

[0096] Step 303, adaptive filter fusion of time-aligned signals;

[0097] Step 304, wavelet coefficient fusion of time-aligned signals;

[0098] Step 305, cross-correlation fusion of time-aligned signals;

[0099] Step 306, energy ratio fusion of time-aligned signals;

[0100] Step 307, principal component analysis fusion of time-aligned signals.

[0101] The step S40 specifically comprises:

[0102] Step 401, define a parameter vector including all parameters to be adjusted;

[0103] Step 402, define a target function using a mean square error loss function;

[0104] Step 403, adjust the weight parameters of the weighted average fusion equation;

[0105] Step 404, adjust the filter order and step size parameters of the adaptive filter fusion equation;

[0106] Step 405, adjust the wavelet basis and decomposition level of the wavelet coefficient fusion equation;

[0107] Step 406, adjust the window size of the cross-correlation fusion equation;

[0108] Step 407, adjust the local energy calculation window size of the energy ratio fusion equation;

[0109] Step 408, adjust the number of principal components of the principal component analysis fusion equation;

[0110] Step 409, use a combination of grid search and random search to optimize parameters;

[0111] Step 410, use cross-validation to evaluate parameter performance and select multiple different parameter combinations for each fusion method.

[0112] The step S50 specifically comprises:

[0113] Step 501, extract time domain features including mean, variance, skewness, kurtosis, peak factor and margin factor from multiple sets of fused signals respectively;

[0114] Step 502, extracting frequency domain features of the multiple groups of fusion signals respectively, including performing fast Fourier transform, calculating power spectral density, extracting frequency spectrum center, extracting frequency spectrum variance and extracting frequency band energy ratio;

[0115] Step 503, extracting time-frequency domain features of the multiple groups of fusion signals respectively, including performing continuous wavelet transform, calculating wavelet energy entropy and extracting wavelet coefficient statistical features;

[0116] Step 504, combining the extracted time domain features, frequency domain features and time-frequency domain features to form a feature vector;

[0117] Step 505, performing principal component analysis on the feature vector for dimension reduction processing.

[0118] The step S60 specifically comprises:

[0119] Step 601, defining a chromosome coding mode, each chromosome representing a group of feature weights;

[0120] Step 602, initializing a population and randomly generating multiple chromosomes;

[0121] Step 603, calculating the fitness of each chromosome, using weighted features for fault diagnosis, and calculating the diagnosis accuracy as the fitness;

[0122] Step 604, performing a selection operation, using a roulette wheel selection method;

[0123] Step 605, performing a crossover operation, using arithmetic crossover;

[0124] Step 606, performing a mutation operation, using Gaussian mutation;

[0125] Step 607, performing elite preservation, preserving the best individual in the current population to the next generation;

[0126] Step 608, repeating steps 603 to 607 until the maximum number of iterations is reached or the optimal fitness of consecutive generations is not significantly improved;

[0127] Step 609, after obtaining the optimal weight, calculating the optimal fusion feature;

[0128] Step 610, using the least mean square error criterion, using gradient descent method to fine-tune the optimal fusion feature.

[0129] The step S70 specifically comprises:

[0130] Step 701, constructing a multi-layer perception network as a fault diagnosis model, including an input layer, multiple fully connected layers, multiple Dropout layers and an output layer;

[0131] Step 702, using the historical data set to train the fault diagnosis model, using cross-entropy loss function and Adam optimizer;

[0132] Step 703, using early stopping strategy and learning rate decay strategy in the training process;

[0133] Step 704, using the test set to evaluate the model performance, calculating the accuracy, precision, recall and F1 score and other indicators;

[0134] Step 705, input the optimal fusion feature into the trained model to get the fault diagnosis vector;

[0135] Step 706, select the type with the highest probability in the fault diagnosis vector as the final diagnosis result.

[0136] Optionally, the step S80 specifically comprises:

[0137] comparing the fault diagnosis result with the preset safety threshold, if the fault diagnosis result exceeds the safety threshold, it is determined that the circuit breaker has potential fault risk; if the fault diagnosis result does not exceed the safety threshold, it is determined that the circuit breaker is in normal working state; based on the comparison result of the fault diagnosis result and the safety threshold, the reliability of the circuit breaker is calculated and output; wherein the formula for calculating the reliability is: R=e ；λt ;

[0138] Wherein:

[0139] R is the reliability of the circuit breaker;

[0140] λ is the failure rate, the calculation formula is:

[0141] λ0 is the reference failure rate, which can be determined according to historical data or empirical value;

[0142] t is the running time.

[0143] Optionally, the grid search method in the step 409 specifically comprises:

[0144] Step 4091, setting a search range and step length for each parameter to be adjusted;

[0145] Step 4092, generating all possible parameter combinations in the set parameter space;

[0146] Step 4093, for each parameter combination, using the training data to evaluate the model performance;

[0147] Step 4094, selecting the best parameter combination as the optimal parameter;

[0148] The random search method specifically comprises the following steps.

[0149] Step 4095, setting a distribution range for each parameter that needs to be adjusted;

[0150] Step 4096, randomly sampling multiple sets of parameter combinations from the set distribution;

[0151] Step 4097, evaluating the model performance using the training data for each randomly generated parameter combination;

[0152] Step 4098, selecting the parameter combination with the best performance as the optimal parameter.

[0153] Optionally, the process of fine-tuning the optimal fusion feature using the gradient descent method in step 610 specifically comprises the following steps.

[0154] Step 6101, calculating the gradient of the objective function with respect to the optimal fusion feature;

[0155] Step 6102, updating the optimal fusion feature according to the gradient and the learning rate;

[0156] Step 6103, repeating step 6101 and step 6102 until the stopping condition is met;

[0157] The stopping condition comprises at least one of the following: reaching the maximum number of iterations, the change in the value of the objective function being less than a preset threshold, and the gradient norm being less than a preset threshold.

[0158] Optionally, the process of outputting the diagnosis result in step 707 specifically comprises the following steps.

[0159] Step 7071, setting a probability threshold, and when the highest probability is lower than the probability threshold, giving a diagnosis result of suspected failure;

[0160] Step 7072, setting an alert threshold, and when the probability of some key failure types exceeds the alert threshold, giving a warning message in the diagnosis result;

[0161] Step 7073, generating corresponding maintenance suggestions or further inspection suggestions according to the diagnosis result;

[0162] Step 7074, integrating the main diagnosis result, the diagnosis confidence, the secondary diagnosis result, and the suggestions to form a complete diagnosis report.

[0163] The second aspect of the application provides a computer readable storage medium, wherein the computer readable storage medium stores program instructions, and the program instructions are used to execute the reliability diagnosis method of the circuit breaker device when running.

[0164] The third aspect of the present application provides a switch safety identification system of a circuit breaker device, wherein the computer readable storage medium is as described above.

[0165] Compared with the prior art, the circuit breaker device reliability diagnosis method, medium and system provided by the present application have the following advantages:

[0166] 1. Multi-signal source fusion: The present application simultaneously collects audio signals and vibration signals during the operation of the circuit breaker, and fully utilizes the complementary information between different signal sources through various signal fusion methods, thereby improving the expression ability and noise resistance of the fault features. Compared with the diagnosis method of a single signal source, the comprehensiveness and reliability of the diagnosis are significantly improved.

[0167] 2. Adaptive signal processing: The present application uses short-time Fourier transform, adaptive median filtering, wavelet denoising and other methods to preprocess the original signals, thereby effectively improving the signal-to-noise ratio and laying a good foundation for subsequent feature extraction.

[0168] 3. Diversified feature extraction: The present application extracts features from three aspects of time domain, frequency domain and time-frequency domain, and comprehensively describes the operating state of the circuit breaker. Dimensionality reduction is performed through principal component analysis, thereby improving the expression efficiency of the features.

[0169] 4. Feature fusion optimization: The present application uses a genetic algorithm to optimize the weights of multiple groups of fused features, and fine-tunes based on the least mean square error criterion to obtain optimal fused features. This method significantly improves the discriminant ability of the features and the performance of the diagnosis model.

[0170] 5. High-performance diagnosis model: The present application uses a deep learning model for fault diagnosis, which has stronger feature learning ability and pattern recognition ability than traditional machine learning methods, and can adapt to complex fault patterns and working condition changes.

[0171] 6. Reliability evaluation: The present application introduces a probability threshold and an alarm threshold to evaluate the reliability of the diagnosis results and provide corresponding warning information and suggestions, thereby providing strong support for maintenance decisions.

[0172] In summary, the circuit breaker device reliability diagnosis method of the present application has made significant progress in terms of diagnosis accuracy, reliability, adaptability and real-time performance, and has solved the technical problems of the prior art, such as the difficulty of a single signal source diagnosis method to fully reflect the operating state of the circuit breaker, the susceptibility to environmental noise and interference, and the instability of the safety identification results. BRIEF DESCRIPTION OF DRAWINGS

[0173] Figure 1 The flowchart of the method provided by the present application is shown. DETAILED DESCRIPTION

[0174] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0175] As Figure 1 shown, it is a reliability diagnosis method flow chart of a circuit breaker device provided by the present application, and the method comprises the following steps:

[0176] S10, continuously acquiring audio signals and vibration signals in a circuit breaker operation process;

[0177] S20, performing short-time Fourier transform and adaptive median filtering on the audio signals to obtain first audio signals, and performing wavelet denoising and envelope analysis on the vibration signals to obtain first vibration signals;

[0178] S30, using a preset fusion equation set to fuse the audio signals and the vibration signals aligned according to time to obtain a fusion signal, wherein the fusion equation set comprises a weighted average fusion equation, an adaptive filtering fusion equation, a wavelet coefficient fusion equation, a cross-correlation fusion equation, an energy ratio fusion equation and a principal component analysis fusion equation;

[0179] S40, adjusting parameters of the fusion equation set to obtain a plurality of groups of fusion signals;

[0180] S50, respectively extracting features from the plurality of groups of fusion signals to obtain a plurality of groups of fusion features;

[0181] S60, using a genetic algorithm to optimize weights of the plurality of groups of fusion features, and using the plurality of groups of fusion features to calculate an optimal fusion feature based on a minimum mean square error criterion;

[0182] S70, inputting the optimal fusion feature into a pre-trained circuit breaker mechanical fault diagnosis model to obtain a fault diagnosis result of the circuit breaker;

[0183] S80, comparing the fault diagnosis result with a preset safety threshold value, if the fault diagnosis result exceeds the safety threshold value, determining that the circuit breaker has a potential fault risk, if the fault diagnosis result does not exceed the safety threshold value, determining that the circuit breaker is in a normal working state, calculating a reliability of the circuit breaker based on a comparison result of the fault diagnosis result and the safety threshold value, and outputting; wherein a formula for calculating the reliability is: R = e ；λt ;

[0184] Wherein:

[0185] R is the reliability of the circuit breaker;

[0186] λ is a failure rate, and a calculation formula is:

[0187] λ0 is a reference failure rate, which can be determined according to historical data or an empirical value;

[0188] t is the running time.

[0189] The specific implementation of the above steps is described in detail as follows:

[0190] The specific implementation of step S10 is as follows: first, high-sensitivity audio sensors and vibration sensors are installed at key positions of the circuit breaker. The audio sensors are capacitive microphones with a wide frequency response range (20 Hz-20 kHz), and the sampling rate is set to 44.1 kHz to capture the sound signals during the operation of the circuit breaker. The vibration sensors are three-axis piezoelectric acceleration sensors with a measurement range of ±100 g and a frequency response range of 0.5 Hz-10 kHz, and the sampling rate is set to 20 kHz. Second, the sensor signals are connected to a data acquisition card with a resolution of not less than 16 bits to ensure the accuracy of signal acquisition. Then, the audio and vibration signals are collected and stored in real time through a customized data acquisition software, and the double-buffering technology is used to ensure the continuity of data. Finally, the collected original signals are marked with time stamps for signal alignment in subsequent processing. The purpose of this step is to obtain high-quality original signals during the operation of the circuit breaker, laying a foundation for subsequent signal processing and feature extraction.

[0191] The specific implementation of step S20 is as follows: for the audio signal, first, short-time Fourier transform (STFT) is performed. The window function of STFT is selected as the Hanning window, the window length is set to 1024 sampling points, and the overlap rate is 50%. The mathematical expression of STFT is as follows:

[0192]

[0193] where x[n] is the input audio signal, w[n] is the window function, m is the time index, and ω is the angular frequency. Through STFT, the time-frequency representation of the audio signal can be obtained. Then, adaptive median filtering is performed on the STFT results to remove impulse noise. The window size of adaptive median filtering is initially set to 3x3 and can be expanded to 7x7 at most. The filtering process is as follows:

[0194] 1. Calculate the minimum value z min , the median value z med , and the maximum value z max in the current window.

[0195] 2. If z min < z med < z max , output z med ; otherwise, expand the window and repeat step 1.

[0196] 3. If the window reaches the maximum size and still does not meet the condition, output z med .

[0197] For the vibration signal, first use the wavelet denoising method. Select db4 wavelet basis, and the decomposition layer is 4. The soft threshold value processing is performed on the wavelet coefficient of each layer, and the threshold value λ is determined through the unbiased risk estimation principle:

[0198]

[0199] Where σ is the noise standard deviation, and N is the signal length. After wavelet denoising, the signal is subjected to envelope analysis. The steps of envelope analysis include:

[0200] 1. Perform Hilbert transform on the signal to obtain the analytic signal.

[0201] 2. Calculate the modulus of the analytic signal to obtain the envelope.

[0202] 3. Smooth the envelope using a moving average filter with a window length of 1% of the signal length.

[0203] Through the above processing, the first audio signal and the first vibration signal after denoising are obtained. The purpose of this step is to improve the signal-to-noise ratio of the signal and highlight the useful feature information, preparing for the subsequent signal fusion.

[0204] The specific implementation of step S30 is: first, time align the first audio signal and the first vibration signal obtained in step S20. Use the cross-correlation method to determine the time delay of the two signals, and then align the signals. The calculation formula of the time delay τ is:

[0205]

[0206] Where A[n] and V[n] are the audio and vibration signals respectively. Then, the aligned signals are fused through six different fusion equations.

[0207] 1. Weighted average fusion: F1(t) = w a A(t) + w v V(t), where w a and w v are initially set to 0.5.

[0208] 2. Adaptive filter fusion: use LMS algorithm, filter order M is set to 32, and step parameter μ initial value is set to 0.01. The fusion formula is:

[0209]

[0210] 3. Wavelet coefficient fusion: db4 wavelet is used, and the decomposition level is 4. For each level of coefficient, the maximum absolute value method is used for fusion:

[0211]

[0212] 4. Cross-correlation fusion: the sliding window method is used, and the window size is set to 10% of the signal length. The fusion formula is:

[0213]

[0214] 5. Energy ratio fusion: the local energy ratio is calculated, and the window size is set to 5% of the signal length. The fusion formula is:

[0215]

[0216] 6. Principal component analysis fusion: the audio and vibration signals are combined into a matrix [A, V], singular value decomposition is performed, and the first k principal components are taken. k is initially set to 2. The fusion formula is:

[0217]

[0218] The purpose of this step is to make full use of the complementary information of audio and vibration signals through various fusion methods, and to improve the information content and feature expression ability of the signal.

[0219] The specific implementation of step S40 is to adjust the parameters of the six fusion equations in step S30 to obtain multiple groups of fusion signals. The adjustment process uses a combination of grid search and random search methods. First, define the parameter vector Θ = [θ1, θ2,..., θN], where N is the total number of parameters to be adjusted. Then define the objective function: N

[0220]

[0221] where L is the mean square error loss function, y i is the true label, is the prediction result obtained using the parameter Θ, and M is the number of samples. For each fusion equation, the parameter adjustment is as follows:

[0222] 1. Weighted average fusion: adjust w a and w v , range [0, 1], step size 0.1.

[0223] 2. Adaptive filter fusion: adjust the filter order M (range: 16-64, step size: 8) and the step size parameter μ (range: 0.001-0.1, logarithmic scale).

[0224] ​3. Wavelet coefficient fusion: Adjust the wavelet basis (db2, db4, db6, db8) and the number of decomposition layers (3-6 layers).

[0225] 4. Cross-correlation fusion: Adjust the window size, ranging from 5% to 20% of the signal length, with a step size of 1%.

[0226] 5. Energy ratio fusion: Adjust the window size for local energy calculation, ranging from 1% to 10% of the signal length, with a step size of 1%.

[0227] 6. Principal component analysis fusion: Adjust the number of principal components k, ranging from 1 to 5.

[0228] To improve search efficiency, first perform a coarse-grained grid search, and then perform a random search near the optimal parameters for refinement. Use 5-fold cross-validation to evaluate parameter performance. Finally, select 3-5 different parameter combinations for each fusion method to generate multiple sets of fused signals. The purpose of this step is to explore the performance of different fusion strategies through parameter adjustment, providing diverse signal representations for subsequent feature extraction and optimization.

[0229] The specific implementation of step S50 is to perform feature extraction on the multiple sets of fused signals obtained in step S40. Feature extraction is divided into three aspects: time domain features, frequency domain features, and time-frequency domain features.

[0230] Time domain feature extraction:

[0231] 1. Calculate the mean:

[0232] 2. Calculate the variance:

[0233] 3. Calculate the skewness:

[0234] 4. Calculate the kurtosis:

[0235] 5. Calculate the peak factor:

[0236] 6. Calculate the margin factor:

[0237] Frequency domain feature extraction:

[0238] 1. Perform Fast Fourier Transform (FFT) on the signal:

[0239] 2. Calculate the power spectral density (PSD):

[0240] 3. Extract the spectral center:

[0241] 4. Extracting spectral variance:

[0242] 5. Extracting band energy ratio: Select 3-5 key frequency bands and calculate the proportion of each band energy in the total energy.

[0243] Time-frequency domain feature extraction:

[0244] 1. Continuous wavelet transform (CWT) is performed on the signal:

[0245] where ψ(t) is the wavelet basis function, and Morlet wavelet is selected.

[0246] 2. Calculate wavelet energy entropy: E = -∑ i,j p i,j log2(p i,j )

[0247] where,

[0248] 3. Extract wavelet coefficient statistical features: Calculate the mean, variance, skewness and kurtosis of wavelet coefficients at different scales.

[0249] For each group of fusion signals, the above features are extracted to form a feature vector. Then principal component analysis (PCA) is used for dimension reduction, and the principal components that explain 95% of the variance are retained. Finally, multiple groups of fusion features are obtained. The purpose of this step is to extract the features of the signal from multiple angles to fully characterize the operating state of the circuit breaker and provide effective feature representation for subsequent fault diagnosis.

[0250] The specific implementation of step S60 is to optimize the weights of multiple groups of fusion features using genetic algorithm, and calculate the optimal fusion features based on the least mean square error criterion. First, define the chromosome coding method, each chromosome represents a set of feature weights, the length is equal to the number of feature groups, and the value range of each gene is [0, 1]. The population size is set to 100 and the iteration number is 200. The specific steps of genetic algorithm are as follows:

[0251] 1. Initialize the population: randomly generate 100 chromosomes, and the sum of the genes of each chromosome is 1.

[0252] 2. Fitness calculation: use weighted features for fault diagnosis, and calculate the diagnosis accuracy as the fitness.

[0253] 3. Selection operation: adopt roulette wheel selection method, and the selection probability is proportional to the fitness.

[0254] 4. Crossover operation: use arithmetic crossover, and the crossover probability is set to 0.8.

[0255] Child1 = a * Parent1 + (1 - a) * Parent2

[0256] Child2 = (1 - a) * Parent1 + a * Parent2

[0257] where a is a random number in the range [0, 1].

[0258] 5. Mutation operation: Gaussian mutation is used with a mutation probability of 0.01.

[0259] Gene new = Gene old + N(0, 0.1)

[0260] where N(0, 0.1) represents a Gaussian random number with a mean of 0 and a standard deviation of 0.1.

[0261] 6. Elite preservation: The top 5% of individuals in the current population are preserved to the next generation.

[0262] Repeat steps 2-6 until the maximum number of iterations is reached or there is no significant improvement in the optimal fitness for 20 consecutive generations.

[0263] After obtaining the optimal weights, calculate the optimal fusion feature:

[0264]

[0265] where F opt is the optimal fusion feature, w i is the weight of the i-th feature group, F i is the i-th feature group, and K is the number of feature groups.

[0266] To further optimize the fusion feature, the least squares error criterion is used. Define the objective function:

[0267]

[0268] where N is the number of samples, y j is the true label of the j-th sample, and f(F opt,j ) is the prediction result obtained using the optimal fusion feature F opt,j .

[0269] Use gradient descent to fine-tune F opt :

[0270]

[0271] where η is the learning rate, initially set to 0.01, and an adaptive learning rate strategy is used. is the gradient of the objective function with respect to F opt .

[0272] Iterate until one of the following stopping conditions is met:

[0273] 1. The maximum number of iterations is reached (set to 1000).

[0274] 2. The objective function value changes less than a preset threshold (e.g., 10 ；6 ).

[0275] 3. The gradient norm is less than a preset threshold (e.g., 10 ；5 ).

[0276] The purpose of this step is to obtain a fusion feature that can best reflect the state of the circuit breaker by combining genetic algorithms and gradient optimization, providing high-quality input for subsequent fault diagnosis.

[0277] The specific implementation of step S70 is to input the optimal fusion feature obtained in step S60 into a pre-trained circuit breaker mechanical fault diagnosis model to obtain a fault diagnosis vector and output the diagnosis result. The specific implementation process is as follows:

[0278] First, build a deep learning model. Choose a multi-layer perceptron (MLP) network as the fault diagnosis model. The network structure is: input layer (node number equal to the dimension of the optimal fusion feature) - fully connected layer (128 neurons) - Dropout layer (dropout rate 0.5) - fully connected layer (64 neurons) - Dropout layer (dropout rate 0.3) - fully connected layer (32 neurons) - output layer (node number equal to the number of fault types). The activation function is ReLU, and the output layer uses Softmax activation.

[0279] The mathematical expression of the model is as follows:

[0280] h1 = ReLU(W1F opt +b1)

[0281] h2 = Dropout(ReLU(W2h1+b2))

[0282] h3 = Dropout(ReLU(W3h2+b3))

[0283] y = Softmax(W4h3+b4)

[0284] Where W i and b i are the weight matrix and bias vector of the i-th layer, F opt is the input optimal fusion feature, and y is the output fault diagnosis vector.

[0285] Then, the model is trained. Using the historical dataset, containing circuit breaker operation data of known fault types. The dataset is divided into training, validation, and test sets in the ratio of 8:1:1. Cross-entropy loss function is adopted:

[0286]

[0287] Where C is the number of fault types, y i is the true label, is the model prediction probability.

[0288] The model is trained using the Adam optimizer, with an initial learning rate of 0.001 and a batch size of 64. Early stopping strategy is used during training, stopping when the validation set loss does not decrease for 10 consecutive epochs. Meanwhile, learning rate decay strategy is adopted, reducing the learning rate to 0.1 times of the original every 50 epochs.

[0289] After the model is trained, the test set is used to evaluate the model performance. Accuracy, precision, recall, and F1 score are calculated. If the model performance does not meet the requirements (such as accuracy less than 95%), the network structure or hyperparameters are adjusted and retrained.

[0290] Finally, the optimal fusion features obtained in step S60 are input into the trained model to obtain the fault diagnosis vector. Each element in the diagnosis vector represents the probability of the corresponding fault type. The type with the highest probability is selected as the final diagnosis result. At the same time, the diagnosis confidence, i.e. the highest probability value, is output.

[0291] To improve the reliability of diagnosis, a probability threshold (such as 0.8) is set. If the highest probability is lower than the threshold, a "suspected fault" diagnosis result is given, and further inspection is recommended.

[0292] In addition, for some key fault types, if their probability exceeds the preset alert threshold (such as 0.3), even if it is not the highest probability, a warning message will be given in the diagnosis result.

[0293] The output format of the diagnosis result is:

[0294] 1. Main diagnosis result: fault type (or "normal" / "suspected fault")

[0295] 2. Diagnosis confidence: highest probability value

[0296] 3. Secondary diagnosis result: other fault types whose probability exceeds the alert threshold (if any)

[0297] 4. Suggestions: give corresponding maintenance or further inspection suggestions according to the diagnosis result

[0298] The purpose of this step is to use the trained deep learning model to quickly and accurately diagnose the mechanical fault type of the circuit breaker based on the optimal fusion features, and provide reliable diagnosis results and suggestions, providing strong support for the maintenance and management of the circuit breaker.

[0299] The specific implementation of step S80 is: comparing the fault diagnosis result with a preset safety threshold, if the fault diagnosis result exceeds the safety threshold, it is determined that the circuit breaker has potential fault risk, if the fault diagnosis result does not exceed the safety threshold, it is determined that the circuit breaker is in a normal working state, based on the comparison result of the fault diagnosis result and the safety threshold, the reliability of the circuit breaker is calculated and output, wherein the formula for calculating the reliability is: R=e ；λt ;

[0300] Wherein:

[0301] R is the reliability of the circuit breaker;

[0302] λ is the failure rate, and the calculation formula is:

[0303] λ0 is the reference failure rate, which can be determined according to historical data or experience value;

[0304] t is the running time.

[0305] In summary, the reliability diagnosis method of the circuit breaker equipment provided by the present application realizes high-precision diagnosis of the mechanical fault of the circuit breaker through the steps of multi-sensor signal acquisition, signal preprocessing, multi-method signal fusion, feature extraction and optimization, and deep learning model diagnosis. The method fully utilizes the complementary information of audio and vibration signals, improves the expression ability of features through various fusion strategies and feature optimization, and captures complex fault patterns by using a deep learning model, thereby improving the accuracy and reliability of the diagnosis.

[0306] Specifically, the principle of the present application is:

[0307] 1. Multi-signal source fusion principle: the audio signals and vibration signals generated by the circuit breaker during operation contain rich state information, but the two kinds of signals differ in time-frequency characteristics, noise sensitivity, etc. The present application adopts various signal fusion methods such as weighted average, adaptive filtering, wavelet coefficient fusion, etc., which can fully utilize the complementary information of different signal sources, improve the integrity and reliability of the information. This multi-signal source fusion method effectively reduces the problem of single signal source being easily disturbed, and improves the robustness of the diagnosis.

[0308] 2. Adaptive signal processing principle: The operating environment of circuit breakers is complex, and the original signal often contains various interference and noise. The short-time Fourier transform used in this invention can reflect the time-frequency characteristics of the signal, and the adaptive median filter can effectively remove impulse noise. The wavelet denoising method can remove random noise while preserving the useful components of the signal. These adaptive signal processing methods can be targeted according to the characteristics of the signal, significantly improving the signal-to-noise ratio and creating favorable conditions for subsequent feature extraction.

[0309] 3. Multi-dimensional feature extraction principle: The fault modes of circuit breakers are complex and diverse, and single-dimensional features are difficult to fully reflect their state. This invention extracts features from three dimensions: time domain, frequency domain, and time-frequency domain, covering various types of statistical features, spectral features, and wavelet features. This multi-dimensional feature extraction method can fully capture the operating state information of the circuit breaker, improving the expression and discrimination ability of the features.

[0310] 4. Feature fusion optimization principle: Different features contribute differently to fault diagnosis, and simple feature combination may result in redundancy or information loss. This invention uses genetic algorithm to optimize feature weights, utilizing its global search ability to find an optimal weight combination in a complex feature space. Meanwhile, the fine-tuning process based on the least mean square error criterion can further optimize feature representation and improve the discriminability of features.

[0311] 5. Deep learning diagnosis model principle: The complexity and diversity of circuit breaker fault modes require a diagnosis model with strong non-linear mapping ability. The multi-layer perceptron network used in this invention has a multi-layer non-linear transformation structure, which can automatically learn the internal features and complex patterns of the data. Through training on a large amount of historical data, the model can establish a complex mapping relationship between features and fault types, achieving high-precision fault recognition.

[0312] 6. Reliability evaluation principle: The reliability of the diagnosis result is crucial for maintenance decisions. The probability threshold and warning threshold mechanism introduced in this invention, based on the probability distribution of the diagnosis model output, can quantitatively evaluate the diagnosis result. This method not only identifies the explicit fault type, but also points out potential risks and issues that need further attention, providing more comprehensive information support for maintenance decisions.

[0313] In summary, the technical scheme of this invention, through a series of technical innovations such as multi-signal source fusion, adaptive signal processing, multi-dimensional feature extraction, feature fusion optimization, deep learning model, and reliability evaluation, has constructed a logically complete and technically advanced circuit breaker fault diagnosis system.

[0314] In order to better understand and implement the present application, the following provides an embodiment of a specific application scenario of the present application: there are multiple circuit breakers in a certain 330kV substation, in order to improve the operation reliability and maintenance efficiency of the circuit breakers, the station decides to use a reliability diagnosis method of a circuit breaker device of the present application for intelligent transformation. The following describes in detail the complete process of fault diagnosis on a ZN65-12 type vacuum circuit breaker.

[0315] 1. Sensor installation and data acquisition

[0316] Firstly, high-sensitivity audio sensors and vibration sensors are installed at key positions of the circuit breaker. The audio sensor selects a condenser microphone with a frequency response range of 20Hz-20kHz, and a sensitivity of -38dB±3dB (0dB=1V / Pa, 1kHz). The vibration sensor selects a three-axis piezoelectric acceleration sensor with a measurement range of ±100g, a frequency response range of 0.5Hz-10kHz, and a sensitivity of 100mV / g.

[0317] These sensors are connected to a 16-bit resolution high-speed data acquisition card through shielded cables. The sampling rate of the audio signal is set to 44.1kHz, and the sampling rate of the vibration signal is set to 20kHz. Customized data acquisition software is used to collect and store signals in real time, and double buffering technology is used to ensure the continuity of data.

[0318] When the circuit breaker is operated, 10 seconds of audio signals and vibration signals are collected simultaneously. The original signals are saved as binary files after being marked with time stamps for subsequent processing.

[0319] 2. Signal preprocessing

[0320] The collected original signals are preprocessed to improve signal quality.

[0321] For audio signals, first perform short-time Fourier transform (STFT). Select a Hanning window, set the window length to 1024 sampling points, and the overlap rate to 50%. Then perform adaptive median filtering on the STFT results, set the initial window size to 3x3, and the maximum expandable to 7x7.

[0322] For vibration signals, use wavelet denoising method. Select db4 wavelet basis, and decomposition level is 4. Use unbiased risk estimation principle to determine the threshold, and perform soft threshold processing on the wavelet coefficients. Then perform envelope analysis, and use a moving average filter with a length of 1% of the signal length to smooth the obtained envelope.

[0323] 3. Signal fusion

[0324] The pre-processed audio signal and vibration signal were time-aligned. Through cross-correlation analysis, the time delay between the two signals was determined to be 23 ms. After alignment, the signals were fused using six different fusion methods:

[0325] (1) Weighted average fusion: The initial weight was set to w a = w v = 0.5.

[0326] (2) Adaptive filter fusion: The filter order M was initially set to 32, and the step parameter μ was initially set to 0.01.

[0327] (3) Wavelet coefficient fusion: db4 wavelet was used, and the decomposition level was 4.

[0328] (4) Cross-correlation fusion: The window size was set to 10% of the signal length.

[0329] (5) Energy ratio fusion: The local energy calculation window size was set to 5% of the signal length.

[0330] (6) Principal component analysis fusion: The initial number of principal components k was set to 2.

[0331] 4. Parameter adjustment and feature extraction

[0332] The parameters of the above six fusion methods were optimized. Taking weighted average fusion as an example, through grid search method, the optimal combination of w a and w v was searched in the range of [0, 1] with a step size of 0.1. Similar parameter optimization was also performed for other fusion methods.

[0333] After optimization, six groups of fused signals were obtained. For each group of fused signals, the following features were extracted:

[0334] Time domain features: mean, variance, skewness, kurtosis, peak factor, and margin factor.

[0335] Frequency domain features: power spectral density was calculated through FFT, spectral center and spectral variance were extracted, and energy ratios of three key frequency bands (0-500Hz, 500-2000Hz, 2000-5000Hz) were calculated.

[0336] Time-frequency domain features: Morlet wavelet was used for continuous wavelet transform, wavelet energy entropy was calculated, and wavelet coefficient statistical features (mean, variance, skewness, kurtosis) of 4 scales were extracted.

[0337] Finally, each group of fused signals obtained a feature vector containing 28 features. Through principal component analysis, the principal components explaining 95% of the variance were retained, and 6 groups of reduced fusion features were obtained.

[0338] 5. Feature fusion optimization

[0339] The weights of the 6 groups of fused features were optimized using a genetic algorithm. The population size was set to 100, and the number of iterations was 200. The crossover probability was set to 0.8, and the mutation probability was set to 0.01. The fitness function was defined as the accuracy of preliminary fault diagnosis using the weighted features.

[0340] After optimization, the optimal weights were obtained as follows: 0.22, 0.18, 0.25, 0.15, 0.12, and 0.08. Using these weights, the optimal fused features were calculated. Then, based on the least mean square error criterion, the gradient descent method was used to fine-tune the optimal fused features. The initial learning rate was set to 0.01, and an adaptive learning rate strategy was used. After 1000 iterations, the final optimized fused features were obtained.

[0341] 6. Fault diagnosis model training

[0342] A multi-layer perceptron (MLP) network was constructed as the fault diagnosis model. The network structure was as follows: input layer (number of nodes equal to the dimension of the optimized fused features) - fully connected layer (128 neurons) - Dropout layer (dropout rate 0.5) - fully connected layer (64 neurons) - Dropout layer (dropout rate 0.3) - fully connected layer (32 neurons) - output layer (number of nodes equal to the number of fault types).

[0343] The model was trained using the historical data set. The data set contained 5000 samples, covering 7 common circuit breaker mechanical fault types and normal states. The data set was divided into training set, validation set and test set in the ratio of 8:1:1.

[0344] The cross-entropy loss function was used, and the Adam optimizer was used for model training. The initial learning rate was set to 0.001, and the batch size was set to 64. The early stopping strategy was used, and the training was stopped when the validation set loss did not decrease for 10 consecutive epochs. At the same time, the learning rate decay strategy was used, and the learning rate was reduced to 0.1 times of the original value every 50 epochs.

[0345] After training, the model achieved an accuracy of 97.8% on the test set. The performance indicators are shown in Table 1:

[0346] Table 1 Model performance table

[0347] Indicator Value Accuracy 97.8％ Mean Precision 97.6％ Mean Recall 97.5％ Mean F1 Score 97.5％

[0348] 7. Actual fault diagnosis

[0349] After completing the model training, actual fault diagnosis is performed on the target circuit breaker. First, the audio and vibration signals of the circuit breaker are collected according to the aforementioned steps, preprocessed, signal fused, features extracted, and optimized. The optimal fusion features obtained are input into the trained MLP model.

[0350] The fault diagnosis vector output by the model is shown in Table 2:

[0351] Table 2 Fault Diagnosis Vector

[0352] Fault Type Probability Normal 0.02 Stuck Operating Mechanism 0.01 Spring Failure 0.03 Abnormal Buffer Device 0.89 Contact Wear 0.02 Vacuum Reduction 0.01 Operating Circuit Failure 0.01 Insulator Damage 0.01

[0353] According to the diagnosis result, the most likely fault type of the circuit breaker is "buffer device abnormality" with a diagnosis confidence of 89%, so the output [buffer device abnormality, 0.89] is taken as the diagnosis result.

[0354] 8. Safety identification analysis and suggestions

[0355] The safety threshold is set to 0.8, and since the buffer device abnormality in the diagnosis result is 0.89, which is greater than the safety threshold, it is determined that there is a potential fault risk, and the reliability is calculated as follows:

[0356] According to the provided information, the reliability of the circuit breaker can be calculated in the following steps:

[0357] Calculate the failure rate λ:

[0358]

[0359] Assuming the reference failure rate λ0 = 0.1, then the failure rate λ is:

[0360]

[0361] Assuming the operating time t = 10 hours, then the reliability R is:

[0362] R = e ；λt = e ；0.1113×10 = 0.8948

[0363] Therefore, under the given conditions, the reliability of the circuit breaker is R = 0.8948.

[0364] Next, the early warning mechanism is triggered, and preventive maintenance suggestions are generated:

[0365] (1) Immediately arrange professional personnel to check the buffer device of the circuit breaker, focusing on checking components such as buffer springs, buffer oil cylinders, etc.

[0366] (2) During the maintenance process, measure the buffer stroke and buffer force and compare them with the standard values.

[0367] (3) Check whether the fixing bolts of the buffer device are loose and whether the buffer oil is leaking.

[0368] (4) If any abnormality is found, replace or repair the buffer device in a timely manner.

[0369] (5) After repair, perform a circuit breaker opening and closing test, and verify the diagnostic system again.

[0370] 9. Verification of Diagnostic Effectiveness

[0371] Maintenance personnel inspected the circuit breaker according to system recommendations. The inspection revealed that the buffer oil in the buffer cylinder was severely aged, with reduced viscosity, resulting in poor buffering performance. After replacing the buffer oil, the circuit breaker underwent opening and closing tests, and operation was normal.

[0372] The diagnostic results obtained by performing the test again using this diagnostic system are shown in Table 3.

[0373] Table 3. Results of Re-diagnosis

[0374] Fault Type Probability Normal 0.97 Stuck Operating Mechanism 0.01 Spring Failure 0.00 Abnormal Buffer Device 0.01 Contact Wear 0.00 Vacuum Reduction 0.00 Operating Circuit Failure 0.01 Insulator Damage 0.00

[0375] The diagnostic results showed that the circuit breaker had returned to normal operation with a confidence level of 97%. The circuit breaker fault diagnosis method of the present invention significantly improves the operational reliability of the circuit breaker.

[0376] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A reliability diagnosis method of a circuit breaker device, characterized by, The method comprises the following steps: S10, continuously acquiring audio signals and vibration signals in the operation process of the circuit breaker; S20, performing short-time Fourier transform and adaptive median filtering on the audio signals to obtain first audio signals, and performing wavelet denoising and envelope analysis on the vibration signals to obtain first vibration signals; S30, fusing the audio signals and the vibration signals aligned in time according to a preset fusion equation set to obtain a fusion signal, the fusion equation set comprising a weighted average fusion equation, an adaptive filtering fusion equation, a wavelet coefficient fusion equation, a cross-correlation fusion equation, an energy ratio fusion equation, and a principal component analysis fusion equation; S40, adjusting parameters of the fusion equation set to obtain multiple groups of fusion signals; S50, respectively extracting features from the multiple groups of fusion signals to obtain multiple groups of fusion features; S60, optimizing weights of the multiple groups of fusion features by using a genetic algorithm, and calculating optimal fusion features based on the multiple groups of fusion features according to a minimum mean square error criterion; S70, inputting the optimal fusion features into a pre-trained circuit breaker mechanical fault diagnosis model to obtain a fault diagnosis result of the circuit breaker; S80, comparing the fault diagnosis result with a preset safety threshold, determining that the circuit breaker has a potential fault risk if the fault diagnosis result exceeds the safety threshold, determining that the circuit breaker is in a normal working state if the fault diagnosis result does not exceed the safety threshold, and calculating the reliability of the circuit breaker based on the comparison result of the fault diagnosis result and the safety threshold and outputting the reliability.

2. The reliability diagnosis method of a circuit breaker device according to claim 1, characterized by, The weighted average fusion equation is specifically represented as: F(t) = w a A(t) + w v V(t); In the formula, F(t) is the fused signal; A(t) is the audio signal; V(t) is the vibration signal; w a and w v are the weights of the audio and vibration signals respectively, satisfying w a +w v =1.

3. The reliability diagnosis method of a circuit breaker device according to claim 2, characterized by, The adaptive filtering fusion equation is specifically represented as: In the formula, F(n) is the fused discrete-time signal; h ξ (n) is the coefficient of the adaptive filter; M is the filter order; w a and w v are the weight coefficients.

4. The reliability diagnosis method of a circuit breaker device according to claim 3, characterized by, The wavelet coefficient fusion equation is specifically represented as: F = W -1 (φ(W A (A), W V (V))) where W A and W V represent wavelet transforms of the audio and vibration signals, respectively; W -1 represents an inverse wavelet transform; and φ is a fusion function of the wavelet coefficients.

5. The reliability diagnosis method of a circuit breaker apparatus according to claim 4, characterized by, The cross-correlation fusion equation is specifically represented as: or a discrete form: where N is a cross-correlation window size, serving as an adjustment parameter.

6. The reliability diagnosis method of a circuit breaker apparatus according to claim 5, characterized by, The energy ratio fusion equation is specifically represented as: where E A and E V are the energies of the audio and vibration signals, respectively:

7. The reliability diagnosis method of a circuit breaker apparatus according to claim 6, characterized by, The principal component analysis fusion equation is specifically represented as: In the formula, [A, V] is a matrix combining the audio and vibration signals; U k ,∑ k ,V k is a matrix obtained by taking the first k principal components after singular value decomposition of [A, V]; and k is the number of principal components.

8. The reliability diagnosis method of a circuit breaker apparatus according to claim 7, characterized by, The training steps of the circuit breaker mechanical fault diagnosis model are specifically: establishing a training data set, specifically, obtaining optimal fusion features of circuit breakers with known mechanical faults according to the steps S10-S60; model training, specifically, establishing a multi-layer perceptron network and training the network by using the training data set, wherein the input of the training is the optimal fusion features of each group of circuit breakers with known mechanical faults, and the output of the training is the corresponding known mechanical faults, to obtain the circuit breaker mechanical fault diagnosis model.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores program instructions, and the program instructions are used to execute the reliability diagnosis method of the circuit breaker device according to any one of claims 1-8 when running.

10. A switch safety recognition system of a circuit breaker apparatus, characterized by, The computer readable storage medium comprises the computer readable storage medium according to claim 9. The computer readable storage medium comprises the computer readable storage medium according to claim 9.

Images