Reliability diagnosis method of circuit breaker equipment, medium and system

Abstract

The invention provides a circuit breaker equipment reliability diagnosis method, a medium and a system, and belongs to the technical field of switch safety identification of circuit breaker equipment.Audio signals and vibration signals in the operation process of a circuit breaker are continuously obtained, and the reliability of the circuit breaker equipment is obtained through short-time Fourier transform, adaptive median filtering, wavelet denoising, envelope analysis and the like. And clean audio and vibration signals are obtained. Then, performing time alignment and fusion on the two types of signals to generate a plurality of groups of fusion signals; feature extraction is carried out on the fusion signals, the feature weight is optimized by adopting a genetic algorithm, and the optimal fusion feature is calculated; and finally, inputting the optimal fusion feature into a pre-trained fault diagnosis model, and outputting a fault diagnosis vector of the circuit breaker. The technical problem that a diagnosis result is unstable due to the fact that a single-signal-source diagnosis method in the prior art is difficult to comprehensively reflect the operation state of the circuit breaker and is easily influenced by environmental noise and interference is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of switch safety identification of circuit breaker equipment, and in particular, relates to a reliability diagnosis method, medium and system for circuit breaker equipment. Background Art

[0002] Circuit breakers are key equipment in power systems, and their reliability directly affects the safe and stable operation of power grids. With the rapid development of smart grids, higher requirements are placed on the reliability and maintenance efficiency of circuit breakers. Traditional circuit breaker maintenance methods mainly rely on regular inspections and manual inspections, which have problems such as high maintenance costs, low efficiency, and failure to detect potential faults in a timely manner.

[0003] In recent years, with the progress of sensor technology, signal processing technology and artificial intelligence technology, circuit breaker fault diagnosis methods based on condition monitoring have been widely studied and applied. Existing circuit breaker fault diagnosis methods mainly include diagnosis methods based on single signal sources such as vibration signal analysis, sound signal analysis, and current signal analysis. These methods have improved the accuracy and real-time performance of circuit breaker fault diagnosis to a certain extent, but the diagnosis methods based on a single signal source are still difficult to fully reflect the operating status of the circuit breaker and are easily affected by environmental noise and interference, resulting in unstable safety identification results. Summary of the invention

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

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides a reliability diagnosis method for a circuit breaker device, comprising the following steps:

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

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

[0009] S30, fusing the time-aligned audio signal and the vibration signal using a preset fusion equation group to obtain a fusion signal, wherein the fusion equation group includes 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 the parameters of the fusion equation group to obtain multiple groups of fusion signals;

[0011] S50, performing feature extraction on the multiple groups of fusion signals respectively to obtain multiple groups of fusion features;

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

[0013] 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;

[0014] S80. Compare the fault diagnosis result with a preset safety threshold. If the fault diagnosis result exceeds the safety threshold, determine that the circuit breaker has a potential fault risk; if the fault diagnosis result does not exceed the safety threshold, determine that the circuit breaker is in a normal working state; based on the comparison result of the fault diagnosis result and the safety threshold, calculate the reliability of the circuit breaker and output it.

[0015] Wherein, each equation in the fusion equation group is specifically expressed as follows:

[0016] 1. Weighted average fusion equation:

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

[0018] Where 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 audio and vibration signals respectively, satisfying w a +w v =1; the weight can be determined by analyzing the signal-to-noise ratio or energy of each signal, and can also be optimized through methods such as cross-validation.

[0019] 2. Adaptive filtering fusion equation:

[0020]

[0021] Where F(n) is the discrete time signal after fusion; h ξ (n) is the coefficient of the adaptive filter; M is the filter order; w a and w v is the weight coefficient; the filter coefficient can be updated by adaptive algorithms such as LMS or RLS:

[0022] h ξ (n+1)=h ξ(n)+μe(n)x(n-ξ);

[0023] Where μ is the step size 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 Respectively represent the wavelet transform of audio and vibration signals; W -1 represents the inverse wavelet transform; φ is the fusion function of the wavelet coefficients, and the maximum value, weighted average, etc. can be selected:

[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 levels 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 serves as an adjustment parameter.

[0036] 5. Energy ratio fusion equation:

[0037]

[0038] Where E A and E V The energy of 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] Where [A, V] is the matrix that combines the audio and vibration signals; U k ,Σ k ,V k It is the matrix obtained by taking the first k principal components after singular value decomposition of [A, V]; k is the number of principal components and can be used as an adjustment parameter.

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

[0045] 1. Define the parameter vector:

[0046] θ=[θ1,θ2,...,θ N ]

[0047] where θ i It 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 is the true label, is the prediction result obtained using parameter Θ, and M is the number of samples.

[0051] 3. Parameter optimization process:

[0052]

[0053] in,

[0054] where Θ * is the optimal parameter combination, and are the lower and upper bounds of the ith 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 cross-validation folds, J k(Θ) is the objective function value on the k-th fold validation set.

[0061] 6. Grid Search:

[0062]

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

[0064] 7. Optional, random search:

[0065]

[0066] in Indicated in Uniform distribution on the interval.

[0067] 8. Optional, Bayesian Optimization:

[0068]

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

[0070] 9. Optional, Genetic Algorithm:

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

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

[0073] 10.Optional, 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 optimal position of the particle, and g is the global optimal position.

[0076] Finally, different parameters of the fusion equation are obtained.

[0077] Furthermore, the training steps of the circuit breaker mechanical fault diagnosis model are specifically:

[0078] Establishing a training data set, specifically, following steps S10-S60 to obtain optimal fusion features of multiple groups of circuit breakers with known mechanical faults;

[0079] Model training specifically involves establishing a multi-layer perceptron network and using the training data set for training, wherein the training input is the optimal fusion feature of each group of circuit breakers with known mechanical faults, and the training output is the corresponding known mechanical faults, thereby obtaining a circuit breaker mechanical fault diagnosis model.

[0080] The multiple groups are generally more than 32 groups, and the known mechanical failures include various types of mechanical failures.

[0081] Specifically, the step S10 includes:

[0082] Step 101: Install a high-sensitivity audio sensor and a vibration sensor at a key position of the circuit breaker, wherein the audio sensor is a condenser microphone with a wide frequency response range, and the vibration sensor is a triaxial piezoelectric acceleration sensor;

[0083] Step 102: Connect the signals of the audio sensor and the vibration sensor to a data acquisition card;

[0084] Step 103: collect and store audio signals and vibration signals in real time through customized data acquisition software;

[0085] Step 104: Time stamp the collected original audio signal and original vibration signal.

[0086] Wherein, the step S20 specifically includes:

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

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

[0089] Step 203, performing noise reduction processing on the vibration signal using a wavelet denoising method;

[0090] Step 204, performing Hilbert transform on the noise-reduced vibration signal to obtain an analytical signal;

[0091] Step 205, calculating the modulus of the analytical signal to obtain the envelope of the vibration signal;

[0092] Step 206: Smooth the envelope of the vibration signal using a moving average filter.

[0093] Wherein, the step S30 specifically includes:

[0094] Step 301: Time-align the first audio signal and the first vibration signal, and determine the time delay of the two signals using a cross-correlation method;

[0095] Step 302: performing weighted average fusion on the time-aligned signals;

[0096] Step 303: performing adaptive filtering fusion on the time-aligned signals;

[0097] Step 304: performing wavelet coefficient fusion on the time-aligned signal;

[0098] Step 305: performing cross-correlation fusion on the time-aligned signals;

[0099] Step 306: Perform energy ratio fusion on the time-aligned signals;

[0100] Step 307: Perform principal component analysis and fusion on the time-aligned signals.

[0101] Wherein, the step S40 specifically includes:

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

[0103] Step 402: define the objective function and use the mean square error loss function;

[0104] Step 403: adjusting the weight parameter of the weighted average fusion equation;

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

[0106] Step 405, adjusting the wavelet basis and the number of decomposition levels of the wavelet coefficient fusion equation;

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

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

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

[0110] Step 409: Optimize parameters by combining grid search and random search;

[0111] Step 410: Use cross-validation to evaluate parameter performance and select multiple groups of different parameter combinations for each fusion method.

[0112] Wherein, the step S50 specifically includes:

[0113] Step 501: extracting time domain features from multiple groups of fusion signals respectively, including mean, variance, skewness, kurtosis, peak factor and margin factor;

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

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

[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: perform principal component analysis on the feature vector and perform dimensionality reduction processing.

[0118] Wherein, the step S60 specifically includes:

[0119] Step 601, define a chromosome encoding method, where each chromosome represents a set of feature weights;

[0120] Step 602: Initialize the population and randomly generate multiple chromosomes;

[0121] Step 603: Calculate the fitness of each chromosome, use weighted features to perform fault diagnosis, and calculate the diagnostic accuracy as the fitness;

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

[0123] Step 605: Perform a crossover operation using arithmetic crossover;

[0124] Step 606: Perform a mutation operation using Gaussian mutation;

[0125] Step 607: Perform elite retention to retain the best individuals 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 there is no significant improvement in the optimal fitness for multiple consecutive generations;

[0127] Step 609: After obtaining the optimal weight, calculate the optimal fusion feature;

[0128] Step 610: Use the minimum mean square error criterion and the gradient descent method to fine-tune the optimal fusion feature.

[0129] Wherein, the step S70 specifically includes:

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

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

[0132] Step 703: Use an early stopping strategy and a learning rate decay strategy during training;

[0133] Step 704: Use the test set to evaluate the model performance and calculate indicators such as accuracy, precision, recall, and F1 score;

[0134] Step 705: input the optimal fusion feature into the trained model to obtain a 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 includes:

[0137] The fault diagnosis result is compared with a preset safety threshold. If the fault diagnosis result exceeds the safety threshold, it is determined that the circuit breaker has a 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 specifically: R = e ；λt ;

[0138] in:

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

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

[0141] λ0 is the benchmark failure rate, which can be determined based on historical data or empirical values;

[0142] t is the running time.

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

[0144] Step 4091: Set a search range and step size for each parameter that needs to be adjusted;

[0145] Step 4092: Generate all possible parameter combinations in the set parameter space;

[0146] Step 4093: For each parameter combination, use the training data to evaluate the model performance;

[0147] Step 4094: Select the parameter combination with the best performance as the optimal parameters;

[0148] The random search method specifically includes:

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

[0150] Step 4096: randomly sample and generate multiple sets of parameter combinations from the set distribution;

[0151] Step 4097: For each randomly generated parameter combination, evaluate the model performance using the training data;

[0152] Step 4098: Select the parameter combination with the best performance as the optimal parameters.

[0153] Optionally, the process of fine-tuning the optimal fusion feature using the gradient descent method in step 610 specifically includes:

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

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

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

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

[0158] Optionally, the process of outputting the diagnosis result in step 707 specifically includes:

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

[0160] Step 7072: setting a warning threshold, when the probability of certain key fault types exceeds the warning threshold, giving a warning message in the diagnosis result;

[0161] Step 7073: Generate corresponding maintenance suggestions or further inspection suggestions based on the diagnosis results;

[0162] Step 7074: Integrate the primary diagnosis results, diagnosis confidence, secondary diagnosis results and recommendations to form a complete diagnosis report.

[0163] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned reliability diagnosis method for circuit breaker equipment.

[0164] A third aspect of the present invention provides a switch safety identification system for a circuit breaker device, which includes the above-mentioned computer-readable storage medium.

[0165] Compared with the prior art, the reliability diagnosis method, medium and system of a circuit breaker device provided by the present invention have the following beneficial effects:

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

[0167] 2. Adaptive signal processing: The present invention uses short-time Fourier transform, adaptive median filtering, wavelet denoising and other methods to pre-process the original signal, which effectively improves the signal-to-noise ratio of the signal and lays a good foundation for subsequent feature extraction.

[0168] 3. Diversified feature extraction: The present invention extracts features from three aspects: time domain, frequency domain and time-frequency domain, comprehensively describing the operating status of the circuit breaker. Dimensionality reduction is performed through principal component analysis, which improves the expression efficiency of features.

[0169] 4. Feature fusion optimization: The present invention uses a genetic algorithm to optimize the weights of multiple sets of fusion features, and fine-tunes based on the minimum mean square error criterion to obtain the optimal fusion feature. This method significantly improves the discriminative ability of the features and the performance of the diagnostic model.

[0170] 5. High-performance diagnostic model: The present invention uses a deep learning model for fault diagnosis. Compared with traditional machine learning methods, it has stronger feature learning and pattern recognition capabilities, and can adapt to complex fault modes and operating condition changes.

[0171] 6. Reliability assessment: The present invention introduces probability thresholds and warning thresholds to conduct reliability assessments on diagnostic results, and gives corresponding warning information and suggestions, providing strong support for maintenance decisions.

[0172] In summary, the reliability diagnosis method of a circuit breaker device of the present invention has made significant progress in diagnostic accuracy, reliability, adaptability and real-time performance, and solves the technical problems that the diagnosis method of a single signal source in the prior art is difficult to fully reflect the operating status of the circuit breaker and is easily affected by environmental noise and interference, resulting in unstable safety identification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0173] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0174] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0175] like Figure 1 FIG. 1 is a flow chart of a reliability diagnosis method for a circuit breaker device provided by the present invention. The method comprises the following steps:

[0176] S10, continuously acquiring audio signals and vibration signals during the operation of the circuit breaker;

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

[0178] S30, fusing the time-aligned audio signal and the vibration signal using a preset fusion equation group to obtain a fusion signal, where the fusion equation group includes 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 the parameters of the fusion equation group to obtain multiple groups of fusion signals;

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

[0181] S60, optimizing weights of multiple groups of fusion features by using a genetic algorithm, and calculating optimal fusion features by using multiple groups of fusion features 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 circuit breaker fault diagnosis result;

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

[0184] in:

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

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

[0187] λ0 is the benchmark failure rate, which can be determined based on historical data or empirical values;

[0188] t is the running time.

[0189] The specific implementation methods of the above steps are described in detail below:

[0190] The specific implementation method of step S10 is: first, install highly sensitive audio sensors and vibration sensors at key positions of the circuit breaker. The audio sensor uses a condenser microphone with a wide frequency response range (20Hz-20kHz), and the sampling rate is set to 44.1kHz to capture the sound signal during the operation of the circuit breaker. The vibration sensor uses a triaxial piezoelectric acceleration sensor with a measurement range of ±100g, a frequency response range of 0.5Hz-10kHz, and a sampling rate of 20kHz. Secondly, connect the sensor signal 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 customized data acquisition software, and double buffering technology is used to ensure the continuity of the data. Finally, the collected original signal is timestamped to facilitate signal alignment during subsequent processing. The purpose of this step is to obtain high-quality original signals during the operation of the circuit breaker, laying the foundation for subsequent signal processing and feature extraction.

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

[0192]

[0193] Among them, 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, the STFT result is adaptively median filtered to remove impulse noise. The window size of the adaptive median filter is initially set to 3×3 and can be expanded to 7×7. The filtering process is as follows:

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

[0195] 2. If z min <z med <z max , then 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 vibration signals, the wavelet denoising method is first used. The db4 wavelet basis is selected and the number of decomposition layers is 4. The wavelet coefficients of each layer are soft-thresholded, and the threshold λ is determined by the unbiased risk estimation principle:

[0198]

[0199] Where σ is the standard deviation of noise 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 analytical signal.

[0201] 2. Calculate the modulus of the analytical signal and 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 noise reduction are obtained. The purpose of this step is to improve the signal-to-noise ratio of the signal, highlight the useful feature information, and prepare for subsequent signal fusion.

[0204] The specific implementation of step S30 is: first, the first audio signal and the first vibration signal obtained in step S20 are time-aligned. The time delay of the two signals is determined by a cross-correlation method, and then the signals are aligned. The calculation formula of the time delay τ is:

[0205]

[0206] Among them, A[n] and V[n] are 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 Initially set to 0.5.

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

[0209]

[0210] 3. Wavelet coefficient fusion: Use db4 wavelet with 4 decomposition layers. For each layer of coefficients, use the maximum absolute value method for fusion:

[0211]

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

[0213]

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

[0215]

[0216] 6. Principal component analysis fusion: Combine the audio and vibration signals into a matrix [A, V], perform singular value decomposition, and take the first k principal components. 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 a variety of fusion methods to improve the information content and feature expression ability of the signal.

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

[0220]

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

[0222] 1. Weighted average fusion: adjust w a and w v , the range is [0,1], and the step size is 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 in the range of 5%-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%-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] In order to improve the search efficiency, a coarse-grained grid search is first performed, and then a random search is performed near the optimal parameters for refinement. Parameter performance is evaluated using 5-fold cross validation. Finally, 3-5 different parameter combinations are selected for each fusion method to generate multiple sets of fusion signals. The purpose of this step is to explore the performance of different fusion strategies through parameter adjustment and provide diversified signal representation for subsequent feature extraction and optimization.

[0229] The specific implementation of step S50 is: extracting features from the multiple groups of fusion signals obtained in step S40 respectively. 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 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 a Fast Fourier Transform (FFT) on the signal:

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

[0240] 3. Extract the center of the spectrum:

[0241] 4. Extract spectrum variance:

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

[0243] Time-frequency domain feature extraction:

[0244] 1. Perform continuous wavelet transform (CWT) on the signal:

[0245] Among them, 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] in,

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

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

[0250] The specific implementation of step S60 is: using a genetic algorithm to optimize the weights of multiple groups of fusion features, and calculating the optimal fusion feature based on the minimum mean square error criterion. First, define the chromosome encoding 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 number of iterations is 200. The specific steps of the 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 diagnostic accuracy as the fitness.

[0253] 3. Selection operation: Roulette selection method is used, and the selection probability is proportional to the fitness.

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

[0255] Child1=α·Parent1+(1-α)·Parent2

[0256] Child2=(1-α)·Parent1+α·Parent2

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

[0258] 5. Mutation operation: Gaussian mutation is used, and the mutation probability is set to 0.01.

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

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

[0261] 6. Elite retention: retain the best 5% of individuals in the current population 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 features:

[0264]

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

[0266] In order to further optimize the fusion features, the minimum mean square error criterion is adopted. Define the objective function:

[0267]

[0268] Where N is the number of samples, y j is the true label of the jth sample, f(F opt,j ) is to use the optimal fusion feature F opt,j The prediction results obtained.

[0269] Use gradient descent to find F opt To make fine adjustments:

[0270]

[0271] Among them, η is the learning rate, the initial value is set to 0.01, and the adaptive learning rate strategy is adopted. is the objective function with respect to F opt gradient.

[0272] The optimization is iterated until one of the following stopping conditions is met:

[0273] 1. Reach the maximum number of iterations (set to 1000).

[0274] 2. The change in the objective function value is less than the preset threshold (such as 10 ；6 ).

[0275] 3. The gradient norm is less than the preset threshold (such as 10 ；5 ).

[0276] The purpose of this step is to obtain a fusion feature that can maximally reflect the state of the circuit breaker through a combination of genetic algorithm and gradient optimization, so as to provide high-quality input for subsequent fault diagnosis.

[0277] The specific implementation of step S70 is: input the optimal fusion feature obtained in step S60 into the 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. Select a multi-layer perceptron (MLP) network as the fault diagnosis model. The network structure is: input layer (the number of nodes is equal to the optimal fusion feature dimension) - fully connected layer (128 neurons) - Dropout layer (discard rate 0.5) - fully connected layer (64 neurons) - Dropout layer (discard rate 0.3) - fully connected layer (32 neurons) - output layer (the number of nodes is 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] Among them, W i and b i are the weight matrix and bias vector of the i-th layer, respectively, and F opt is the optimal fusion feature of the input, and y is the output fault diagnosis vector.

[0285] Then, the model is trained. A historical dataset is used, which contains the operation data of circuit breakers with known fault types. The dataset is divided into training set, validation set and test set in a ratio of 8:1:1. The cross entropy loss function is used:

[0286]

[0287] Where C is the number of fault types, y i is the true label, Predict probabilities for the model.

[0288] The Adam optimizer is used for model training, with the initial learning rate set to 0.001 and the batch size set to 64. The early stopping strategy is used during training, and training is stopped when the validation set loss does not decrease for 10 consecutive epochs. At the same time, the learning rate decay strategy is adopted, and the learning rate is reduced to 0.1 times the original every 50 epochs.

[0289] After the model training is completed, the model performance is evaluated using the test set. Indicators such as accuracy, precision, recall, and F1 score are calculated. If the model performance does not meet the requirements (such as accuracy is less than 95%), the network structure or hyperparameters are adjusted and retrained.

[0290] Finally, the optimal fusion feature obtained in step S60 is input into the trained model to obtain a 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, that is, the highest probability value, is output.

[0291] In order to improve the reliability of diagnosis, a probability threshold is set (such as 0.8). 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 warning threshold (such as 0.3), a warning message will be given in the diagnosis result even if it is not the highest probability.

[0293] The output format of the diagnostic results is:

[0294] 1. Main diagnostic results: Fault type (or "normal" / "suspected fault")

[0295] 2. Diagnostic confidence: highest probability value

[0296] 3. Secondary diagnostic results: other fault types with probabilities exceeding the warning threshold (if any)

[0297] 4. Recommendations: Provide corresponding maintenance or further inspection recommendations based on the diagnosis results

[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 diagnostic results and suggestions to provide 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, determining that the circuit breaker has a potential fault risk; if the fault diagnosis result does not exceed the safety threshold, determining that the circuit breaker is in a normal working state; based on the comparison result of the fault diagnosis result and the safety threshold, calculating the reliability of the circuit breaker and outputting it; wherein the formula for calculating the reliability is specifically: R = e ；λt ;

[0300] in:

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

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

[0303] λ0 is the benchmark failure rate, which can be determined based on historical data or empirical values;

[0304] t is the running time.

[0305] In summary, the reliability diagnosis method of a circuit breaker device proposed in the present invention realizes high-precision diagnosis of mechanical faults of circuit breakers through steps such as multi-sensor signal acquisition, signal preprocessing, multi-method signal fusion, feature extraction and optimization, and deep learning model diagnosis. This method makes full use of the complementary information of audio and vibration signals, improves the expression ability of features through multiple fusion strategies and feature optimization, and uses deep learning models to capture complex fault modes, thereby improving the accuracy and reliability of diagnosis.

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

[0307] 1. Principle of multi-signal source fusion: The audio signal and vibration signal generated by the circuit breaker during operation contain rich state information, but the two signals differ in terms of time-frequency characteristics and noise sensitivity. The present invention adopts a variety of signal fusion methods, such as weighted averaging, adaptive filtering, wavelet coefficient fusion, etc., which can make full use of the complementary information of different signal sources and improve the integrity and reliability of the information. This multi-signal source fusion method effectively reduces the problem of single signal source being susceptible to interference and improves the robustness of diagnosis.

[0308] 2. Principle of adaptive signal processing: The operating environment of the circuit breaker is complex, and the original signal often contains various interferences and noises. The short-time Fourier transform used in the present invention can reflect the time-frequency characteristics of the signal, the adaptive median filter can effectively remove the pulse noise, and the wavelet denoising method can remove the random noise while retaining the useful components of the signal. These adaptive signal processing methods can perform targeted processing according to the characteristics of the signal, significantly improve the signal-to-noise ratio of the signal, and create favorable conditions for subsequent feature extraction.

[0309] 3. Principle of multi-dimensional feature extraction: The failure modes of circuit breakers are complex and diverse, and single-dimensional features are difficult to fully reflect their status. The present invention extracts features from three dimensions: time domain, frequency domain, and time-frequency domain, covering various types such as statistical features, spectral features, and wavelet features. This multi-dimensional feature extraction method can fully capture the operating status information of circuit breakers and improve the expression and differentiation capabilities of features.

[0310] 4. Feature fusion optimization principle: Different features contribute to different degrees to fault diagnosis, and simple feature combinations may lead to redundancy or information loss. The present invention uses a genetic algorithm to optimize feature weights, and uses its global search capability to find a near-optimal weight combination in a complex feature space. At the same time, the fine-tuning process based on the minimum mean square error criterion can further optimize feature representation and improve the feature discrimination capability.

[0311] 5. Principle of deep learning diagnostic model: The complexity and diversity of circuit breaker failure modes require the diagnostic model to have strong nonlinear mapping capabilities. The multi-layer perceptron network used in the present invention has a multi-layer nonlinear transformation structure, which can automatically learn the intrinsic characteristics and complex patterns of data. Through the training of a large amount of historical data, the model can establish a complex mapping relationship between features and fault types, and achieve high-precision fault identification.

[0312] 6. Reliability evaluation principle: The reliability of the diagnosis results is crucial to maintenance decisions. The probability threshold and warning threshold mechanism introduced in the present invention can quantitatively evaluate the diagnosis results based on the probability distribution output by the diagnosis model. This method can not only identify clear fault types, but also point out potential risks and issues that require further attention, providing more comprehensive information support for maintenance decisions.

[0313] In summary, the technical solution of the present invention constructs a logically complete and technologically advanced circuit breaker fault diagnosis system 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.

[0314] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: There are multiple circuit breakers in a 330kV substation. In order to improve the operating 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 invention for intelligent transformation. The following describes in detail the complete process of fault diagnosis of one of the ZN65-12 vacuum circuit breakers.

[0315] 1. Sensor installation and data collection

[0316] First, highly sensitive audio sensors and vibration sensors were installed at key locations on the circuit breaker. The audio sensor used a condenser microphone with a frequency response range of 20Hz-20kHz and a sensitivity of -38dB±3dB (0dB=1V / Pa, 1kHz). The vibration sensor used a triaxial piezoelectric accelerometer 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 high-speed data acquisition card with 16-bit resolution via 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 data continuity.

[0318] When the circuit breaker is opening and closing, the audio signal and vibration signal are collected for 10 seconds. The original signal is marked with a timestamp and saved as a binary file for subsequent processing.

[0319] 2. Signal preprocessing

[0320] The collected original signal is preprocessed to improve the signal quality.

[0321] For audio signals, short-time Fourier transform (STFT) is first performed. The Hanning window is selected, the window length is set to 1024 sampling points, and the overlap rate is 50%. Then the STFT result is adaptively median filtered, and the initial window size is set to 3×3, which can be expanded to 7×7.

[0322] For vibration signals, wavelet denoising method is used. The db4 wavelet base is selected and the decomposition level is 4. The threshold is determined using the unbiased risk estimation principle, and the wavelet coefficients are soft-thresholded. Then envelope analysis is performed, and the obtained envelope is smoothed using a moving average filter with a length of 1% of the signal length.

[0323] 3. Signal Fusion

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

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

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

[0327] (3) Wavelet coefficient fusion: Use db4 wavelet with 4 decomposition levels.

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

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

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

[0331] 4. Parameter adjustment and feature extraction

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

[0333] After optimization, six groups of fusion signals were obtained. The following features were extracted for each group of fusion signals:

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

[0335] Frequency domain features: The power spectral density is calculated by FFT, the spectrum center and spectrum variance are extracted, and the energy ratio of three key frequency bands (0-500Hz, 500-2000Hz, 2000-5000Hz) is calculated.

[0336] Time-frequency domain features: Morlet wavelet is used to perform continuous wavelet transform, calculate wavelet energy entropy, and extract statistical features of wavelet coefficients at four scales (mean, variance, skewness, and kurtosis).

[0337] Finally, each set of fused signals obtained a feature vector containing 28 features. Dimensionality reduction was performed through principal component analysis, and the principal components that explained 95% of the variance were retained, resulting in 6 sets of fusion features after dimensionality reduction.

[0338] 5. Feature fusion optimization

[0339] The genetic algorithm is used to optimize the weights of the six groups of fusion features. The population size is set to 100 and the number of iterations is 200. The crossover probability is set to 0.8 and the mutation probability is set to 0.01. The fitness function is defined as the accuracy of preliminary fault diagnosis using weighted features.

[0340] After optimization, the optimal weights are: 0.22, 0.18, 0.25, 0.15, 0.12, 0.08. The optimal fusion features are calculated using these weights. Then, based on the minimum mean square error criterion, the gradient descent method is used to fine-tune the optimal fusion features. The initial value of the learning rate is set to 0.01, and the adaptive learning rate strategy is adopted. After 1000 iterations, the final optimized fusion features are obtained.

[0341] 6. Fault diagnosis model training

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

[0343] The model is trained using a historical dataset. The dataset contains 5,000 samples, covering seven common circuit breaker mechanical fault types and normal states. The dataset is divided into training set, validation set, and test set in a 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 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 to reduce the learning rate to 0.1 times 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] index value Accuracy 97.8％ Average precision 97.6％ Average recall 97.5％ Average 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 above steps, and preprocessing, signal fusion, feature extraction and optimization are performed. 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 Operating mechanism stuck 0.01 Spring failure 0.03 Abnormal buffer device 0.89 Contact wear 0.02 Vacuum reduction 0.01 Operation circuit failure 0.01 Insulator damage 0.01

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

[0354] 8. Security identification analysis and suggestions

[0355] The safety threshold is set to 0.8. 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 failure risk. The reliability is calculated as follows:

[0356] Based on the information provided, the reliability of the circuit breaker can be calculated as follows:

[0357] Calculate the failure rate λ:

[0358]

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

[0360]

[0361] Assuming the running time t = 10 hours, 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 to generate preventive maintenance recommendations:

[0365] (1) Immediately arrange for professionals to check the buffer device of the circuit breaker, focusing on the buffer spring, buffer cylinder and other components.

[0366] (2) During the inspection and 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 time.

[0369] (5) After repair, perform opening and closing tests and use this diagnostic system for verification again.

[0370] 9. Verification of diagnostic effect

[0371] The maintenance personnel inspected the circuit breaker according to the system's recommendations. The inspection found that the buffer oil in the buffer oil cylinder had aged severely and the viscosity had decreased, resulting in poor buffering effect. After replacing the new buffer oil, the circuit breaker was tested for opening and closing, and the operation was normal.

[0372] The diagnostic system was used to perform the test again, and the diagnostic results obtained are shown in Table 3:

[0373] Table 3 Re-diagnosis results

[0374] Fault type Probability normal 0.97 Operating mechanism stuck 0.01 Spring failure 0.00 Abnormal buffer device 0.01 Contact wear 0.00 Vacuum reduction 0.00 Operation circuit failure 0.01 Insulator damage 0.00

[0375] The diagnosis result shows that the circuit breaker has returned to a normal state, with a confidence level of 97%. The circuit breaker fault diagnosis method of the present invention significantly improves the operation reliability of the circuit breaker.

[0376] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A reliability diagnosis method for a circuit breaker device, characterized in that: The following steps are involved: S10, continuously acquiring audio signals and vibration signals during the operation of the circuit breaker; S20, performing short-time Fourier transform and adaptive median filtering on the audio signal to obtain a first audio signal; performing wavelet denoising and envelope analysis on the vibration signal to obtain a first vibration signal; S30, fusing the audio signal and the vibration signal aligned in time using a preset fusion equation group to obtain a fusion signal, wherein the fusion equation group includes: 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 the parameters of the fusion equation group to obtain multiple groups of fusion signals; S50, performing feature extraction on the multiple groups of fusion signals respectively to obtain multiple groups of fusion features; S60, optimizing the weights of the multiple groups of fusion features by using a genetic algorithm, and calculating the optimal fusion feature by using the multiple groups of fusion features based on a minimum mean square error criterion; 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; S80. Compare the fault diagnosis result with a preset safety threshold. If the fault diagnosis result exceeds the safety threshold, determine that the circuit breaker has a potential fault risk; if the fault diagnosis result does not exceed the safety threshold, determine that the circuit breaker is in a normal working state; based on the comparison result of the fault diagnosis result and the safety threshold, calculate the reliability of the circuit breaker and output it.

2. A reliability diagnosis method for circuit breaker equipment according to claim 1, characterized in that: The weighted average fusion equation is specifically expressed as: F(t)=w a A(t)+w v V(t); Where 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 audio and vibration signals respectively, satisfying w a +w v =1.

3. A reliability diagnosis method for circuit breaker equipment according to claim 2, characterized in that: The adaptive filtering fusion equation is specifically expressed as: Where F(n) is the discrete time signal after fusion; h ξ (n) is the coefficient of the adaptive filter; M is the filter order; w a and w v is the weight coefficient.

4. A reliability diagnosis method for circuit breaker equipment according to claim 3, characterized in that: The wavelet coefficient fusion equation is specifically expressed as: F=W -1 (φ(W A (A),W V (V))); Where W A and W V Respectively represent the wavelet transform of audio and vibration signals; W -1 represents inverse wavelet transform; φ is the fusion function of wavelet coefficients.

5. A reliability diagnosis method for circuit breaker equipment according to claim 4, characterized in that: The cross-correlation fusion equation is specifically expressed as: Or in discrete form: Where N is the cross-correlation window size, which serves as an adjustment parameter.

6. A reliability diagnosis method for circuit breaker equipment according to claim 5, characterized in that: The energy ratio fusion equation is specifically expressed as: Where E A and E V The energy of audio and vibration signals respectively:

7. A reliability diagnosis method for circuit breaker equipment according to claim 6, characterized in that: The principal component analysis fusion equation is specifically expressed as: Where [A, V] is the matrix that combines the audio and vibration signals; U k ,Σ k ,V k It is the matrix obtained by taking the first k principal components after singular value decomposition of [A,V]; k is the number of principal components.

8. A reliability diagnosis method for circuit breaker equipment according to claim 7, characterized in that: The training steps of the circuit breaker mechanical fault diagnosis model are specifically: Establishing a training data set, specifically, following steps S10-S60 to obtain optimal fusion features of multiple groups of circuit breakers with known mechanical faults; Model training specifically involves establishing a multi-layer perceptron network and using the training data set for training, wherein the training input is the optimal fusion feature of each group of circuit breakers with known mechanical faults, and the training output is the corresponding known mechanical faults, thereby obtaining a circuit breaker mechanical fault diagnosis model.

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

10. A switch safety identification system for a circuit breaker device, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images