High-voltage circuit breaker fault diagnosis method based on image partitioning method and vibration-sound combination

Abstract

The invention provides a high-voltage circuit breaker fault diagnosis method based on an image partitioning method and vibration-sound combination, and relates to the technical field of fault diagnosis. The method comprises the following steps: performing short-time Fourier transform and graying processing on a vibration signal and a sound signal of the high-voltage circuit breaker, and partitioning; extracting a first feature and a second feature of each small picture, and constructing a feature data set; performing feature screening by using a Laplace score method; and inputting the optimal feature subset into a support vector machine fault diagnosis model, taking the highest classification accuracy as an objective function, combining with a dung beetle optimization algorithm, outputting preliminary diagnosis results, performing decision-making layer fusion on the preliminary diagnosis results by using a subjective Bayesian network, and outputting a final diagnosis result of the high-voltage circuit breaker. According to the method, the vibration signal and the sound signal of the high-voltage circuit breaker are processed in an image partitioning mode, and the two signals are subjected to combined diagnosis analysis by adopting an information fusion method, so that the fault diagnosis precision of the high-voltage circuit breaker is improved.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method for high-voltage circuit breakers based on image segmentation and vibration-acoustic combination. Background Technology

[0002] Against the backdrop of the accelerated construction of new power systems and the digital transformation of the power industry, high-voltage circuit breakers, as core equipment for the safe operation of the power grid, have made fault diagnosis technology a crucial link in ensuring the reliability of the power system. High-voltage circuit breaker fault diagnosis not only directly relates to the state awareness capability of power grid equipment but also serves as an important technical support for preventing major power accidents and achieving full life-cycle management of equipment, significantly contributing to improving the intelligent level of power grid operation and maintenance. Currently, commonly used methods for high-voltage circuit breaker fault diagnosis include traditional threshold methods, single-signal diagnostic methods, and multi-source information fusion diagnostic methods. However, traditional threshold methods often suffer from low sensitivity and high false alarm rates, resulting in insufficient diagnostic accuracy. While single-signal diagnostic methods can improve the accuracy of high-voltage circuit breaker fault diagnosis by extracting signal features and utilizing classification models, they lack anti-interference capabilities and struggle to handle feature extraction for progressive faults. Summary of the Invention

[0003] This application provides a high-voltage circuit breaker fault diagnosis method based on image segmentation and vibration-sound combination to solve the problems of poor diagnostic accuracy of traditional threshold method and insufficient anti-interference capability of single signal diagnosis method. It uses image segmentation to process the vibration signal and sound signal of high-voltage circuit breaker, and adopts information fusion method to combine the advantages of vibration signal and sound signal, and jointly analyzes the two signals to improve the accuracy and robustness of high-voltage circuit breaker fault diagnosis.

[0004] This application provides a high-voltage circuit breaker fault diagnosis method based on image block method, characterized in that the method includes:

[0005] Acquire vibration and sound signals from high-voltage circuit breakers;

[0006] Short-time Fourier transforms are performed on the vibration signal and the sound signal respectively to obtain two-dimensional time-frequency diagrams of the vibration signal and the sound signal. Then, each two-dimensional time-frequency diagram is grayscaled to obtain grayscaled two-dimensional time-frequency diagrams.

[0007] The grayscale two-dimensional time-frequency image is divided into blocks using the image block method to obtain multiple small images of equal size.

[0008] Extract the first and second features of each of the small images to construct a feature dataset; wherein, the first feature is the feature of the gray-level-gradient co-occurrence matrix, and the second feature is the feature of the Gaussian-Markov random field;

[0009] The Laplace score method is used to filter features in the feature dataset and construct the optimal feature subset.

[0010] The optimal feature subset is input into the support vector machine fault diagnosis model. The objective function is to maximize classification accuracy. The model parameters are then evaluated to determine if they meet the iteration conditions. If they do, the diagnostic results for the vibration or sound signal are output. If they do not meet the iteration conditions, the dung beetle optimization algorithm is used to optimize the penalty factor c and kernel parameter g of the support vector machine. The optimized parameters are then input into the support vector machine fault diagnosis model for model update. The updated support vector machine fault diagnosis model is then re-processed with the optimal feature subset, and the iteration conditions are evaluated again. The iteration conditions are defined as either a classification accuracy reaching a preset threshold or a set number of iterations.

[0011] Furthermore, the formula for the short-time Fourier transform is as shown in equation (1):

[0012]

[0013] In the formula, (τ-t) is a window function centered at τ; x(τ) is the time-domain signal to be processed; STFT(t,f) is the output of the short-time Fourier transform, e is the natural constant, approximately equal to 2.718, j is the imaginary unit, f is the frequency, and τ is the time variable;

[0014] Furthermore, the features of the gray-level-gradient co-occurrence matrix include gray-level average, gradient average, gray-level standard deviation, gradient standard deviation, small gradient advantage, large gradient advantage, non-uniform gray-level distribution, non-uniform gradient distribution, energy, correlation, gray-level entropy, gradient entropy, mixed entropy, inertia and / or inverse gap.

[0015] Furthermore, the second feature is extracted in the following manner:

[0016] A second feature is extracted from the small image based on a Gaussian-Markov random field; wherein:

[0017] The conditional probability formula for the Gauss-Markov random field is:

[0018] P(y(m)|y(m+E),E∈G) (2)

[0019] In the formula, y(m) is the pixel value of any pixel m in the image, P represents the probability that the value of the center pixel m is y(m) given the values ​​of the neighboring pixels, y(m+E) is the pixel value of a certain pixel m+E in G, G is a symmetric neighborhood with m as the center and E as the radius but not including m, and E is the radius of the symmetric neighborhood G.

[0020] The model formula for the Gauss-Markov random field is as follows:

[0021]

[0022] In the formula, y(S) is the image on the point set S, and G... m Let S be the neighborhood of a Gaussian-Markov random field with m pixels, S be the set of points on a W×W image patch, y1(m+E) be a point in S, and θ be a point in S. E is the weight of the symmetrical neighboring pixels, and e(m) is Gaussian noise with a mean of 0;

[0023] Substituting the points on the small image into equation (3) yields:

[0024]

[0025] In the formula, Let be the transpose of the matrix y(m); θ is the eigenvector to be estimated in the model;

[0026] The following equation (4) is obtained by estimating and solving using the least squares error criterion:

[0027]

[0028] In the formula, T u To extract features from the Gaussian-Markov random field, Z m It is a matrix about y(m).

[0029] Furthermore, when using the Laplace score method to filter features in the feature dataset and construct the optimal feature subset, the weight matrix expression corresponding to the adjacency matrix in the Laplace score method is:

[0030]

[0031] In the formula, x h and x j They are the h-th and j-th nodes in the adjacency matrix, where e is the natural constant and S is the adjacency matrix. hj To evaluate the weight matrix of the similarity between the h-th node and the j-th node, G hj Let h×j be the adjacency matrix;

[0032] The formula for calculating the Laplace fraction of the r-th feature is:

[0033]

[0034] In the formula, L r Let f represent the Laplace score of the r-th feature. rh and f rj Var(f) represents the r-th feature of the h-th sample and the j-th sample.r ) represents the variance of the r-th feature relative to all samples.

[0035] Furthermore, the support vector machine fault diagnosis model is expressed as:

[0036]

[0037] In the formula, w is the normal vector of the optimal hyperplane; b is the threshold; y h The class label of the h-th sample is represented by T, which is the matrix transpose; h represents the sample index; N represents the number of samples; x h Let represent the classification feature of the h-th sample, st represent the constraint condition, and min represent the minimum value function;

[0038] Based on the aforementioned support vector machine fault diagnosis model, after introducing a penalty factor and a Lagrange operator, we obtain:

[0039]

[0040] In the formula, max is the maximum value function; α is the Lagrange operator; α h It is the Lagrange operator for the h-th sample; α j It is the Lagrange operator for the j-th sample; y j x is the classification label of the j-th sample; j is the classification feature of the j-th sample; C is the penalty factor;

[0041] Furthermore, a kernel function is introduced, which is a Gaussian radius basis kernel function, expressed as:

[0042]

[0043] In the formula, K is the kernel function, exp is an exponential function with the natural constant as the base, and g is the parameter to be optimized in the Gaussian radial basis kernel function;

[0044] Substituting equation (10) into the solution, we obtain the following formula for the linear classification function:

[0045]

[0046] In the formula, f(x) is the classification function, sgn is the sign of the score calculated by the function, and the final classification result is output accordingly.

[0047] Furthermore, in the dung beetle optimization algorithm, the position update formula for the rolling ball group in the obstacle-free mode is as follows:

[0048]

[0049] In the formula, t represents the current iteration number; Let represent the position of the i-th dung beetle in the population at the t-th iteration; k, β, and b are all constants; Indicates the worst position in the population; This represents the position of the i-th dung beetle in the population at the (t+1)-th iteration;

[0050] The position update formula for the rolling ball group with obstacles is:

[0051]

[0052] In the formula, μ is the angle at which the dung beetle changes its posture to obtain a new direction when it encounters an obstacle and cannot move forward. This represents the position of the i-th dung beetle in the population during the (t-1)th iteration.

[0053] The formula for updating the position of chicks in a breeding population is:

[0054]

[0055] In the formula, Ub represents the position of the κ-th primordial ball in the t-th iteration; * and Lb * These are the upper and lower bounds of the spawning region; b1 and b2 represent independent random vectors of size 1×D; D represents the dimension optimized in the support vector machine. Indicates the globally optimal position;

[0056] The formula for updating the location of foraging groups is:

[0057]

[0058] In the formula, C1 is a random number following a normal distribution, and C2 is a random vector of size 1×D between (0,1); Lb l and Ub l These are the upper and lower limits of the foraging area;

[0059] The formula for updating the location of the feces-stealing group is:

[0060]

[0061] In the formula, F is a constant value, and z represents a normal distribution. It is the optimal position for the current population.

[0062] Secondly, this application provides a vibration-acoustic combined fault diagnosis method for high-voltage circuit breakers, the vibration-acoustic combined fault diagnosis method for high-voltage circuit breakers comprising:

[0063] Acquire vibration signals from one channel and sound signals from three channels;

[0064] The vibration signal from one channel and the sound signals from three channels are preprocessed to obtain a preprocessed signal;

[0065] The high-voltage circuit breaker fault diagnosis method based on image segmentation, as described above, is used to process the preprocessed signals to obtain preliminary diagnostic results for each signal.

[0066] The preliminary diagnostic results are fused at the decision level using a subjective Bayesian network to output the final diagnostic results for the high-voltage circuit breaker.

[0067] Furthermore, the vibration signal from one channel and the audio signals from three channels are preprocessed to obtain the preprocessed signal, including the following methods:

[0068] The vibration signal of one channel is denoised to obtain a denoised vibration signal;

[0069] The audio signals from the three channels are denoised using wavelet packets and then reconstructed to obtain the reconstructed signal.

[0070] The denoised vibration signal and the reconstructed signal are used as preprocessed signals.

[0071] Furthermore, the subjective Bayesian network is based on Bayes' theorem, which is:

[0072]

[0073] In the formula, P(B) u () represents the u-th event B u The prior probability; P(A|B) u ) is the u-th event B u The probability of event A occurring given the given conditions; P(B) u |A) is the u-th event B given that event A has occurred. u The probability of occurrence; n is the number of events;

[0074] Knowledge can be represented in the following ways:

[0075] if A then(LS,LN)B (18)

[0076] In the formula, LS and LN represent the sufficiency and necessity of the rule being valid, respectively, and are defined as follows:

[0077]

[0078] By introducing a probability function and establishing its probability relationship with the occurrence of event B, the posterior probability formula is derived as follows:

[0079]

[0080] In the formula, P(B) is the probability of event B occurring.

[0081] The high-voltage circuit breaker fault diagnosis method based on image segmentation and vibration-acoustic combination provided in this application has at least the following beneficial effects:

[0082] This application utilizes image segmentation to process vibration and sound signals from high-voltage circuit breakers, and employs an information fusion method to combine the advantages of vibration and sound signals, performing joint diagnostic analysis of the two signals to improve the accuracy and robustness of high-voltage circuit breaker fault diagnosis. Attached Figure Description

[0083] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0084] Figure 1 A data processing flowchart based on an optimized support vector machine diagnostic model using dung beetles is provided in this application embodiment;

[0085] Figure 2 A flowchart of a high-voltage circuit breaker fault diagnosis method based on image segmentation provided in this application embodiment;

[0086] Figure 3 A schematic diagram of STFT conversion and grayscale conversion of vibration and acoustic signals provided in an embodiment of this application;

[0087] Figure 4 The image processing effect of vibration and acoustic signal image block method under the 3×2 block method provided in the embodiment of this application is shown in the figure.

[0088] Figure 5 The flowchart of a vibration-acoustic combined high-voltage circuit breaker fault diagnosis method provided in the embodiments of this application Figure 1 ;

[0089] Figure 6 The flowchart of a vibration-acoustic combined high-voltage circuit breaker fault diagnosis method provided in the embodiments of this application Figure 2 ;

[0090] Figure 7 This is a flowchart illustrating the multi-image information fusion process provided in an embodiment of this application.

[0091] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0092] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0093] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0094] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0095] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0096] Example 1:

[0097] This application provides a high-voltage circuit breaker fault diagnosis method based on image segmentation. For example... Figure 1 The diagram shows a data processing flowchart of a dung beetle-based optimized support vector machine diagnostic model provided in this embodiment of the application. This embodiment provides a dung beetle-based optimized support vector machine diagnostic model, which can be used to implement the high-voltage circuit breaker fault diagnosis method based on image segmentation. It takes vibration and sound signals as inputs, performs short-time Fourier transform and grayscale processing, then performs image segmentation. Based on the image segmentation results, 27 features are extracted, and the Laplace score method is used to select features from these 27 features. The selected features are input into the SVM fault diagnosis model, and it is determined whether the iteration conditions are met. If the iteration conditions are not met, the parameters of the SVM fault diagnosis model are optimized using the dung beetle algorithm, and the segmentation is repeated until the iteration conditions are met, at which point the diagnostic result is output. The iteration conditions refer to whether the classification accuracy reaches a set value or whether the number of iterations reaches a set number.

[0098] Specifically, such as Figure 2The diagram shown is a flowchart of a high-voltage circuit breaker fault diagnosis method based on image segmentation provided in an embodiment of this application. The high-voltage circuit breaker fault diagnosis method based on image segmentation includes the following steps S201-S208.

[0099] S201: Acquire vibration and sound signals from the high-voltage circuit breaker.

[0100] S202: Perform short-time Fourier transform on the vibration signal and the sound signal respectively to obtain two-dimensional time-frequency graphs of the vibration signal and the sound signal, and perform grayscale processing on each two-dimensional time-frequency graph to obtain grayscale two-dimensional time-frequency graphs.

[0101] The Short-Time Fourier Transform (STFT) is a method for converting a one-dimensional signal into a two-dimensional time-frequency graph to obtain the signal's time and frequency characteristics. The STFT first divides the signal into multiple segments of the same length using a window function, then performs a Fast Fourier Transform on the time signal within each window. The resulting set of spectral functions for each time window is then concatenated to obtain a final two-dimensional time-frequency graph. The transformation formula is shown below:

[0102]

[0103] In the formula, (τ-t) is a window function centered at τ; x(τ) is the time-domain signal to be processed; STFT(t,f) is the output of the short-time Fourier transform, e is the natural constant, approximately equal to 2.718, j is the imaginary unit, f is the frequency, and τ is the time variable;

[0104] like Figure 3 The diagram shown is an exemplary illustration of the STFT conversion and grayscale conversion of vibration and acoustic signals in this embodiment.

[0105] S203: The grayscale two-dimensional time-frequency image is divided into blocks using the image block method to obtain multiple small images of equal size.

[0106] Image segmentation (IS) is a signal processing method that facilitates the extraction of local signal features and improves the richness of model feature extraction. It involves randomly dividing an image into M×N small images of equal size. In this embodiment, image segmentation is used to segment the time-frequency diagrams of vibration and acoustic signals into M×N equal-sized blocks, where M is the number of horizontal blocks and N is the number of vertical blocks. Taking a 3×2 block as an example... Figure 4 The image shown is a block diagram of the vibration and acoustic signal image processing effect under the 3×2 block method.

[0107] S204: Extract the first and second features of each small image to construct a feature dataset.

[0108] In this embodiment, the first feature is the feature of the gray-level-gradient co-occurrence matrix, and the second feature is the feature of the Gauss-Markov random field.

[0109] The gray-gradient co-occurrence matrix (GGCM) reflects the relationship between the gray level and gradient of each pixel in an image. For example, element H(i,j) represents an element in the image with gray level i and gradient j, and the H value indicates the number of pixels with the same gray level. P(i,j) represents the probability of obtaining the total number of pixels in the image after normalization. The gray-gradient co-occurrence matrix obtained from the multiple small images processed in step S203 contains a total of 15 features, as shown in Table 1.

[0110] Table 1 Gray-level-gradient co-occurrence matrix

[0111]

[0112] A Gaussian Markov random field (GMRF) is a probabilistic model for describing the structure of graphics and the texture structure of images. The conditional probability of texture features in an image is represented by GMRF as shown in equation (2):

[0113] P(y(m)|y(m+E),E∈G) (2)

[0114] In the formula, y(m) is the pixel value of any pixel m in the image, P represents the probability that the value of the center pixel m is y(m) given the values ​​of the neighboring pixels, y(m+E) is the pixel value of a certain pixel m+E in G, G is a symmetric neighborhood with m as the center and E as the radius but not including m, and E is the radius of the symmetric neighborhood G.

[0115] The model formula for the Gauss-Markov random field is as follows:

[0116]

[0117] In the formula, y(S) is the image on the point set S, and G... m Let S be the neighborhood of a Gaussian-Markov random field with m pixels, S be the set of points on a W×W image patch, y1(m+E) be a point in S, and θ be a point in S. E is the weight of the symmetrical neighborhood pixels, and e(m) is Gaussian noise with a mean of 0;

[0118] Substituting the points on the small image into equation (3) yields:

[0119]

[0120] In the formula, Let be the transpose of the matrix y(m); θ is the eigenvector to be estimated in the model;

[0121] The following equation (4) is obtained by estimating and solving using the least squares error criterion:

[0122]

[0123] In the formula, T u To extract features from the Gaussian-Markov random field, Z m It is a matrix about y(m).

[0124] S205: Use the Laplace score method to filter features in the feature dataset and construct the optimal feature subset.

[0125] The Laplacian Score (LS) assesses feature importance by measuring the difference between a feature and other features in the dataset that have a similar distribution to the target feature. A smaller Laplacian score indicates greater importance of the feature in the feature set.

[0126] When performing feature selection, LS first constructs an m×m adjacency matrix G. The h-th node in the matrix corresponds to x. h When x h and x j When the types are the same, G hj =1, otherwise 0, to obtain the weight matrix S used to evaluate the similarity between two nodes, the expression of which is shown in equation (6):

[0127]

[0128] In the formula, x h and x j They are the h-th and j-th nodes in the adjacency matrix, where e is the natural constant and S is the adjacency matrix. hj To evaluate the weight matrix of the similarity between the h-th node and the j-th node, G hj Let h×j be the adjacency matrix;

[0129] The formula for calculating the Laplace fraction of the r-th feature is:

[0130]

[0131] In the formula, L r Let f represent the Laplace score of the r-th feature. rh and f rj Var(f) represents the r-th feature of the h-th sample and the j-th sample. r ) represents the variance of the r-th feature relative to all samples.

[0132] Substitute the 27 features listed in step S204 into the above formula for analysis, and then sort them according to the Laplace score obtained by formula (7). Select the top-ranked features as the optimal features. The number of optimal features can be preset, for example, it can be 8, 10, 12, etc. This embodiment does not specifically limit it here. Multiple optimal features form an optimal feature subset.

[0133] S206: Input the optimal feature subset into the support vector machine fault diagnosis model, take the highest classification accuracy as the objective function, and determine whether the model parameters meet the iteration conditions; if they meet the iteration conditions, proceed to step S207; if they do not meet the iteration conditions, proceed to step S208.

[0134] In this embodiment, the iteration condition is that the classification accuracy reaches a preset threshold or the number of iterations reaches a set number.

[0135] The process of fault diagnosis implemented by the support vector machine fault diagnosis model is as follows:

[0136] Support Vector Machine (SVM) is a classification algorithm based on a linear classification model with the objective function of maximizing the classification margin. Its algorithmic idea is to construct an optimal hyperplane to maximize 2 / ||w||

[8175] . 2 / ||w|| refers to the distance between two optimal hyperplanes, and its mathematical model is shown in equation (8):

[0137]

[0138] In the formula, w is the normal vector of the optimal hyperplane; b is the threshold; y h The class label of the h-th sample is represented by T, which is the matrix transpose; h represents the sample index; N represents the number of samples; x h Let represent the classification feature of the h-th sample, st represent the constraint condition, and min represent the minimum value function;

[0139] Based on the aforementioned support vector machine fault diagnosis model, after introducing a penalty factor and a Lagrange operator, we obtain:

[0140]

[0141] In the formula, max is the maximum value function; α is the Lagrange operator; α h It is the Lagrange operator for the h-th sample; α j It is the Lagrange operator for the j-th sample; y j x is the classification label of the j-th sample; j is the classification feature of the j-th sample; C is the penalty factor;

[0142] Furthermore, a kernel function is introduced, which is a Gaussian radius basis kernel function, expressed as:

[0143]

[0144] In the formula, K is the kernel function, exp is an exponential function with the natural constant as the base, and g is the parameter to be optimized in the Gaussian radial basis kernel function;

[0145] Substituting equation (10) into the solution, we obtain the following formula for the linear classification function:

[0146]

[0147] In the formula, f(x) is the classification function, sgn is the sign of the score calculated by the function, and the final classification result is output accordingly.

[0148] S207: Diagnostic results of output vibration or sound signals.

[0149] S208: The dung beetle optimization algorithm is used to optimize the penalty factor c and kernel parameter g of the support vector machine. The optimized parameters are input into the support vector machine fault diagnosis model for model update. Based on the updated support vector machine fault diagnosis model, step S203 is executed again.

[0150] The Dung Beetle Optimizer (DBO) is a heuristic optimization algorithm with faster convergence and stronger optimization ability compared to existing optimization algorithms. By dividing the dung beetle population into four groups in a ratio of 5:7:7:6, the groups perform four behaviors: rolling balls, reproduction, foraging, and dung stealing, respectively [82-83].

[0151] Among them, the group responsible for rolling the ball will have two modes in the rolling process, one is the barrier-free mode. In the barrier-free mode, the dung beetle rolling process will be affected by the light intensity, and the position update formula for the rolling process is as shown in equation (12):

[0152]

[0153] In the formula, t represents the current iteration number; Let represent the position of the i-th dung beetle in the population at the t-th iteration; k, β, and b are all constants; Indicates the worst position in the population; This represents the position of the i-th dung beetle in the population at the (t+1)-th iteration;

[0154] The position update formula for the rolling ball group with obstacles is:

[0155]

[0156] In the formula, μ is the angle at which the dung beetle changes its posture to obtain a new direction when it encounters an obstacle and cannot move forward. This represents the position of the i-th dung beetle in the population during the (t-1)th iteration.

[0157] The formula for updating the position of chicks in a breeding population is:

[0158]

[0159] In the formula, Ub represents the position of the κ-th primordial ball in the t-th iteration; * and Lb * These are the upper and lower bounds of the spawning region; b1 and b2 represent independent random vectors of size 1×D; D represents the dimension optimized in the support vector machine. Indicates the globally optimal position;

[0160] The formula for updating the location of foraging groups is:

[0161]

[0162] In the formula, C1 is a random number following a normal distribution, and C2 is a random vector of size 1×D between (0,1); Lb l and Ub l These are the upper and lower limits of the foraging area;

[0163] The formula for updating the location of the feces-stealing group is:

[0164]

[0165] In the formula, F is a constant value, and z represents a normal distribution. It is the optimal position for the current population.

[0166] The dung beetle algorithm continuously updates the positions of the dung beetles, determines whether each target exceeds the boundary, and iteratively searches for the optimal solution that meets the fitness function. This paper sets the dung beetle population size to 25 and the maximum number of iterations to 200.

[0167] This embodiment uses the dung beetle algorithm to optimize the penalty factor c and kernel parameter g of the support vector machine, where the value of c ranges from [0.1, 10] and the value of g ranges from [0.03, 256]. Using SVM classification accuracy as the fitness function, the optimal parameter combination is determined, and the optimal parameters are used for fault diagnosis of vibration and sound signals. The optimal parameter combination obtained through the dung beetle algorithm is: c = 9.3209, g = 0.3905, at which point the model achieves the best diagnostic effect.

[0168] Example 2:

[0169] This application provides a vibration-acoustic combined fault diagnosis method for high-voltage circuit breakers, such as... Figure 5 As shown, this vibration-sound combined high-voltage circuit breaker fault diagnosis method acquires vibration signals from one channel and sound signals from three channels, where the three sound signals are designated as sound signal 1, sound signal 2, and sound signal 3. These vibration and sound signals are then input into a dung beetle-based optimized support vector machine diagnostic model. The flowchart of the dung beetle-based optimized support vector machine diagnostic model is shown below. Figure 1 As shown, the dung beetle-based optimized support vector machine diagnostic model yields four classification results: classification result 1, classification result 2, classification result 3, and classification result 4. Subjective Bayesian fusion diagnosis is then performed on the four classification results to finally output the diagnostic result.

[0170] Specifically, such as Figure 6 As shown, the vibration-acoustic combined high-voltage circuit breaker fault diagnosis method includes the following steps S601-S604.

[0171] S601: Acquires vibration signals from one channel and sound signals from three channels;

[0172] S602: Preprocess the vibration signal from one channel and the audio signal from three channels to obtain the preprocessed signal;

[0173] In some embodiments, the preprocessing of a vibration signal from one channel and an audio signal from three channels to obtain a preprocessed signal includes: denoising the vibration signal from one channel to obtain a denoised vibration signal; reconstructing the audio signal from three channels after denoising with wavelet packets to obtain a reconstructed signal; and using the denoised vibration signal and the reconstructed signal as the preprocessed signal.

[0174] S603: A high-voltage circuit breaker fault diagnosis method based on image segmentation is adopted to process the preprocessed signals and obtain the preliminary diagnosis results of each signal.

[0175] It should be noted that the specific implementation process of the high-voltage circuit breaker fault diagnosis method based on image segmentation is described in Example 1, and will not be repeated here.

[0176] S604: Use a subjective Bayesian network to perform decision-level fusion of the preliminary diagnostic results and output the final diagnostic results of the high-voltage circuit breaker.

[0177] In this embodiment, the subjective Bayesian network in step S604 is based on the subjective Bayesian decision-making method, which is based on Bayes' formula in probability theory, assuming events B1, B2, ... B n Mutually exclusive, B1∪B2∪…∪B n =Ω, event A and events B1, B2, ... B nThey are mutually independent, and P(A) > 0, P(B) > 0. i )>0, i=1,2,…,n, then:

[0178]

[0179] In the formula, P(B) u () represents the u-th event B u The prior probability; P(A|B) u ) is the u-th event B u The probability of event A occurring given the given conditions; P(B) u |A) is the u-th event B given that event A has occurred. u The probability of occurrence; n is the number of events;

[0180] Knowledge can be represented in the following ways:

[0181] if A then(LS,LN)B (18)

[0182] In the formula, LS and LN represent the sufficiency and necessity of the rule being valid, respectively, and are defined as follows:

[0183]

[0184] By introducing a probability function and establishing its probability relationship with the occurrence of event B, the posterior probability formula is derived as follows:

[0185]

[0186] In the formula, P(B) is the probability of event B occurring.

[0187] In practical applications, both LS and LN are derived from expert experience, so when dealing with uncertain derivation problems, it is only necessary to know the prior probability P(B). i That's it. Since the data collected by the four sensors in this embodiment can each independently identify nine types of faults, the prior probability set in this embodiment is 1 / 9.

[0188] In some embodiments, one implementation of the vibration-sound combined high-voltage circuit breaker fault diagnosis method is as follows: Noise is removed from the sound signal using a Fast-ICA and VMD-WPD combined model, and the one-dimensional vibration and sound signals are converted into a two-dimensional grayscale time-frequency image using the STFT algorithm. Next, the grayscale image is divided into M×N equal-sized small images using an image segmentation method, and 27 features are extracted from each small image based on the gray-level-gradient co-occurrence matrix and Gauss-Markov random field. Then, the optimal feature subset is selected using the Laplace score method. Next, the feature subset is input into an SVM classification model for preliminary fault diagnosis, and the dung beetle optimization algorithm is used to determine four parameters in the model: the number of horizontal blocks M, the number of vertical blocks N, the penalty factor c of the support vector machine, and the kernel parameter g, to further improve the recognition accuracy of each preliminary diagnosis model. Finally, a subjective Bayesian network is used to fuse the preliminary diagnosis results of the vibration and sound signals at the decision layer to obtain the final diagnosis result.

[0189] In some embodiments, such as Figure 7 As shown, the vibration-sound combined high-voltage circuit breaker fault diagnosis method, in its specific implementation, first processes three sound signals and one vibration signal, where the three sound signals are sound signal 1, sound signal 2, and sound signal 3. For the sound signals, environmental noise is first added to simulate the substation noise environment. Then, environmental noise is removed using Fast-ICA, RIME-VMD, wavelet packet denoising, and signal reconstruction to obtain noise-free sound signals 1, 2, and 3. The vibration signal can be denoised using conventional denoising methods. The noise-free sound signals 1, 2, and 3, as well as the vibration signal, are then subjected to short-time Fourier transform to time-frequency graphs and grayscale processing before image segmentation, thus realizing the processing of vibration and sound signals. Subsequently, feature extraction is performed, including GGCM texture feature extraction and GMRF texture feature extraction, and then training and testing sample sets are formed. After selecting features from the training sample set using the Laplace score method, the samples are input into the SVM fault type diagnosis model to determine if the iteration conditions are met. If yes, the model enters the IS-SVM model and outputs classification result 1, classification result 2, classification result 3, and classification result 4. If not, DBO optimization is performed on the parameters M, N, c, and g (dung beetle optimization algorithm). Finally, the final result is obtained through subjective Bayesian fusion diagnosis.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A fault diagnosis method for high-voltage circuit breakers based on image block segmentation, characterized in that, The method includes: Acquire vibration and sound signals from high-voltage circuit breakers; Short-time Fourier transforms are performed on the vibration signal and the sound signal respectively to obtain two-dimensional time-frequency diagrams of the vibration signal and the sound signal. Then, each two-dimensional time-frequency diagram is grayscaled to obtain grayscaled two-dimensional time-frequency diagrams. The grayscale two-dimensional time-frequency image is divided into blocks using the image block method to obtain multiple small images of equal size. Extract the first and second features of each of the small images to construct a feature dataset; wherein, the first feature is the feature of the gray-level-gradient co-occurrence matrix, and the second feature is the feature of the Gaussian-Markov random field; The Laplace score method is used to filter features in the feature dataset and construct the optimal feature subset. The optimal feature subset is input into the support vector machine fault diagnosis model. The objective function is to maximize classification accuracy. The model parameters are then evaluated to determine if they meet the iteration conditions. If they do, the diagnostic results for the vibration or sound signal are output. If they do not meet the iteration conditions, the dung beetle optimization algorithm is used to optimize the penalty factor c and kernel parameter g of the support vector machine. The optimized parameters are then input into the support vector machine fault diagnosis model for model update. The updated support vector machine fault diagnosis model is then re-processed with the optimal feature subset, and the iteration conditions are evaluated again. The iteration conditions are defined as either a classification accuracy reaching a preset threshold or a set number of iterations.

2. The high-voltage circuit breaker fault diagnosis method based on image segmentation according to claim 1, characterized in that, The formula for the short-time Fourier transform is as shown in equation (1): In the formula, (τ-t) is a window function centered at τ; x(τ) is the time-domain signal to be processed; STFT(t,f) is the output of the short-time Fourier transform, e is the natural constant, j is the imaginary unit, f is the frequency, and τ is the time variable.

3. The high-voltage circuit breaker fault diagnosis method based on image segmentation according to claim 1, characterized in that, The characteristics of the gray-gradient co-occurrence matrix include gray-level average, gradient average, gray-level standard deviation, gradient standard deviation, small gradient dominance, large gradient dominance, non-uniform gray-level distribution, non-uniform gradient distribution, energy, correlation, gray-level entropy, gradient entropy, mixed entropy, inertia and / or inverse gap.

4. The high-voltage circuit breaker fault diagnosis method based on image segmentation according to claim 1, characterized in that, The second feature is extracted using the following method: A second feature is extracted from the small image based on a Gaussian-Markov random field; wherein: The conditional probability formula for the Gauss-Markov random field is: P(y(m)|y(m+E),E∈G) (2) In the formula, y(m) is the pixel value of any pixel m in the image, P represents the probability that the value of the center pixel m is y(m) given the values ​​of the neighboring pixels, y(m+E) is the pixel value of a certain pixel m+E in G, G is a symmetric neighborhood with m as the center and E as the radius but not including m, and E is the radius of the symmetric neighborhood G. The model formula for the Gauss-Markov random field is as follows: In the formula, y(S) is the image on the point set S, and G... m Let S be the neighborhood of a Gaussian-Markov random field with m pixels, S be the set of points on a W×W image patch, y1(m+E) be a point in S, and θ be a point in S. E is the weight of the symmetrical neighboring pixels, and e(m) is Gaussian noise with a mean of 0; Substituting the points on the small image into equation (3) yields: In the formula, Let be the transpose of the matrix y(m); θ is the eigenvector to be estimated in the model; The following equation (4) is obtained by estimating and solving using the least squares error criterion: In the formula, T u To extract features from the Gaussian-Markov random field, Z m It is a matrix about y(m).

5. The high-voltage circuit breaker fault diagnosis method based on image segmentation according to claim 1, characterized in that, When using the Laplace score method to filter features in the feature dataset and construct the optimal feature subset, the weight matrix expression corresponding to the adjacency matrix in the Laplace score method is: In the formula, x h and x j They are the h-th and j-th nodes in the adjacency matrix, where e is the natural constant and S is the adjacency matrix. hj To evaluate the weight matrix of the similarity between the h-th node and the j-th node, G hj Let h×j be the adjacency matrix; The formula for calculating the Laplace fraction of the r-th feature is: In the formula, L r Let f represent the Laplace score of the r-th feature. rh and f rj Var(f) represents the r-th feature of the h-th sample and the j-th sample. r ) represents the variance of the r-th feature relative to all samples.

6. The high-voltage circuit breaker fault diagnosis method based on image segmentation according to claim 1, characterized in that, The support vector machine fault diagnosis model is represented as follows: In the formula, w is the normal vector of the optimal hyperplane; b is the threshold; y h The class label of the h-th sample is represented by T, which is the matrix transpose; h represents the sample index; N represents the number of samples; x h Let represent the classification feature of the h-th sample, and st represent the constraint condition; min represents the minimum value function; Based on the aforementioned support vector machine fault diagnosis model, after introducing the penalty factor and the Lagrange operator, we obtain: In the formula, max is the maximum value function; α is the Lagrange operator; α h It is the Lagrange operator for the h-th sample; α j It is the Lagrange operator for the j-th sample; y j x is the classification label of the j-th sample; j It is the classification feature of the j-th sample; C is the penalty factor; Furthermore, a kernel function is introduced, which is a Gaussian radius basis kernel function, expressed as: In the formula, K is the kernel function, exp is an exponential function with the natural constant as the base, and g is the parameter to be optimized in the Gaussian radial basis kernel function; Substituting equation (10) into the solution, we obtain the following formula for the linear classification function: In the formula, f(x) is the classification function, sgn is the sign of the score calculated by the function, and the final classification result is output accordingly.

7. The high-voltage circuit breaker fault diagnosis method based on image segmentation according to claim 1, characterized in that, In the dung beetle optimization algorithm, the position update formula for the rolling ball group in the obstacle-free mode is as follows: In the formula, t represents the current iteration number; Let represent the position of the i-th dung beetle in the population at the t-th iteration; k, β, and b are all constants; Indicates the worst position in the population; This represents the position of the i-th dung beetle in the population at the (t+1)-th iteration; The position update formula for the rolling ball group with obstacles is: In the formula, μ is the angle at which the dung beetle changes its posture to obtain a new direction when it encounters an obstacle and cannot move forward. This represents the position of the i-th dung beetle in the population during the (t-1)th iteration. The formula for updating the position of chicks in a breeding population is: In the formula, Ub represents the position of the κ-th primordial ball in the t-th iteration; * and Lb * These are the upper and lower bounds of the spawning region; b1 and b2 represent independent random vectors of size 1×D; D represents the dimension optimized in the support vector machine. Indicates the globally optimal position; The formula for updating the location of foraging groups is: In the formula, C1 is a random number following a normal distribution, and C2 is a random vector of size 1×D between (0,1); Lb l and Ub l These are the upper and lower limits of the foraging area; The formula for updating the location of the feces-stealing group is: In the formula, F is a constant value, and z represents a normal distribution. It is the optimal position for the current population.

8. A vibration-acoustic combined fault diagnosis method for high-voltage circuit breakers, characterized in that, The combined vibration-acoustic fault diagnosis method for high-voltage circuit breakers includes: Acquire vibration signals from one channel and sound signals from three channels; The vibration signal from one channel and the sound signals from three channels are preprocessed to obtain a preprocessed signal; The high-voltage circuit breaker fault diagnosis method based on image segmentation, as described in any one of claims 1 to 7, is used to process the preprocessed signals to obtain preliminary diagnostic results for each signal. The preliminary diagnostic results are fused at the decision level using a subjective Bayesian network to output the final diagnostic results for the high-voltage circuit breaker.

9. The high-voltage circuit breaker fault diagnosis method combining vibration and acoustic analysis as described in claim 8, characterized in that, The vibration signal from one channel and the audio signals from three channels are preprocessed to obtain the preprocessed signal in the following ways: The vibration signal of one channel is denoised to obtain a denoised vibration signal; The audio signals from the three channels are denoised using wavelet packets and then reconstructed to obtain the reconstructed signal. The denoised vibration signal and the reconstructed signal are used as preprocessed signals.

10. The high-voltage circuit breaker fault diagnosis method based on vibration and acoustic combined according to claim 8, characterized in that, The subjective Bayesian network is based on Bayes' theorem, which is as follows: In the formula, P(B) u () represents the u-th event B u The prior probability; P(A|B) u ) is the u-th event B u The probability of event A occurring given the given conditions; P(B) u |A) is the u-th event B given that event A has occurred. u The probability of occurrence; n is the number of events; Knowledge can be represented in the following ways: if A then(LS,LN)B (18) In the formula, LS and LN represent the sufficiency and necessity of the rule being valid, respectively, and are defined as follows: By introducing a probability function and establishing its probability relationship with the occurrence of event B, the posterior probability formula is derived as follows: In the formula, P(B) is the probability of event B occurring.

Images