Switch cabinet defect diagnosis method based on multi-information fusion and related product

Abstract

The invention relates to the field of electrical equipment fault diagnosis, in particular to a switch cabinet defect diagnosis method based on multi-information fusion and a related product, and the method comprises the steps: detecting a switch cabinet by adopting multiple detection methods, and extracting a feature vector from a detection signal; constructing different defect diagnosis models for different detection methods by using the feature set after dimension reduction; obtaining a to-be-detected sample of a to-be-detected switch cabinet through multiple detection methods, and inputting the to-be-detected sample to the corresponding defect diagnosis model; obtaining a basic distribution probability, fusing to obtain a final comprehensive distribution probability, and determining a final prediction result; according to the invention, through combination of multiple detection means, limitation of a single detection method is made up, multiple defect types in the switch cabinet can be identified more comprehensively and accurately, outputs of different defect detection models are fused, and robustness and accuracy of diagnosis are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of electrical equipment fault diagnosis, and in particular to a switch cabinet defect diagnosis method based on multi-information fusion and related products. Background Art

[0002] As an important equipment in the power system, switchgear is widely used in power transmission and distribution, substations and other places. The reliability of its operating status is crucial to the stability of the entire power system. However, switchgear is easily affected by environmental factors, operating errors and equipment aging during long-term operation, resulting in various defects. These defects may cause electrical failures and even cause serious consequences such as equipment damage and power outages. Therefore, it is of great significance to effectively detect and diagnose defects in switchgear.

[0003] In the existing technology, defect detection methods for switch cabinets are mostly based on a single detection technology. However, due to the complex manifestations of electrical equipment failures, a single detection method is often difficult to fully and accurately identify and diagnose various types of defects. Even though there are some fusion technologies at this stage, how to effectively fuse the signal information of different detection methods to obtain a unified and reliable diagnosis result is still a technical problem.

[0004] Therefore, there is an urgent need for a new method that can integrate the characteristics of multiple detection methods and use more intelligent fusion technology to improve the accuracy and reliability of switchgear defect diagnosis. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a switch cabinet defect diagnosis method and related products based on multi-information fusion. Through comprehensive analysis of multiple detection methods (such as transient ground voltage detection, ultrasonic detection and ultra-high frequency detection), advanced dimensionality reduction algorithm and DS evidence theory fusion technology are adopted to effectively improve the accuracy, robustness and reliability of switch cabinet defect diagnosis.

[0006] The present invention is achieved through the following technical solutions:

[0007] A switch cabinet defect diagnosis method based on multi-information fusion, comprising:

[0008] Establish a switchgear test platform to simulate various typical defects;

[0009] Use multiple detection methods to detect the switch cabinet and extract feature vectors from the detection signals;

[0010] Reduce the dimension of the feature vector to obtain the feature set after dimension reduction;

[0011] Using the feature set after dimensionality reduction, different defect diagnosis models are constructed for different detection methods;

[0012] Acquire test samples of the switch cabinet to be tested by using a variety of detection methods, and input the test samples into the corresponding defect diagnosis model;

[0013] The basic distribution probability of the samples to be tested is obtained through different defect diagnosis models, and the basic distribution probabilities are fused to obtain the final comprehensive distribution probability. The final prediction result is determined by the maximum support principle.

[0014] Specifically, the switch cabinet test platform is equipped with TEV sensors, ultrasonic sensors, UHF sensors, data acquisition cards, preamplifiers, and synchronous collectors;

[0015] The detection methods include: transient ground voltage detection, ultrasonic detection and UHF detection; transient ground voltage detection is performed through TEV sensors, data acquisition cards and synchronous collectors, ultrasonic detection is performed through ultrasonic sensors, data acquisition cards, preamplifiers and synchronous collectors, and UHF detection is performed through UHF sensors, data acquisition cards, preamplifiers and synchronous collectors;

[0016] Obtain the original transient ground voltage signal, filter and denoise it; calculate the maximum amplitude A of the processed transient ground voltage signal max , amplitude mean μ A , amplitude standard deviation σ A , phase mean μ φ , phase standard deviation σ φ , total energy E total , frequency mean μ f and the frequency standard deviation σ f ; Construct feature vector F TEV =[A max , μ A , σ A , μ φ , σ φ μ φ , E total , μ f , σ f ] T ;

[0017] Get the original ultrasonic signal, remove high-frequency noise and low-frequency interference by filtering; calculate the peak amplitude U of the processed ultrasonic signal peak , effective value μ U , amplitude standard deviation σ U , Instantaneous limit mean Instantaneous phase standard deviation δ U , total signal energy E U , spectrum center frequency f U and spectrum bandwidth B U ; Construct feature vector

[0018] Get the original UHF signal and perform bandpass filtering on it; extract the maximum signal amplitude P of the processed signal max , mean amplitude μ P , amplitude standard deviation σ P , Phase Center Phase standard deviation Total energy E P , main frequency f max and frequency extension bandwidth W P ; Construct feature vector

[0019]

[0020] Specifically, the method of obtaining the feature set after dimensionality reduction includes:

[0021] Calculate the mean of the eigenvectors F for F TEV 、F US or F UHF , F (i) For the first i feature components;

[0022] Calculate the standard deviation of the eigenvector Where N is N TEV 、N US or N UHF ;

[0023] Normalize the feature vector to obtain the standardized feature vector

[0024] Determine the Gaussian kernel function Among them, x p and x q is the first feature vector in the original p The feature components and q feature components, σ is the kernel width parameter automatically selected based on cross-validation;

[0025] Constructing the kernel matrix K , the element K in the kernel matrix pq = k(x p , x q ), and centralize the kernel matrix K′=K-1 N K-K1 N +1 N K1 N , where 1 N Each element is N×N matrix;

[0026] Perform eigenvalue decomposition on the centralized kernel matrix K′, K′v k =λ k v k , and obtain the eigenvalue λ k and its corresponding eigenvector v k ;

[0027] Sort the eigenvalues ​​by size, determine the dimension M of the vector after dimensionality reduction, select the first M eigenvectors to form the feature matrix V, and project the original eigenvector to obtain the eigenvector F1=V after dimensionality reduction T K′;

[0028] Reduce the dimension of multiple feature vectors to construct a feature vector set.

[0029] Furthermore, the method for obtaining the feature set after dimensionality reduction also includes:

[0030] After obtaining the feature vector F1 after the first dimension reduction, the independent components are obtained by ICA separation. in, For the r Separation matrix of independent components;

[0031] Evaluate each independent component s based on the mutual information r (n) to the recognition task, and set the contribution threshold, select the independent components greater than the contribution threshold, and construct the secondary dimension reduction feature vector F2 = [s1(n), s2(n), ..., s K (n)] T , where K is the number of independent components retained after adaptive selection;

[0032] Reduce the dimension of multiple eigenvectors to construct a set of secondary eigenvectors.

[0033] Optionally, a feature vector set corresponding to transient ground voltage detection and a corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set;

[0034] The support vector machine is trained through the training set, and the model is evaluated through the validation set, and the parameters are adjusted to obtain the defect detection model based on transient ground voltage;

[0035] The feature vector set corresponding to ultrasonic testing and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set;

[0036] The random forest model is trained through the training set, and the model is evaluated through the validation set, and the parameters are adjusted to obtain the ultrasonic-based defect detection model;

[0037] The feature vector set corresponding to the UHF detection and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set;

[0038] The naive Bayes classifier is trained through the training set, and the model is evaluated through the validation set, and the parameters are adjusted to obtain the UHF-based defect detection model.

[0039] Specifically, the method of fusing the outputs of multiple defect diagnosis models includes:

[0040] Based on the number of types of simulated typical defects, a defect type set Θ = {θ1, θ2, ..., θ M}, where M is the total number of defect types;

[0041] Input the sample u of the switch cabinet to be tested into the defect detection model, and obtain the predicted probability distribution of the defect type of the sample u in, For the vth model, sample u belongs to defect θ M The probability of; v = 1, 2, 3, and correspond to the defect detection model based on transient ground voltage detection, the defect detection model based on ultrasonic detection and the defect detection model based on ultra-high frequency detection respectively;

[0042] Order v Defect detection model for the sample to be tested u Give defects t The basic probability of distribution and

[0043] Obtaining the basic distribution probability of the defect detection model based on transient ground voltage detection Obtaining the basic distribution probability of the defect detection model based on ultrasonic testing Obtaining the basic distribution probability of the defect detection model based on UHF detection

[0044] right and Perform fusion to obtain the basic distribution probability after fusion

[0045] right and Perform fusion to obtain the basic distribution probability after fusion

[0046] right and Perform fusion to obtain the basic distribution probability after fusion

[0047] right and Perform fusion to obtain the basic distribution probability after fusion

[0048] right and Fusion to obtain the final comprehensive allocation probability

[0049] Normalize the final comprehensive allocation probability

[0050] Loop to obtain samples to be tested u The final combined assigned probability relative to all defect types;

[0051] Determine the output sample through the maximum support principle u Predicted defect types

[0052] Optionally, and Methods for performing fusion include:

[0053] Determine the conflict coefficient in, A and B is the set of defect types Θ Different subsets of For subset A The basic distribution probability of the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection;

[0054] Calculate the basic distribution probability after fusion

[0055] The basic assignment probability of the output fusion after normalization

[0056] right and Methods for performing fusion include:

[0057] Determine the conflict coefficient in, A and B are different subsets of the defect type set Θ, For subset A The basic distribution probability of the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection;

[0058] Calculate the basic distribution probability after fusion

[0059] The basic assignment probability of the output fusion after normalization

[0060] right and Methods for performing fusion include:

[0061] Determine the conflict coefficient in, A and B is the set of defect types Θ Different subsets of For subset A The basic distribution probability of the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection;

[0062] Calculate the basic distribution probability after fusion

[0063] The basic assignment probability of the output fusion after normalization

[0064] A switch cabinet defect diagnosis terminal based on multi-information fusion includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a switch cabinet defect diagnosis method based on multi-information fusion as described above is implemented.

[0065] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the switch cabinet defect diagnosis method based on multi-information fusion as described above is implemented.

[0066] A computer program product includes a computer program / instruction, which, when executed by a processor, implements a switch cabinet defect diagnosis method based on multi-information fusion as described above.

[0067] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0068] The present invention establishes a test platform, simulates a variety of typical defects, uses a variety of detection methods such as transient ground voltage, ultrasound and ultra-high frequency, extracts feature vectors from the detection signal, and performs dimensionality reduction processing to obtain a feature set after dimensionality reduction. Based on the feature set after dimensionality reduction, different defect diagnosis models are constructed respectively, and then the feature vectors of the switch cabinet to be tested are obtained and input into the corresponding model, and the outputs of each model are fused to finally obtain a comprehensive diagnosis result.

[0069] The present invention makes up for the limitations of a single detection method by combining multiple detection means, can more comprehensively and accurately identify multiple defect types in the switch cabinet, and fuse the outputs of different defect detection models, significantly improving the robustness and accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, are used to explain the principles of the present invention. These drawings are included to provide a further understanding of the present invention, and the accompanying drawings are included in and constitute a part of this specification and do not constitute a limitation of the embodiments of the present invention.

[0071] Figure 1 It is a structural schematic diagram of a switch cabinet defect diagnosis method based on multi-information fusion according to the present invention. DETAILED DESCRIPTION

[0072] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and implementation methods. It is understood that the specific implementation methods described herein are only used to explain the relevant content, rather than to limit the present invention.

[0073] It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings.

[0074] In the absence of conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0075] Embodiment 1

[0076] like Figure 1 As shown, a switch cabinet defect diagnosis method based on multi-information fusion includes:

[0077] Establish a switchgear test platform to simulate a variety of typical defects; build a switchgear test platform that meets actual working conditions, can truly simulate the operating status of the switchgear, and introduce artificial defects into the test platform, including typical defect types such as metal protrusions, air gaps, metal suspensions and internal insulation discharges.

[0078] A variety of detection methods are used to detect the switchgear and extract feature vectors from the detection signals; the methods include transient earth voltage detection (TEV detection), ultrasonic detection and ultra-high frequency detection (UHF detection). After signal processing, the data collected by each detection method are extracted into feature vectors, which can reflect the physical characteristics of various defects in the switchgear, such as signal amplitude, frequency, phase and other information.

[0079] The feature vector is reduced in dimension to obtain a reduced feature set; after dimensionality reduction, a simplified feature set that can effectively retain information is obtained, providing efficient input for subsequent diagnosis.

[0080] Using the feature set after dimensionality reduction, different defect diagnosis models are constructed for different detection methods; the data of each detection method (such as TEV, ultrasonic, and UHF) corresponds to an independent model.

[0081] The test samples of the switch cabinet to be tested are obtained through various detection methods, and the test samples are input into the corresponding defect diagnosis model; the signal data is obtained from the detection platform, and the feature vector is extracted, and then the feature vector of the sample is input into the corresponding defect diagnosis model.

[0082] The basic distribution probability of the sample to be tested is obtained through different defect diagnosis models. The model will output the basic distribution probability of the sample for each defect type. The basic distribution probability reflects the possibility that the sample belongs to a certain defect type.

[0083] The basic allocation probabilities of each model are integrated to obtain the final comprehensive allocation probability, and the final prediction result is determined by the maximum support principle. That is, the defect type with the largest comprehensive allocation probability is selected as the final prediction result.

[0084] The switch cabinet test platform is equipped with TEV sensors, ultrasonic sensors, UHF sensors, data acquisition cards, preamplifiers, and synchronous collectors. The detection methods include: transient ground voltage detection, ultrasonic detection, and UHF detection; transient ground voltage detection through TEV sensors, data acquisition cards, and synchronous collectors, ultrasonic detection through ultrasonic sensors, data acquisition cards, preamplifiers, and synchronous collectors, and UHF detection through UHF sensors, data acquisition cards, preamplifiers, and synchronous collectors.

[0085] TEV sensors are used to detect transient earth voltage, mainly capturing voltage fluctuations caused by partial discharge in switch cabinets. Fluctuations reflect the local insulation performance of electrical equipment and can reveal potential discharge defects. During the detection process, the sensor collects signals and then converts them through a data acquisition card, and ensures signal timing consistency through a synchronous collector.

[0086] Ultrasonic sensors are used to detect ultrasonic signals generated by partial discharge. This signal can penetrate the insulating medium and reflect the location and intensity of internal defects. The signal collected by the ultrasonic sensor is enhanced by the preamplifier, transmitted to the data acquisition card for sampling, and finally synchronized by the synchronous collector for signal timing.

[0087] UHF sensors are used to capture the ultra-high frequency electromagnetic wave signals (Ultra-High Frequency) generated by partial discharge, and are particularly suitable for the detection of partial discharge in gas-insulated equipment. The signals collected by the sensor are processed by bandpass filtering and preamplifiers, sampled by data acquisition cards, and synchronized by synchronous collectors.

[0088] The original transient voltage signal is obtained and filtered and denoised; that is, the signal usually contains noise components after acquisition, and filtering technology (such as low-pass filter or high-pass filter) is used to pre-process the signal to remove irrelevant noise and retain the effective signal.

[0089] Calculate the maximum amplitude A of the transient ground voltage signal after processing max , amplitude mean μ A , amplitude standard deviation σ A , phase mean μ φ , phase standard deviation σ φ , total energy E total , frequency mean μ f and the frequency standard deviation σ f ; Construct feature vector F TEV =[A max , μ A , σ A , μ φ , σ φ , E total , μ f , σ f ] T ;

[0090] Obtain the original ultrasonic signal and remove high-frequency noise and low-frequency interference through filtering; use the filter to remove high-frequency noise and low-frequency interference in the ultrasonic signal and retain the core information of the signal.

[0091] Calculate the peak amplitude U of the processed ultrasonic signal peak , effective value μ U , amplitude standard deviation σ U , Instantaneous limit mean Instantaneous phase standard deviation δ U , total signal energy E U , spectrum center frequency f U and spectrum bandwidth B U ; Construct feature vector

[0092] The original UHF signal is obtained and band-pass filtered; the UHF signal is band-pass filtered to remove irrelevant frequency components and retain the effective frequency band.

[0093] Extract the maximum signal amplitude P of the processed signal max, mean amplitude μ P , amplitude standard deviation σ P , Phase Center Phase standard deviation Total energy E P , main frequency f max and frequency extension bandwidth W P ; Construct feature vector

[0094] After obtaining the above feature vectors, methods for obtaining the feature set after dimensionality reduction include:

[0095] Calculate the mean of the eigenvectors F for F TEV 、F US or F UHF , F (i) For the first i feature components; by summing up each component in the feature vector and dividing it by the number of components, the average level of the overall feature can be obtained, which can reflect the central tendency of the feature vector, that is, describe the central position of the feature vector as a whole.

[0096] To measure the breadth of the distribution of eigenvalues, the standard deviation of the eigenvector is calculated Where N is N TEV 、N US or N UHF ; By calculating the difference between each feature component and the mean, then squaring these differences, and finally taking the square root, the dispersion of the feature components can be obtained. A large standard deviation indicates that the feature components fluctuate greatly, while a small standard deviation indicates less fluctuation.

[0097] Standardize the feature vector to eliminate the scale differences between different feature components and obtain the standardized feature vector By standardization, each component of the feature vector will have zero mean and unit variance. Standardization is to scale each feature component by its mean and standard deviation, ensuring that the feature component has a mean of zero and a standard deviation of one.

[0098] The Gaussian kernel function is used to measure the similarity between different feature components in the feature vector. Among them, x p and x q is the pth eigenvalue and the pth eigenvalue in the original eigenvector q feature components, σ The kernel width parameter is automatically selected based on cross-validation; it measures the difference between two feature components based on the Euclidean distance. The Gaussian kernel function decays exponentially, making feature components with smaller distances have higher similarity, while the similarity of components with larger distances decreases rapidly.

[0099] Construct the kernel matrix K, the element K in the kernel matrix pq = k(x p , x q ), and centralize the kernel matrix K′=K-1 N K-K1 N +1 N K1 N , where 1 N Each element is N×N matrix; each element K of the kernel matrix pq = k(x p , x q ) represents the pth eigenvalue and the pth eigenvalue of the eigenvector q The similarity between the feature components is calculated by centering the entire kernel matrix (removing the mean).

[0100] Perform eigenvalue decomposition on the centralized kernel matrix K′, K′v k =λ k v k , and obtain the eigenvalue λ k and its corresponding eigenvector v k ; The eigenvalue represents the importance of the eigenvector, and the eigenvector is the new feature axis after dimensionality reduction.

[0101] Sort the eigenvalues ​​by size, determine the dimension M of the vector after dimensionality reduction, select the first M eigenvectors to form the feature matrix V, and project the original eigenvector to the new space to obtain the eigenvector F1=V after dimensionality reduction T K′;

[0102] Reduce the dimension of multiple feature vectors to obtain a feature set with reduced dimension.

[0103] In order to further reduce the complexity of the data, after obtaining the feature vector F1 after the first dimension reduction, the independent components are obtained by ICA separation. in, For the r The separation matrix of the independent components can be optimized iteratively to maximize the non-Gaussianity, and independent signals can be obtained. Independent component analysis (ICA) is a signal separation technique used to decompose linear combination signals into independent source signals.

[0104] Evaluate each independent component s based on the mutual information r (n) Contribution to the recognition task. The higher the mutual information value, the greater the contribution of the component to the task. According to the set contribution threshold, the independent components above the threshold are selected to form the feature vector set after secondary dimensionality reduction.

[0105] And set the contribution threshold, select the independent components greater than the contribution threshold, and construct the secondary dimension reduction feature vector F2 = [s1(n), s2(n), ..., s K (n)] T , where K is the number of independent components retained after adaptive selection;

[0106] Finally, multiple feature vectors are reduced in dimension to construct a feature vector set.

[0107] Embodiment 3

[0108] The feature vector set corresponding to transient ground voltage detection (such as maximum amplitude, mean amplitude, total energy, etc.) and the corresponding defect type label (such as partial discharge, insulation damage, etc.) are constructed as a data set, and the data set is divided into a training set and a validation set; the training set is used to train the model, and the validation set is used to evaluate the performance of the model during the training process to avoid overfitting of the model. The common division method is 80% of the data is used for training and 20% of the data is used for validation.

[0109] The support vector machine is trained through the training set, and the model is evaluated through the validation set. The parameters are adjusted to obtain a defect detection model based on transient ground voltage. The support vector machine (SVM) performs classification by finding a hyperplane that maximizes the class interval. Using the feature vectors and labels in the training set, the SVM model learns how to distinguish different defect types.

[0110] The feature vector set corresponding to ultrasonic testing (such as peak amplitude, effective value, spectrum center frequency, etc.) and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set;

[0111] The random forest model is trained through the training set, and the model is evaluated through the validation set. The parameters are adjusted to obtain the ultrasonic-based defect detection model; random forest is classified by building multiple decision trees. Each tree randomly extracts samples from the training set for training, and finally determines the prediction result through majority voting. Random forest reduces overfitting and improves generalization performance through this "multiple trees" approach.

[0112] The feature vector set corresponding to the UHF detection and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set;

[0113] The naive Bayes classifier is trained through the training set, and the model is evaluated through the validation set. The parameters are adjusted to obtain a defect detection model based on UHF. The naive Bayes classifier assumes that the features are independent of each other, calculates the posterior probability of each category, and selects the category with the largest posterior probability as the prediction result. The naive Bayes classifier is trained using the UHF detection feature vector and label data in the training set.

[0114] This embodiment uses three different training models, namely support vector machine (SVM), random forest (RF) and naive Bayes (NB), for three detection methods, namely transient earth voltage (TEV) detection, ultrasonic detection and ultra-high frequency (UHF) detection. The reason is that the signal characteristics and data features of each detection method are taken into consideration.

[0115] Transient ground voltage signals are highly complex and have high-dimensional features, especially transient signals generated during partial discharge. The data features of these signals are usually distributed in high-dimensional space. Support vector machines (SVMs) are good at processing high-dimensional data and can effectively find the optimal classification hyperplane in complex data sets. By using kernel functions, SVMs can also handle linear inseparable situations, which is consistent with the nonlinear characteristics of transient ground voltage signals.

[0116] Ultrasonic signals are affected by environmental noise, equipment status and other factors during the detection process, and signal characteristics often show high randomness and nonlinear characteristics. Random Forest (RF) performs classification by integrating multiple decision trees, and each tree is randomly sampled from the data set during training. RF can effectively process the noise in ultrasonic signals through this "randomness", and the RF model is highly adaptable to nonlinear and complex feature distributions.

[0117] UHF signals often show features with clear statistical distribution in partial discharge detection, such as signal amplitude, frequency, phase, etc. Naive Bayes (NB) is a classification algorithm based on Bayes' theorem. It assumes that features are independent of each other and is suitable for processing data sets with clear probability distribution. The various feature dimensions of UHF detection signals often have high independence, so the use of NB classifiers can quickly and accurately infer the defect type to which the signal belongs.

[0118] Embodiment 4

[0119] Methods for fusing the outputs of multiple defect diagnosis models include:

[0120] Based on the number of types of simulated typical defects, a defect type set Θ = {θ1, θ2, ..., θ M}, where M is the total number of defect types; the set includes all possible defect types in the system, ensuring that each detection model outputs based on the same defect type.

[0121] The switch cabinet to be tested u Input the defect detection model and obtain the sample to be tested u Defect type prediction probability distribution in, For the vth model, sample u belongs to defect θ MThe probability of the sample being classified is obtained by using the predicted probability. v = 1, 2, 3, which correspond to the defect detection model based on transient ground voltage detection, the defect detection model based on ultrasonic detection, and the defect detection model based on ultra-high frequency detection. The defect prediction probability distribution of each model is obtained, which provides the preliminary results of sample classification. By predicting the probability, the possibility that the sample may belong to each defect type can be obtained.

[0122] Order v Defect detection model for the sample to be tested u Give defects t The basic probability of distribution and Convert the predicted probability distribution of each detection model into basic assignment probability. Basic assignment probability is an important concept in DS evidence theory, which represents the support of the model for each defect type, and the sum of the basic assignment probabilities of all defect types is equal to 1.

[0123] Obtaining the basic distribution probability of the defect detection model based on transient ground voltage detection Obtaining the basic distribution probability of the defect detection model based on ultrasonic testing Obtaining the basic distribution probability of the defect detection model based on UHF detection

[0124] right and Perform fusion to obtain the basic distribution probability after fusion

[0125] right and Perform fusion to obtain the basic distribution probability after fusion

[0126] right and Perform fusion to obtain the basic distribution probability after fusion

[0127] right and Perform fusion to obtain the basic distribution probability after fusion

[0128] right and Fusion to obtain the final comprehensive allocation probability

[0129] By gradually integrating the outputs of all models, we finally get the comprehensive basic distribution probability and normalize the final comprehensive distribution probability.

[0130] Loop to obtain samples to be testedu The final combined assigned probability relative to all defect types;

[0131] Determine the output sample through the maximum support principle u Predicted defect types

[0132] The three preliminary fusion methods are the same.

[0133] right and Methods for performing fusion include:

[0134] The conflict coefficient is used to measure the inconsistency between the outputs of the two models, reflecting the degree of conflict between the results of the transient ground voltage detection model and the ultrasonic detection model. in, A and B are different subsets of the defect type set Θ, For subset A The basic distribution probability of the defect detection model based on transient ground voltage detection, For subset B In the basic distribution probability of the defect detection model based on ultrasonic testing, Indicates that there is no intersection; that is, a conflict will only occur when the two models make inconsistent judgments on different subsets (i.e., there is no intersection).

[0135] Calculate the basic distribution probability after fusion A∩B=θ t It means that both the transient voltage detection model and the ultrasonic detection model believe that the sample may belong to the defect type θ t situation.

[0136] The basic assignment probability of the output fusion after normalization

[0137] right and Methods for performing fusion include:

[0138] Determine the conflict coefficient in, A and B is the set of defect types Θ Different subsets of For subset A The basic distribution probability of the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection;

[0139] Calculate the basic distribution probability after fusion

[0140] The basic assignment probability of the output fusion after normalization

[0141] right and Methods for performing fusion include:

[0142] Determine the conflict coefficient in, A and B are different subsets of the defect type set Θ, For subset A The basic distribution probability of the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection;

[0143] Calculate the basic distribution probability after fusion

[0144] The basic assignment probability of the output fusion after normalization

[0145] Embodiment 5

[0146] A switch cabinet defect diagnosis terminal based on multi-information fusion includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a switch cabinet defect diagnosis method based on multi-information fusion as described above is implemented.

[0147] The memory can be used to store software programs and modules. The processor executes various functional applications and data processing of the terminal by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an execution program required for at least one function, etc.

[0148] The data storage area can store data created according to the use of the terminal, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0149] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the switch cabinet defect diagnosis method based on multi-information fusion as described above is implemented.

[0150] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instruction data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technology, CD-ROM, DVD or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory and mass storage devices can be collectively referred to as memory.

[0151] A computer program product includes a computer program / instruction, which, when executed by a processor, implements a switch cabinet defect diagnosis method based on multi-information fusion as described above.

[0152] A computer program product includes a computer program or set of instructions for performing specific tasks or implementing specific functions. These programs or instructions are designed to be executed by a processor to implement a series of predefined steps or operations. The program product may be stored in various forms of computer storage media, such as memory, hard disk, solid-state drive, optical disk or other forms of digital storage devices. It may exist in the form of compiled binary code or in the form of scripts or bytecodes that can be executed by an interpreter. The program product uses carefully designed algorithms and logical instructions to enable the processor to process data in a specific order and manner to complete various functions such as data analysis, user interaction, device control, etc.

[0153] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, unless they are contradictory.

[0154] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0155] It should be understood by those skilled in the art that the above embodiments are only for the purpose of clearly illustrating the present invention, and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications may be made based on the above invention, and these changes or modifications are still within the scope of the present invention.

Claims

1. A switch cabinet defect diagnosis method based on multi-information fusion, characterized in that: include: Establish a switchgear test platform to simulate various typical defects; Use multiple detection methods to detect the switch cabinet and extract feature vectors from the detection signals; Reduce the dimension of the feature vector to obtain the feature set after dimension reduction; Using the feature set after dimensionality reduction, different defect diagnosis models are constructed for different detection methods; Acquire test samples of the switch cabinet to be tested by using a variety of detection methods, and input the test samples into the corresponding defect diagnosis model; The basic distribution probability of the samples to be tested is obtained through different defect diagnosis models, and the basic distribution probabilities are fused to obtain the final comprehensive distribution probability. The final prediction result is determined by the maximum support principle.

2. A switch cabinet defect diagnosis method based on multi-information fusion according to claim 1, characterized in that: The switch cabinet test platform is equipped with TEV sensors, ultrasonic sensors, UHF sensors, data acquisition cards, preamplifiers, and synchronous collectors; The detection methods include: transient ground voltage detection, ultrasonic detection and UHF detection; transient ground voltage detection is performed through TEV sensors, data acquisition cards and synchronous collectors, ultrasonic detection is performed through ultrasonic sensors, data acquisition cards, preamplifiers and synchronous collectors, and UHF detection is performed through UHF sensors, data acquisition cards, preamplifiers and synchronous collectors; Obtain the original transient ground voltage signal, filter and denoise it; calculate the maximum amplitude A of the processed transient ground voltage signal max , amplitude mean μ A , amplitude standard deviation σ A , phase mean μ φ , phase standard deviation σ φ , total energy E total , frequency mean μ f and the frequency standard deviation σ f ; Construct feature vector F TEV =[A max , μ A , σ A , μ φ , σ φ , E total , μ f , σ f ] T ; Get the original ultrasonic signal, remove high-frequency noise and low-frequency interference by filtering; calculate the peak amplitude U of the processed ultrasonic signal peak , effective value μ U , amplitude standard deviation σ U , Instantaneous limit mean Instantaneous phase standard deviation δ U , total signal energy E U , spectrum center frequency f U and spectrum bandwidth B U ; Construct feature vector Get the original UHF signal and perform bandpass filtering on it; extract the maximum signal amplitude P of the processed signal max , mean amplitude μ P , amplitude standard deviation σ P , Phase Center Phase standard deviation Total energy E P , main frequency f max and frequency extension bandwidth W P ; Construct feature vector 3. The switch cabinet defect diagnosis method based on multi-information fusion according to claim 2 is characterized in that: Methods for obtaining the feature set after dimensionality reduction include: Calculate the mean of the eigenvectors F for F TEV 、F US or F UHF , F (i) is the i-th characteristic component in F; Calculate the standard deviation of the eigenvector Where N is N TEV 、N US or N UHF ; Normalize the feature vector to obtain the standardized feature vector Determine the Gaussian kernel function Among them, x p and x q are the p-th and q-th feature components in the original feature vector, and σ is the kernel width parameter automatically selected based on cross-validation; Construct the kernel matrix K, the element K in the kernel matrix pq = k(x p ,x q ), and centralize the kernel matrix K′=K-1 N K-K1 N +1 N K1 N , where 1 N Each element is N×N matrix; Perform eigenvalue decomposition on the centralized kernel matrix K′, K′v k =λ k v k , and obtain the eigenvalue λ k and its corresponding eigenvector v k ; Sort the eigenvalues ​​by size, determine the dimension H of the vector after dimensionality reduction, select the first H eigenvectors to form the feature matrix V, and project the original eigenvector to obtain the eigenvector F1=V after dimensionality reduction T K′; Reduce the dimension of multiple feature vectors to construct a feature vector set.

4. The switch cabinet defect diagnosis method based on multi-information fusion according to claim 3 is characterized in that: Methods for obtaining feature sets after dimensionality reduction also include: After obtaining the feature vector F1 after the first dimension reduction, the independent components are obtained by ICA separation. in, is the separation matrix of the rth independent component; Evaluate each independent component s based on the mutual information r (n) to the recognition task, and set the contribution threshold, select the independent components greater than the contribution threshold, and construct the secondary dimension reduction feature vector F2 = [s1(n), s2(n), ..., s K (n)] T , where K is the number of independent components retained after adaptive selection; Reduce the dimension of multiple eigenvectors to construct a set of secondary eigenvectors.

5. The switch cabinet defect diagnosis method based on multi-information fusion according to claim 2 is characterized in that: The feature vector set corresponding to the transient ground voltage detection and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set; The support vector machine is trained through the training set, and the model is evaluated through the validation set, and the parameters are adjusted to obtain the defect detection model based on transient ground voltage; The feature vector set corresponding to ultrasonic testing and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set; The random forest model is trained through the training set, and the model is evaluated through the validation set, and the parameters are adjusted to obtain the ultrasonic-based defect detection model; The feature vector set corresponding to the UHF detection and the corresponding defect type label are constructed as a data set, and the data set is divided into a training set and a validation set; The naive Bayes classifier is trained through the training set, and the model is evaluated through the validation set, and the parameters are adjusted to obtain the UHF-based defect detection model.

6. The switch cabinet defect diagnosis method based on multi-information fusion according to claim 5 is characterized in that: Methods for fusing the outputs of multiple defect diagnosis models include: Based on the number of types of simulated typical defects, a defect type set Θ = {θ1, θ2, ..., θ M }, where M is the total number of defect types; Input the sample u of the switch cabinet to be tested into the defect detection model, and obtain the predicted probability distribution of the defect type of the sample u in, For the u-th model, sample u belongs to defect θ M The probability of; v = 1, 2, 3, and correspond to the defect detection model based on transient ground voltage detection, the defect detection model based on ultrasonic detection and the defect detection model based on ultra-high frequency detection respectively; Let the vth defect detection model assign the basic distribution probability of defect t to the sample u to be tested and Obtaining the basic distribution probability of the defect detection model based on transient ground voltage detection Obtaining the basic distribution probability of the defect detection model based on ultrasonic testing Obtaining the basic distribution probability of the defect detection model based on UHF detection right and Perform fusion to obtain the basic distribution probability after fusion right and Perform fusion to obtain the basic distribution probability after fusion right and Perform fusion to obtain the basic distribution probability after fusion right and Perform fusion to obtain the basic distribution probability after fusion right and Fusion to obtain the final comprehensive allocation probability Normalize the final comprehensive allocation probability Circularly obtain the final comprehensive distribution probability of the sample u to be tested relative to all defect types; Determine the predicted defect type of the output sample u through the maximum support principle 7. The switch cabinet defect diagnosis method based on multi-information fusion according to claim 5 is characterized in that: right and Methods for performing fusion include: Determine the conflict coefficient Among them, A and B are different subsets of the defect type set Θ, The defects of subset A based on transient ground voltage detection The basic assignment probability of the detection model, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection; Calculate the basic distribution probability after fusion The basic assignment probability of the output fusion after normalization right and Methods for performing fusion include: Determine the conflict coefficient Among them, A and B are different subsets of the defect type set Θ, is the basic allocation probability of subset A in the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection; Calculate the basic distribution probability after fusion The basic assignment probability of the output fusion after normalization right and Methods for performing fusion include: Determine the conflict coefficient Among them, A and B are defects different subsets of the set Θ of types, is the basic allocation probability of subset A in the defect detection model based on transient ground voltage detection, is the basic distribution probability of subset B in the defect detection model based on ultrasonic testing, Indicates no intersection; Calculate the basic distribution probability after fusion The basic assignment probability of the output fusion after normalization 8. A switch cabinet defect diagnosis terminal based on multi-information fusion, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, a switch cabinet defect diagnosis method based on multi-information fusion as described in any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a switch cabinet defect diagnosis method based on multi-information fusion as described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by the processor, a switch cabinet defect diagnosis method based on multi-information fusion as described in any one of claims 1 to 7 is implemented.

Images