High-voltage switch cabinet characteristic gas inversion partial discharge detection method and device

Abstract

The invention provides a high-voltage switch cabinet characteristic gas inversion partial discharge detection method and device, and belongs to the technical field of high-voltage switch cabinet partial discharge monitoring. The method comprises the following steps: acquiring field fault data representing mixed gas components in the high-voltage switch cabinet when an insulation defect occurs in the high-voltage switch cabinet, obtaining simulated fault data representing the mixed gas components through simulation, and generating an original data set based on the field fault data and the simulated fault data; wherein the insulation defect types comprise an N-type metal protrusion defect, a P-type metal particle defect and a G-type insulation layer air gap defect; expanding the original data set by using a segmented cubic spline interpolation PSCI method to obtain an expanded data set; introducing an MCMC algorithm in a Gaussian process GP, and resampling the extended data set to obtain a sample set F used for estimating posterior distribution; and carrying out characteristic gas inversion by utilizing a Bagging-DF model, training the model by utilizing a sample set F, and testing the model by utilizing an original data set to obtain an insulation defect type.

Description

Technical Field

[0001] The present invention relates to the technical field of partial discharge monitoring of high-voltage switch cabinets, and in particular to a method and device for detecting partial discharge of a high-voltage switch cabinet by characteristic gas inversion. Background Art

[0002] High Voltage Switchgear is mainly used to ensure the safe and stable operation of power during the transmission, transformation and distribution process. It is usually composed of high-voltage circuit breakers, high-voltage load switches, high-voltage contactors, high-voltage fuses, high-voltage disconnectors, high-voltage grounding switches, high-voltage transformers, station transformers and other high-voltage electrical equipment, as well as related control, measurement, protection, regulation devices, internal connectors, accessories, housings and support parts. It is a widely used power transmission and distribution equipment in the power system.

[0003] With the rapid development of my country's economy, the power consumption and installed capacity of the power system have grown rapidly. When the power system has a short circuit, circuit breaker fault or needs to be powered off for maintenance, the high-voltage switchgear can safely disconnect the power through relevant devices, which can not only protect the downstream power equipment but also ensure the safety of the power operator. However, during the manufacturing, transportation and operation and maintenance of the high-voltage switchgear, improper operation may cause insulation defects inside the cabinet. Typical defect types can be divided into the following three types: (1) Metal protrusion defects: metal burrs protruding from the inner wall of the high-voltage switchgear cavity or the high-voltage (HV) conductor; (2) Metal particle defects: conductive particles left in the switchgear cavity during the manufacturing and installation process and metal powder formed by friction of metal devices during operation; (3) Insulation layer air gap defects: solid insulation layer due to aging, production process and other factors, resulting in air gaps inside. Air gaps may also appear in components due to different material expansion coefficients, aging of plastics and other problems. Partial discharge caused by insulation defects is one of the common faults of high-voltage switchgear. Early detection and treatment of partial discharge is of great significance for preventing equipment failures.

[0004] In recent years, researchers have found that partial discharge inside high-voltage switchgear can cause the air composition around the insulation defect to change and produce new gases. The type, content and generation rate of these characteristic gases are closely related to the defect type and energy density. Therefore, by effectively extracting the characteristic information of the characteristic gas decomposition components, the internal insulation fault of the high-voltage switchgear can be effectively diagnosed. However, the characteristic gas is actually affected by the internal environmental factors of the high-voltage switchgear (such as temperature, humidity, gas mixture, partial discharge mode, etc.), as well as maintenance status factors (operation methods, inspection methods and equipment conditions of maintenance personnel), gas collection methods, equipment conditions and other factors on the fault diagnosis results, which may cause the concentration and composition of the characteristic gas to change; In addition, there may be limitations on the amount of effective decomposition data: Insufficient data: Due to the complex and dynamic gas generation process caused by partial discharge, a variety of characteristic gases may be generated, and in actual applications, the amount of effective gas decomposition data obtained is limited. This means that when fault diagnosis is based on limited data, the model may not be able to fully capture all possible fault modes, resulting in limited accuracy of fault diagnosis. Missing or incomplete data: In some cases, due to the limitations of gas collection technology or insufficient gas sampling when a fault occurs, the data available for diagnosis is incomplete. Due to the above-mentioned environmental and data limitations, misjudgments and missed diagnoses are prone to occur in the existing diagnostic process. Misjudgments may be caused by the similar gas characteristics produced by different types of faults, which leads to the model misclassifying them; missed diagnoses may occur when the amount of data is insufficient or the characteristic gas changes are relatively weak, and the fault cannot be accurately detected. Summary of the invention

[0005] In view of this, the present invention provides a method and device for detecting partial discharge by characteristic gas inversion of a high-voltage switch cabinet, which can quickly and accurately identify the type of fault defects inside the switch cabinet based on gas component data affected by other factors such as internal environmental factors of the high-voltage switch cabinet, maintenance status factors, body collection methods, equipment conditions, etc., avoid misjudgment and missed diagnosis, and improve diagnostic efficiency.

[0006] The technical solution adopted by the embodiment of the present invention to solve the technical problem is:

[0007] A method for detecting partial discharge of a high-voltage switch cabinet by characteristic gas inversion, comprising:

[0008] Step S1, obtaining field fault data representing the components of the mixed gas in the cabinet when an insulation defect occurs in the high-voltage switch cabinet, simulating the insulation defect in the high-voltage switch cabinet to obtain simulated fault data representing the components of the mixed gas, and generating an original data set based on the field fault data and the simulated fault data; wherein the insulation defect types include N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects; the mixed gas is composed of SO2F2, SOF4, SO2, CF4, and CO2;

[0009] Step S2, using the piecewise cubic spline interpolation PSCI method, performing piecewise cubic spline interpolation fitting on all data in the original data set to obtain a fitting curve, extracting data points from the fitting curve to expand the original data set to obtain an extended data set;

[0010] Step S3, introducing the Monte Carlo Markov chain algorithm MCMC into the Gaussian process GP, resampling the extended data set to obtain a sample set F = {f1, f2, ..., f N};

[0011] Step S4, introducing bagging bootstrap sampling ensemble learning and decision fusion into the generalized prediction model to become a bagging-DF model, performing characteristic gas inversion, using the sample set F to train the bagging-DF model, inputting the original data set into the trained bagging-DF model for testing, and obtaining the insulation defect type in the switch cabinet;

[0012] Step S41, using the Bagging method to generate M base classifiers C m ;

[0013] Step S42: Use the Bootstrap sampling method to perform N1 samplings with replacement on the sample set F to obtain a training subset Fm with the same length as X. Perform M Bootstrap samplings in total to obtain M training subsets F. m , m∈[1,M];

[0014] Step S43, using each of the training subsets F m Train M base classifiers C separately m , we get M base classifiers C with differences m ;

[0015] Step S44, using the trained M base classifiers C m The data in the original data set are predicted and analyzed respectively to obtain the base classifiers C m The basic probability assignment set m m ;

[0016] Step S45, using the Dp synthesis rule to synthesize the output results to obtain the synthesis result m(A):

[0017]

[0018]

[0019] In the formula, m M (B M ) represents m M Neutron Set B M The probability of; K is the conflict coefficient;

[0020] Step S46, calculate the trust level Bel(A):

[0021]

[0022] The label corresponding to the A value with the largest Bel(A) value is selected as the final output classification, that is, the insulation defect type in the switch cabinet.

[0023] Preferably, the step S1 obtains simulated fault data representing the mixed gas components by simulating insulation defects in the high-voltage switch cabinet, including:

[0024] A high-voltage switchgear model with three types of internal insulation defects in the cavity is established, and sensors are installed at N2 collection positions selected in the high-voltage switchgear model;

[0025] A single insulation defect type is set to perform fault simulation, and the concentrations of mixed gas components in N directions are collected as a single set of simulated fault data;

[0026] For N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects, N3, N4, and N5 fault simulations are performed respectively to obtain the simulated fault data for each insulation defect type.

[0027] Preferably, the on-site fault data and the simulated fault data are both gas component data affected by internal environmental factors of the high-voltage switchgear, maintenance status factors, gas collection methods, and equipment factory parameters.

[0028] Preferably, the step S3 comprises:

[0029] Step S31, define the posterior distribution: for a given ={x1,x2,...,x N} and the class label set Y = {y1,y2,...,y N}, n∈[1,N], where y n Represents x nFor the corresponding internal insulation defect type, the posterior distribution p(f|X,y) of the potential function f is defined as:

[0030]

[0031] Among them, p(y|X) approximates the posterior distribution by the MCMC method; p(f|X) is the prior distribution of the Gaussian process; p(y|f) is the likelihood of the class label given the potential function value f, expressed as:

[0032]

[0033] In the formula, σ(f(x)) is the Sigmoid function;

[0034] Step S32, the MCMC sampling process of p(y|X) for a single element is as follows:

[0035] Initialization: Select an element from the extended dataset X as the initial potential function value f (0) ;

[0036] Iterative update: Let t represent the current iteration number, t ∈ [1, T], f (t) represents the new sample obtained after the t-th iteration based on the previous sample f (t-1) ; The iterative process is as follows:

[0037] Starting from the current sample f (t-1) , generate a new candidate sample f′ ∼ N(f (t-1) , Σ) based on the jump distribution, where Σ is a covariance matrix;

[0038] Calculate the acceptance probability α(f (t) , f′):

[0039]

[0040] Generate a random number A uniformly distributed in [0, 1], and compare the acceptance probability α(f (t) , f′) and A: p(f'|X) is the target distribution probability density of the candidate sample f′ under the extended dataset X; q(f (t) |X) represents the probability distribution of f (t) under the condition of the given data X;

[0041] If α(f (t) , f′) ≥ A, then take the candidate sample f′ as the output result f of the t-th iteration (t) ;

[0042] If α(f (t) , f′) < A, then reject the candidate sample f′ and take the sample f (t-1)As the output result f of the tth iteration (t) ;

[0043] When the maximum number of iterations T is reached, the iteration is stopped and f is obtained. (0) The final iteration result f (T) ;

[0044] Step S33: Process each element in the extended data set X based on the process of step S32 to obtain F = {f1, f2, ..., f N}, f n For x n The processing result.

[0045] The present invention provides a high-voltage switch cabinet characteristic gas inversion partial discharge detection device, characterized in that it is used to implement the above-mentioned method.

[0046] It can be seen from the above technical solution that the method and device for detecting partial discharge by characteristic gas inversion of high-voltage switch cabinet provided by the embodiment of the present invention. First, an original database is established to obtain field fault data representing the components of the mixed gas in the cabinet when insulation defects occur in the high-voltage switch cabinet, and simulated fault data representing the components of the mixed gas are obtained by simulating the insulation defects in the high-voltage switch cabinet, and an original data set is generated based on the field fault data and the simulated fault data; wherein, the insulation defect types include N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects; the mixed gas is composed of SO2F2, SOF4, SO2, CF4, and CO2; using the piecewise cubic spline interpolation PSCI method, piecewise cubic spline interpolation fitting is performed on all data in the original data set to obtain a fitting curve, and data points are extracted from the fitting curve to expand the original data set to obtain an extended data set; the Monte Carlo Markov chain algorithm MCMC is introduced into the Gaussian process GP to resample the extended data set to obtain a sample set F = {f1, f2, ..., f N}; Introduce Bagging bootstrap sampling integrated learning and decision fusion into the generalized prediction model to become the Bagging-DF model, perform characteristic gas inversion, use sample set F to train the Bagging-DF model, input the original data set into the trained Bagging-DF model for testing, and obtain the insulation defect type in the switch cabinet. The present invention can quickly and accurately identify the fault defect type inside the switch cabinet based on the gas component data affected by other factors such as the internal environmental factors of the high-voltage switch cabinet, the collection method of the maintenance status factor, and the equipment conditions, thereby improving the diagnostic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is the overall technical roadmap of the characteristic gas inversion partial discharge detection method of the high-voltage switch cabinet of the present invention;

[0048] Figure 2 The extended data scheme flow chart of the piecewise cubic spline interpolation of the present invention;

[0049] Figure 3 It is a block diagram of the establishment of the Bagging-DF model in the present invention; DETAILED DESCRIPTION

[0050] The technical scheme and technical effects of the present invention are further elaborated in detail below in conjunction with the accompanying drawings of the present invention.

[0051] The purpose of the present invention is to design a set of characteristic gas inversion partial discharge detection methods for high-voltage switch cabinets based on Gaussian process classification (GPC) and decision fusion (DF), which can effectively improve the generalization ability of diagnosis of different insulation defects (metal protrusion defects, metal particle defects, insulation layer air gap defects) inside the high-voltage switch cabinet, and use the piecewise cubic spline interpolation method to expand the large amount of laboratory simulation and field fault data accumulated in the early stage, and use the expanded data as the training data of the fault diagnosis model. On this basis, the Bagging ensemble learning model is introduced and combined with the Gaussian process classification (GPC) method as the base classifier, and the Markov chain Monte Carlo (MCMC) is introduced in the generalized predictive control model for sampling, and the Bagging ensemble model is further improved, and multiple evidence theory combination rules are used for integration.

[0052] The field fault data adopted in the solution of the present invention integrates the internal environmental factors of the high-voltage switchgear (such as temperature, humidity, gas mixture, partial discharge mode, etc.), as well as the influence of other factors such as maintenance status factors (operating methods, inspection methods and equipment conditions of maintenance personnel), gas collection methods, equipment conditions, etc. on the measurement results, and effectively decomposes the limitation of data volume (limited data). Compared with creating original data only using simulation data, it is more realistic, can improve the accuracy of classification prediction when the model is applied to actual scenarios, and effectively reduce the occurrence of misjudgment and missed diagnosis.

[0053] refer to Figure 1 As shown, the present invention provides a method for detecting partial discharge of a high-voltage switch cabinet by characteristic gas inversion, the steps comprising:

[0054] Step S1, obtaining field fault data representing the components of the mixed gas in the cabinet when an insulation defect occurs in the high-voltage switch cabinet, obtaining simulated fault data representing the components of the mixed gas by simulating the insulation defect occurring in the high-voltage switch cabinet, and generating an original data set based on the field fault data and the simulated fault data; wherein the insulation defect types include N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects; the mixed gas is composed of SO2F2, SOF4, SO2, CF4, and CO2;

[0055] Step S2, using the piecewise cubic spline interpolation PSCI method, performing piecewise cubic spline interpolation fitting on all data in the original data set to obtain a fitting curve, extracting data points from the fitting curve to expand the original data set to obtain an extended data set;

[0056] Step S3, introducing the Monte Carlo Markov chain algorithm MCMC into the Gaussian process GP, resampling the extended data set to obtain a sample set F = {f1, f2, ..., f N};

[0057] Step S4, introducing bagging bootstrap sampling ensemble learning and decision fusion into the generalized prediction model to become a bagging-DF model, performing characteristic gas inversion, using sample set F to train the bagging-DF model, inputting the original data set into the trained bagging-DF model for testing, and obtaining the insulation defect type in the switch cabinet.

[0058] Preferably, step S1 obtains simulated fault data representing the mixed gas components by simulating insulation defects in the high-voltage switch cabinet, including:

[0059] A high-voltage switchgear model with three types of internal insulation defects in the cavity is established, and N2 collection positions are selected in the high-voltage switchgear model to install sensors. Among them, not all sensors are of the same type and model, and the emphasis on sensitivity may also be different. The on-site fault data and simulated fault data are both gas component data affected by the internal environmental factors of the high-voltage switchgear, maintenance status factors, gas collection methods, and equipment factory parameters.

[0060] Set a single insulation defect type for fault simulation, and collect the concentrations of mixed gas components in N directions as a single set of simulated fault data;

[0061] For N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects, N3, N4, and N5 fault simulations are performed respectively to obtain simulated fault data for each insulation defect type.

[0062] As an optional implementation, the data ratios corresponding to the three insulation defects in the original data set are 1:1:1;

[0063] The PSCI method using piecewise cubic spline interpolation is as follows Figure 2 , Xnew represents the horizontal coordinate of the sampling point on the fitting curve, i is the horizontal coordinate of the original data, j represents the j-th type of insulation defect, w is the growth feature, and Ynew(w) represents the new data generated by the growth feature in the j-th type of insulation defect fitting curve to ensure effective training of the model.

[0064] Preferably, the process of using the Gaussian process GP to introduce the Monte Carlo Markov chain algorithm MCMC to process data in step S3 is:

[0065] Step S31, define the posterior distribution: for a given ={x1,x2,...,x N} and the class label set Y = {y1,y2,...,y N}, n∈[1,N], where y n Represents x n Corresponding to the internal insulation defect type, the posterior distribution p(f|X,y) of the potential function f is defined as:

[0066]

[0067] Among them, p(y|X) is the posterior distribution approximated by the MCMC method; p(f|X) is the prior distribution of the Gaussian process, and it is usually assumed that f(x) obeys the Gaussian process GP(0,k(x,x′)), that is, zero mean and covariance defined by the kernel function k(x,x′); p(y|f) is the likelihood of the class label given the potential function value f, expressed as:

[0068]

[0069] Where, σ(f(x)) is the Sigmoid function;

[0070] Step S32, the MCMC sampling process of p(y|X) for a single element is:

[0071] Initialization: Select an element from the extended dataset X as the initial potential function value f (0) ;

[0072] Iterative update: Let t represent the current iteration number, t∈[1,T], f (t) Indicates that based on the previous sample f (t-1) The new sample obtained after the tth iteration; the iterative process is:

[0073] From the current sample f (t-1) Starting from, generate new candidate samples f′~N(f (t-1) ,∑), where Σ is a covariance matrix, which represents the distribution width of generating new samples;

[0074] Calculate the acceptance probability α(f (t) ,f′):

[0075]

[0076] Generate a uniformly distributed random number A∈[0,1] and compare the acceptance probability α(f (t), f') and A: p(f'|X) is the target distribution probability density of the candidate sample f' under the extended dataset X; q(f (t) |X) represents the probability distribution of f (t) under the condition of the given data X;

[0077] If α(f (t) , f') ≥ A, then the candidate sample f' is used as the output result f (t) of the t-th iteration;

[0078] If α(f (t) , f') < A, then the candidate sample f' is rejected, and the sample f (t-1) is used as the output result f (t) of the t-th iteration;

[0079] When the maximum number of iterations T is reached, the iteration stops, and f (0) the final iteration result f (T) is obtained;

[0080] Step S33, based on the process of step S32, process each element in the extended dataset X to obtain F = {f1, f2,..., f N}, where f n is the processing result of x n ;

[0081] As an alternative implementation, the Bagging-DF model refers to Figure 3 shown in the figure, and the specific implementation process of step S4 for establishing the model and performing training and testing is as follows:

[0082] Step S41, use the Bagging method to generate M base classifiers C m ;

[0083] Step S42, use the Bootstrap sampling method to perform N1 times of sampling with replacement on the sample set F to obtain a training subset Fm with the same length as X, and perform the Bootstrap sampling M times to obtain M training subsets F m , m ∈ [1, M];

[0084] Step S43, use each training subset F m to train M base classifiers C m respectively, to obtain M different base classifiers C m ; For each classifier C m , calculate the BPA according to the output category. Assume that the output of each classifier is a category yi ∈ Y, representing the prediction result of the classifier for a certain sample;

[0085] Step S44, use the trained M base classifiers Cm The data in the original data set are predicted and analyzed respectively to obtain the base classifiers C m The basic probability assignment set m m ;

[0086] Step S45, using the Dp synthesis rule to synthesize the output results to obtain the synthesis result m(A):

[0087]

[0088] In the formula, m M (B M ) represents m M Neutron Set B M The probability of; K is the conflict coefficient;

[0089] Step S46, calculate the trust Bel(A):

[0090]

[0091] The label corresponding to the A value with the largest Bel(A) value is selected as the final output classification, that is, the insulation defect type in the switch cabinet.

[0092] The present invention provides a high-voltage switch cabinet characteristic gas inversion partial discharge detection device, characterized in that it is used to implement the above-mentioned method.

[0093] The design of the present invention has the following technical features:

[0094] 1. Based on the generalized predictive control of MCMC sampling, the piecewise cubic spline interpolation method is introduced to expand the 352 sets of data accumulated in the early stage to obtain 1200 sets of new data. The model training method with new data as training set and original data as test set improves the robustness and generalization ability of air decomposition component fault diagnosis.

[0095] 2. The sampling scheme of Monte Carlo Markov chain algorithm (MCMC) is introduced into Gaussian process (GP) to quantify the error introduced by approximation. For support vector machine, hinge loss function is used to calculate empirical risk, and regularization term is added to the solution system to optimize structural risk. Three layers of RBM units are set in deep belief network (DBN) and stacked in sequence. The optimization algorithm of DBN is small batch gradient descent method with learning rate 0.8. For the diagnosis problem of defect type in partial discharge fault, extended data is used as training set, which proves that GPC is the best fault identification model. Gaussian process classification (GPC): GPC can provide flexible modeling capabilities when dealing with fault diagnosis problems with uncertainty and complexity. It models characteristic gas data through Gaussian process, which helps to overcome the problem of insufficient data and improve the accuracy of classification. Decision fusion (DF): The decision fusion method can effectively combine the results of multiple classifiers to reduce the misjudgment and missed diagnosis problems that may be caused by a single classifier, and enhance the stability and reliability of diagnosis.

[0096] 3. Bagging uses Bootstrap sampling to make each base classifier have different prediction performance, solving the overfitting problem that may occur in a single classifier. In the Bagging framework, DF fusion complementary information is introduced to reduce the uncertainty of the model, and the fusion of redundant information is used to improve the stability of the model to construct a Bagging-DF model. In the Bagging-DF model, the base classifier is a GPC model with MCMC sampling.

[0097] According to the embodiments disclosed in the present invention, the present invention also discloses an electronic device, a readable storage medium and a computer program product. The electronic device is intended to represent various forms of digital computers, including a computing unit, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The computing unit, ROM and RAM are connected to each other via a bus. The input / output (I / O) interface is also connected to the bus.

[0098] Multiple components in the device are connected to the I / O interface, including: input unit, output unit, storage unit, and communication unit. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0099] The computing unit may be a variety of general and / or special processing components with processing and computing capabilities to execute the various methods and processes described above, such as the GIS sub-millimeter particle defect accumulation morphology and development stage identification method.

[0100] What is disclosed above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A method for detecting partial discharge by characteristic gas inversion of a high-voltage switch cabinet, characterized in that: include: Step S1, obtaining field fault data representing the components of the mixed gas in the cabinet when an insulation defect occurs in the high-voltage switch cabinet, simulating the insulation defect in the high-voltage switch cabinet to obtain simulated fault data representing the components of the mixed gas, and generating an original data set based on the field fault data and the simulated fault data; wherein the insulation defect types include N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects; the mixed gas is composed of SO2F2, SOF4, SO2, CF4, and CO2; Step S2, using the piecewise cubic spline interpolation PSCI method, performing piecewise cubic spline interpolation fitting on all data in the original data set to obtain a fitting curve, extracting data points from the fitting curve to expand the original data set to obtain an extended data set; Step S3, introducing the Monte Carlo Markov chain algorithm MCMC into the Gaussian process GP, resampling the extended data set to obtain a sample set F for estimating the posterior distribution; Step S4, introducing bagging bootstrap sampling ensemble learning and decision fusion into the generalized prediction model to become a bagging-DF model, performing characteristic gas inversion, using the sample set F to train the bagging-DF model, inputting the original data set into the trained bagging-DF model for testing, and obtaining the insulation defect type in the switch cabinet; Step S41, using the Bagging method to generate M base classifiers C m ; Step S42: Use the Bootstrap sampling method to perform N1 samplings with replacement on the sample set F to obtain a training subset Fm with the same length as X. Perform M Bootstrap samplings in total to obtain M training subsets F. m , m∈[1,M]; Step S43, using each of the training subsets F m Train M base classifiers C separately m , we get M base classifiers C with differences m ; Step S44, using the trained M base classifiers C m The data in the original data set are predicted and analyzed respectively to obtain the base classifiers C m The basic probability assignment set m m ; Step S45, using the Dp synthesis rule to synthesize the output results to obtain the synthesis result m(A): In the formula, m M (B M ) represents m M Neutron Set B M The probability of; K is the conflict coefficient; Step S46, calculate the trust Bel(A): The label corresponding to the A value with the largest Bel(A) value is selected as the final output classification, that is, the insulation defect type in the switch cabinet.

2. The method for detecting partial discharge by characteristic gas inversion of a high-voltage switch cabinet according to claim 1, characterized in that: The step S1 obtains simulated fault data representing the mixed gas components by simulating insulation defects in the high-voltage switch cabinet, including: A high-voltage switchgear model with three types of internal insulation defects in the cavity is established, N2 acquisition positions are selected in the high-voltage switchgear model to install sensors, and the internal environment of the high-voltage switchgear model is set; A single insulation defect type is set to perform fault simulation, and the concentrations of mixed gas components in N directions are collected as a single set of simulated fault data; For N-type metal protrusion defects, P-type metal particle defects, and G-type insulation layer air gap defects, N3, N4, and N5 fault simulations are performed respectively to obtain the simulated fault data for each insulation defect type.

3. The method for detecting partial discharge by characteristic gas inversion of high-voltage switch cabinet according to claim 2, characterized in that: The on-site fault data and the simulated fault data are both gas component data affected by the internal environmental factors of the high-voltage switchgear, maintenance status factors, gas collection methods, and equipment factory parameters.

4. The method for detecting partial discharge by characteristic gas inversion of a high-voltage switch cabinet according to claim 3, characterized in that: The step S3 comprises: Step S31, define the posterior distribution: for a given ={x1,x2,...,x N } and the class label set Y = {y1,y2,...,y N }, n∈[1,N], where y n Represents x n Corresponding to the internal insulation defect type, the posterior distribution p(f|X,y) of the potential function f is defined as: Among them, p(y|X) is the posterior distribution approximated by the MCMC method; p(f|X) is the prior distribution of the Gaussian process; p(y|f) is the likelihood of the class label given the potential function value f, expressed as: Where, σ(f(x)) is the Sigmoid function; Step S32, the MCMC sampling process of p(y|X) for a single element is: Initialization: Select an element from the extended dataset X as the initial potential function value f (0) ; Iterative update: Let t represent the current iteration number, t∈[1,T], f (t) Indicates that based on the previous sample f (t-1) The new sample obtained after the tth iteration; the iterative process is: From the current sample f (t-1) Starting from, generate new candidate samples f′~N(f (t-1) ,∑), where Σ is a covariance moment; Calculate the acceptance probability α(f (t) ,f′): Generate a uniformly distributed random number A∈[0,1] and compare the acceptance probability α(f (t) ,f′) and A: p(f'|X) is the target distribution probability density of candidate sample f′ under the extended data set X; q(f (t) |X) means that under the given data X, f (t) The probability distribution of If α(f (t) ,f′)≥A, then the candidate sample f′ is taken as the output result f of the tth iteration (t) ; If α(f (t) , f′) < A, then reject the candidate sample f′ and take the sample f (t-1) as the output result f of the t-th iteration (t) ; when the maximum number of iterations T is reached, stop the iteration and obtain f (0) as the final iteration result f (T) ; Step S33: Process each element in the extended dataset X based on the process in Step S32 to obtain F = {f1, f2,..., f N}, where f n is the processing result of x n .

5. A high-voltage switchgear characteristic gas inversion partial discharge detection device, characterized in that: Used to implement the method according to any one of claims 1 to 3.

6. An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-4.

8. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 4.

Images