Power transformer fault diagnosis method, equipment and medium

Abstract

The invention discloses a power transformer fault diagnosis method and device and a medium, and relates to the technical field of industrial fault diagnosis, and the method comprises the steps: carrying out the normalization processing of the analysis data of gas in oil of a power transformer, and determining various observation state data; inputting the various observation state data as samples into a plurality of classification models, and determining a fault diagnosis result corresponding to each classification model; according to the fault diagnosis result corresponding to each classification model, calculating the credibility weight of each classification model, and based on the credibility weight of each classification model, determining a final fault diagnosis result corresponding to each classification model; and based on an adaptive Kalman filter, fusing the final fault diagnosis results corresponding to the classification models, and determining a power transformer fault diagnosis result. The power transformer fault diagnosis result comprises a fault diagnosis result corresponding to each fault type. The precision of the fault diagnosis result of the power transformer is improved.

Description

Technical Field

[0001] The present application relates to the technical field of industrial fault diagnosis, and in particular to a method, device and medium for diagnosing faults of power transformers. Background Art

[0002] Power transformers are critical equipment in power systems, known as the "heart" of the grid. They carry out the crucial tasks of transmitting, distributing, and converting voltage levels for electrical energy, making them a crucial component in ensuring efficient power system operation. Power plants, transmission networks, and end users all rely on the reliable support of transformers. Transformers not only ensure efficient transmission of electrical energy and reduce energy losses during long-distance transmission, but also, by adjusting voltage levels, provide a safe and stable power supply for industrial, commercial, and residential users.

[0003] In modern power systems, with the integration of new energy sources, the development of smart grids, and the continued growth of electricity load, transformers are becoming increasingly important. They not only fulfill the fundamental functions of traditional power grids but also play an irreplaceable role in ensuring system stability, supporting the integration of new energy sources, and improving power quality. As the core hub of the power system, the operating status of transformers is directly related to the safety and economic viability of the grid. Their performance and operational reliability have become key indicators of overall power system efficiency.

[0004] However, with the passage of time, fluctuations in electrical loads, and complex operating environments, transformers are highly susceptible to multiple stresses, including thermal, mechanical, and electrical stresses, leading to failures. These failures can cause power outages, impacting both residential and industrial operations. For example, failure of a primary main transformer can lead to widespread power outages, directly resulting in industrial shutdowns, commercial losses, and social inconvenience. Furthermore, the high cost of repairing these failures, including equipment replacement, emergency dispatch, and restoration of operations, significantly increases operational expenditures. Furthermore, frequent transformer failures can weaken power system reliability, increase the pressure on backup power supply dispatch, and impact the stable operation of the power grid. Therefore, accurately identifying transformer failures and implementing effective preventive and remedial measures have become critical issues in ensuring the safe and efficient operation of power systems. This is crucial for improving enterprise economic efficiency, reducing energy waste, and ensuring social stability. Summary of the Invention

[0005] The purpose of this application is to provide a power transformer fault diagnosis method, device and medium, which can solve the problem of low power transformer fault diagnosis accuracy in related technologies.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In the first aspect, the present application provides a method for diagnosing faults of power transformers, comprising: normalizing the oil gas analysis data of the power transformer, and determining a plurality of observation state data; each observation state data represents a gas concentration data; inputting the plurality of observation state data as samples into a plurality of classification models respectively, and determining the fault diagnosis results corresponding to each classification model; calculating the credibility weight of each classification model according to the fault diagnosis results corresponding to each classification model, and determining the final fault diagnosis results corresponding to each classification model based on the credibility weight of each classification model; based on an adaptive Kalman filter, fusing the final fault diagnosis results corresponding to each classification model to determine the fault diagnosis results of the power transformer; the fault diagnosis results of the power transformer include the fault diagnosis results corresponding to each fault type.

[0008] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the power transformer fault diagnosis method described above.

[0009] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the power transformer fault diagnosis method described above when executed by a processor.

[0010] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0011] The present application provides a method, device, and medium for diagnosing faults in power transformers. The present application first normalizes the oil-gas analysis data of the power transformer, determines a plurality of observation state data, and then inputs the plurality of observation state data as samples into a plurality of classification models to determine the fault diagnosis results corresponding to each classification model, thereby enabling different classification models to diagnose the oil-gas analysis data, that is, to obtain fault diagnosis results with different performances. Then, based on the fault diagnosis results corresponding to each classification model, the credibility weight of each classification model is calculated to determine the final fault diagnosis result corresponding to each classification model. This accurately quantifies the proportion of each classification model in the subsequent adaptive Kalman filter fusion process. Finally, based on the adaptive Kalman filter, the final fault diagnosis results corresponding to each classification model are fused, thereby improving the accuracy of the power transformer fault diagnosis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic flow chart of a method for diagnosing a power transformer fault provided in an embodiment of the present application;

[0014] Figure 2 This is a flow chart of the overall process of a power transformer fault diagnosis method provided in an embodiment of the present application;

[0015] Figure 3 This is a flowchart of the credibility weight module of each classification model provided in the embodiments of the present application;

[0016] Figure 4 This is a flow chart of the adaptive Kalman filter fusion process provided in an embodiment of the present application;

[0017] Figure 5 This is a structural diagram of the adaptive Kalman filter multi-model fusion process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] With the rapid development of fault diagnosis technology, the focus of existing transformer fault diagnosis techniques has shifted from single-classifier diagnosis to multi-classifier fusion diagnosis. Multi-classifier fusion technology comprehensively considers the advantages of each classifier, thereby overcoming the limitation of single-classifier methods, where diagnostic performance depends on the performance of each classifier. In practical applications, by using multi-classifier fusion technology to fuse the diagnostic results of various machine learning models, more accurate fault diagnosis results can be obtained compared to those obtained using a single classifier.

[0020] Among the multi-classifier fusion technologies, the multi-model fusion method represented by Kalman filter fusion is the current mainstream technology. Among them, the theoretical basis of Kalman filter is relatively deep, and the parameter configuration (process noise covariance matrix Q k and the measurement noise covariance matrix R k ) is suitable, the fusion performance is better, but a constant Q is used in the fusion processk and R k It will cause the problem of inaccurate state recursion, which will lead to reduced fault diagnosis accuracy. Therefore, in the fusion process, the appropriate Q k and R k It plays an important role in improving the fusion performance of Kalman filter and accurately completing fault diagnosis.

[0021] In the related art, the fault diagnosis algorithm based on Kalman filter multi-model fusion includes the following steps: first, three classification models are used to process the pre-processed Dissolved Gas Analysis (DGA) data to obtain the probability diagnosis results of each fault for different models, and then the initialization parameters of the Kalman filter (process noise covariance matrix Q k and the measurement noise covariance matrix R k Finally, three Kalman filters are used to sequentially fuse the probabilistic diagnosis results of different models, and the fault type with the highest probability is selected as the output result. Although this method has a simple algorithm structure and is easy to implement, the fusion performance of the Kalman filter is greatly dependent on the pre-set process noise covariance matrix Q. k and the measurement noise covariance matrix R k , but in the fusion process Q k and R k Not constant, inaccurate Q k and R k This will lead to unreasonable use of state data and observation data, resulting in inaccurate state recursion and reduced fault diagnosis accuracy.

[0022] Specifically, too large (small) Q k This will result in too little (too much) use of state data during state recursion, and too large (small) R k This will result in too little (too much) use of observation data during the state recursion process. In addition, the weights of the fault diagnosis probability results obtained by each classification model are not consistent. If the same weight is used, the diagnostic performance of each model may not be fully utilized.

[0023] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0024] like Figure 1 As shown, the present application provides a power transformer fault diagnosis method, comprising:

[0025] Step 101: After normalizing the gas analysis data in the oil of the power transformer, multiple observation state data are determined; each observation state data represents a type of gas concentration data.

[0026] Step 102: Inputting a plurality of observation state data as samples into a plurality of classification models respectively, and determining the fault diagnosis result corresponding to each classification model.

[0027] Step 103: Calculate the credibility weight of each classification model according to the fault diagnosis result corresponding to each classification model, and determine the final fault diagnosis result corresponding to each classification model based on the credibility weight of each classification model.

[0028] Step 104: Based on the adaptive Kalman filter, the final fault diagnosis results corresponding to the classification models are integrated to determine the power transformer fault diagnosis result; the power transformer fault diagnosis result includes the fault diagnosis result corresponding to each fault type.

[0029] In some embodiments, step 103 specifically includes:

[0030] Step 201: For any classification model, calculate the accuracy index, precision index and consistency index of the classification model.

[0031] Step 202: Determine the credibility weight of the classification model according to the accuracy index, precision index and consistency index corresponding to the classification model.

[0032] Step 203: Based on the credibility weight of each classification model and the fault diagnosis result corresponding to each classification model, determine the final fault diagnosis result corresponding to each classification model.

[0033] In some embodiments, step 202 specifically includes: determining the total number of samples correctly identified for each type of fault type in the classification model based on the fault diagnosis results of the classification model and the actual fault labels of the samples; determining the accuracy index corresponding to each type of fault type in the classification model based on the total number of samples correctly identified and the total number of samples corresponding to each type of fault type; determining the precision index corresponding to each type of fault type in the classification model based on the total number of fault types, the number of samples correctly identified for each type of fault type, and the number of samples identified for each type of fault type; determining the consistency index corresponding to each type of fault type in the classification model based on the total number of fault types, the number of samples correctly identified for each type of fault type, and the total number of samples for each type of fault type; and determining the mean of the accuracy index, precision index, and consistency index corresponding to each type of fault type in the classification model, and using the mean as the credibility weight of the classification model.

[0034] In some embodiments, step 104 specifically includes steps 301 to 308 .

[0035] Step 301: sorting the final fault diagnosis results corresponding to the classification models to determine a fault diagnosis result sequence; a final fault diagnosis result in the fault diagnosis result sequence is used as an observation vector.

[0036] Step 302: Based on the initialized adaptive Kalman filter, determine the initial state vector and the initial state error covariance matrix.

[0037] Step 303: According to the fault diagnosis result sequence, each observation vector in the fault diagnosis result sequence is used as the current observation vector.

[0038] Step 304: Input the current observation vector, the initial state vector and the initial state error covariance matrix into an adaptive Kalman filter to determine a current state prediction value of the current observation vector.

[0039] Step 305: Determine the current innovation vector based on the current state prediction value and the current observation vector.

[0040] Step 306: Determine the current state error covariance matrix based on the current innovation vector.

[0041] Step 307: The current state prediction value and the current state error covariance matrix of the current observation vector are used as the initial state prediction value and the initial state error covariance matrix of the next observation vector, respectively, until the current state prediction value of the last observation vector is determined; the current state prediction value of the last observation vector is the fused current state prediction value.

[0042] Among them, the current state prediction value and the current state error covariance matrix of the current observation vector are used as the initial state prediction value and the initial state error covariance matrix of the next observation vector, respectively, and return to step 304 to determine the current state prediction value of the next observation vector until the current state prediction value of the last observation vector is determined.

[0043] Step 308: Using the fused current state prediction value as a final state vector, and determining a power transformer fault diagnosis result based on the final state vector.

[0044] In some embodiments, step 306 specifically includes steps 401 and 402.

[0045] Step 401: Determine a current adaptive coefficient based on the current innovation vector.

[0046] Step 402: Adaptively adjust the noise covariance matrix and the observation noise covariance matrix of the initial adaptive Kalman filter according to the current adaptive coefficient to determine the current state error covariance matrix.

[0047] In some embodiments, the current adaptive coefficient is: Among them, α k is the current adaptive coefficient; k is the number of observation vectors; is the new information vector corresponding to the k-th observation vector; T is the transpose; b is the forgetting factor; G k is a customized comprehensive evolution indicator; λ is a weighting coefficient; is the new information vector corresponding to the lth observation vector.

[0048] In some embodiments, step 402 specifically includes the following.

[0049] Adaptively adjust the noise covariance matrix in the adaptive Kalman filter according to the current adaptive coefficient to determine the prior state error covariance matrix; based on the prior state error covariance matrix, adaptively adjust the observation noise covariance matrix in the adaptive Kalman filter according to the current adaptive coefficient to determine the posterior state error covariance matrix; and use the posterior state error covariance matrix as the current state error covariance matrix.

[0050] In some embodiments, the plurality of classification models include at least a BPNN classifier, a KNN classifier, and a RF classifier.

[0051] In practical applications, this application can fuse the probabilistic diagnosis results of various faults obtained by processing DGA data using different models, and then determine the fault state of the power transformer. The specific process is as follows.

[0052] Before fusion, the probabilistic diagnosis results of each fault of different classification models are preprocessed, and the mean of the accuracy, precision and consistency index of each classification model is introduced as the credibility weight of the probabilistic diagnosis results of different classification models; during fusion, the adaptive coefficient α is obtained for different state data and observation data. k Thus the process noise covariance matrix Q is corrected k and the observation noise covariance matrix R k Complete the adaptive Kalman filter sequential fusion to obtain the final probabilistic diagnosis results for each fault type, select the fault type with the maximum probability as the result output, and achieve accurate fault diagnosis.

[0053] To achieve the above purpose, this application adopts the following technical solutions: Figure 2 shown.

[0054] Step 1: Collect DGA data from the power system transformer. When the power system transformer has abnormal working conditions (different faults), the amount of gas generated by the aging and decomposition of the oil-paper insulation material will increase sharply and continue to dissolve in the oil. Therefore, in order to correctly classify the fault type of the transformer, it is necessary to first collect DGA data from the power system transformer. The DGA data includes six states: partial discharge (PD), low-energy discharge fault (LED), high-energy discharge fault (HED), low-temperature thermal fault (LTT), high-temperature thermal fault (HTT) and normal condition (NC). The six states are six types of observation state data, representing the original concentration data of six gases.

[0055] Specifically, DGA data is collected from power system transformers. DGA data is usually used for status monitoring of power equipment such as transformers. The power transformer DGA data set used in the implementation case of this application is IEC TC 10, which is obtained from the document "Duval M, DePabla A. Interpretation of gas-in-oil analysis using new IEC publication 60599 and IEC TC 10 databases [J]. IEEE Electrical Insulation Magazine, 2001, 17 (2): 31-41." As shown in Table 1, Table 1 is an example table of some IEC TC 10 data sets. Among them, IEC TC 10 is the 10th Technical Committee of the International Electrotechnical Commission (IEC Technical Committee 10), which focuses on the standardization of power networks and their related equipment. This data set gives examples of gas concentration values ​​of various states observed in use of six different types of transformer equipment, including PD, LED, HED, LTT, HTT and NC; among them, the data set contains various observed state data.

[0056] Table 1: Some examples of IEC TC 10 datasets

[0057]

[0058] Step 2: After data preprocessing, the data is divided into an 8:2 ratio. The DGA data is preprocessed using logarithmic transformation to reduce the magnitude differences between gas concentration data, and then normalized. Finally, the training and test data sets are divided into an 8:2 ratio, and the calculation is as follows:

[0059] X i =log 10 x i +1(1)

[0060]

[0061] Among them, x i is the original concentration data of the i-th gas, X i is the concentration data of the i-th gas after logarithmic change, E(X i ) and Var(X i ) are the expectation and variance of the concentration data of the i-th gas after logarithmic transformation, D i is the normalized concentration data of the i-th gas, and i is the number of gas types.

[0062] After data preprocessing, the data were divided into an 8:2 ratio. Equation (1) was used to reduce the magnitude differences between the gas data using logarithmic transformations, and then the DGA data were normalized using equation (2). After data preprocessing, the analysis was repeated 30 times using each randomly mixed dataset with an 8:2 ratio to create new training and test datasets. Finally, the overall average diagnostic accuracy was obtained by averaging the diagnostic accuracy of these repeated studies.

[0063] Step 3: Train and establish three classification models: BPNN, KNN, and RF. After setting parameters such as the number of hidden layer neurons, activation function, optimizer, and maximum number of iterations for the Back Propagation Neural Network (BPNN), the number of neighbors for the K-Nearest Neighbors (KNN), and the number of base estimators for the Random Forest (RF), the training data from Step 2 was used to train and establish the BPNN, KNN, and RF classification models.

[0064] Specifically, train and establish BPNN, KNN, and RF models. The training set obtained in step 2 is used to train and establish BPNN, KNN, and RF classification models. The BPNN uses 64 hidden layer neurons, the activation function is the Rectified Linear Unit (ReLU), the optimizer is the Adaptive Moment (AdaM), and the maximum number of iterations is 500. The number of neighbors for the KNN is 5, and the number of base estimators for the RF is 100.

[0065] Step 4: Calculate the fault diagnosis results of different fault types for each model (i.e., each classification model). Use the established BPNN, KNN, and RF models to perform fault diagnosis classification on the test set data and obtain the fault diagnosis probability results of different classification models. The fault diagnosis probability results are the fault diagnosis results.

[0066] Calculate the fault diagnosis probability of each model. Use the three classification models established in step 3 to classify the test set for fault diagnosis and obtain the fault diagnosis probability results of each model.

[0067] Step 5: Establish the credibility weight of each model. After obtaining the fault diagnosis probability results of different models, the mean of the accuracy, precision and consistency index of each model is used as the credibility weight of the fault diagnosis probability results of each model. Figure 3 As shown, the steps are as follows:

[0068] Step 5-1: Calculate the accuracy, precision, and consistency indicators of each model as follows:

[0069]

[0070] Among them, A is the accuracy index, N ri To correctly identify the total number, N al is the total number of samples; P is the precision index, R i is the number of samples correctly identified in the i-th fault type, P i is the number of samples identified as the i-th fault type, n is the total number of fault types; C is the consistency index, N i is the total number of samples of the i-th fault type, where the samples are the DGA data.

[0071] In step 5-2, the average of the three obtained indicators is used as the credibility weight of each model, which is calculated as follows:

[0072]

[0073] Step 5-3: Allocate the credibility weight of each model to the fault diagnosis probability result of each model.

[0074] The credibility weight of each classification model is established, and the mean of the accuracy, precision and consistency index of each classification model is used as the credibility weight of the fault diagnosis probability result of each model.

[0075] (6) Adaptive Kalman filter multi-model fusion. After initializing the parameters of each adaptive Kalman filter, the fault diagnosis probabilities of different models are used as observation vectors of adaptive Kalman filter 1, adaptive Kalman filter 2 and adaptive Kalman filter 3 respectively. Each adaptive Kalman filter selects the appropriate Q k and R k , thus completing the adaptive Kalman filter multi-model fusion. Specifically, in the fusion process, by obtaining the adaptive coefficient α k Post-parameter Q k and R k Correction is performed to complete the prediction and update steps in each adaptive Kalman filter, and finally the three adaptive Kalman filters are sequentially integrated to obtain the probabilistic diagnosis results of each fault, and the fault type with the maximum probability is selected as the result output. The probabilistic diagnosis result is the fault diagnosis probability result. Figure 4 and Figure 5 As shown, the steps are as follows.

[0076] Step 6-1, initialize the adaptive Kalman filter parameters. Specifically, set the process noise covariance matrix Q in the adaptive Kalman filter algorithm k , the observation noise covariance matrix R k , state vector x, state error covariance matrix P, state transfer matrix F and observation matrix H. Q0, R0 and P0 of the 0th state can be set according to the actual state. The default setting of x0 of the 0th state is 0*l6, and the default setting of F of the kth state is k and H k The default setting is I6, where l6 is a 6×1 column vector of all ones and I6 is the 6-dimensional identity matrix.

[0077] Step 6-2, perform the prediction step, first get the one-step state prediction value The calculation is as follows:

[0078]

[0079] Among them, F k is the k-th state transition matrix, is the state estimate for the k-1th state.

[0080] Among them, the state estimate of the k-1th state is the state prediction value of the k-1th observation vector.

[0081] Select the fault diagnosis probability result of one of the classification models as the kth observation vector zk , by obtaining and The difference between the two gets the new information vector of the kth state Right now:

[0082]

[0083] Among them, z k is the kth observation vector, is the observed predicted value of the kth state, H k is the observation matrix of the kth state; wherein the kth observation vector is taken as the kth state.

[0084] Then, the exponential fading memory weighted average method is used to construct the weight coefficient λ, which is used to control the created comprehensive evolution index G k The weight ratio of the innovation vector modulus and G k Compare and get the adaptive coefficient α k , α k >1 indicates that more (less) state (observation) data should be used in the state recursion process; and α k <1 indicates that more (less) observation (state) data should be used in the state recursion process, which is calculated as follows:

[0085]

[0086]

[0087] in, is the innovation vector of the kth state The transpose of . b is the forgetting factor, which is set to 0.95 by default.

[0088] Then according to α k Q k Adaptively adjust to obtain the prior state error covariance matrix and the prediction error covariance matrix P k∣k-1 After completing the prediction step, the calculation is as follows:

[0089]

[0090] in, is the state transition matrix F of the kth state k The transpose of P k-1∣k-1 is the state error covariance matrix of the k-1th state, P k∣k-1 is the prior state error covariance matrix, and k is the number of observation vectors.

[0091] Step 6-3, perform the update step, first according to the obtained α k Rk Perform adaptive adjustment to obtain the Kalman gain K k , calculated as follows:

[0092]

[0093] Then get the updated value of the k-th state prediction value And the k-th state posterior state error covariance matrix P k∣k .

[0094]

[0095] P k∣k =(IK k H k )P k∣k-1 (14)

[0096] Step 6-4, if the current (kth) observation vector z k If the fault diagnosis result of the last classification model is obtained, the final state vector is output, otherwise steps 6-2 and 6-3 are repeated.

[0097] (7) Output the fault diagnosis results. Select the fault type with the highest probability from the final state vector as the output result.

[0098] The prediction step primarily predicts the next state. The update step uses the observation vector to modify the state prediction obtained in the previous step. The observation vector is the fault diagnosis result output by different classification models.

[0099] For example, there are three adaptive Kalman filters, wherein adaptive Kalman filter 1 is adaptive Kalman filter 2 after updating its parameters, and adaptive Kalman filter 2 is adaptive Kalman filter 3 after updating its parameters. z1, z2, z3 represent the fault diagnosis result sequence determined after sorting the final fault diagnosis results corresponding to each classification model, such as Figure 5 shown.

[0100] Among them, the adaptive Kalman filter multi-model fusion. First, initialize the parameters of the adaptive Kalman filter Q0 to 0.6*I6, R0 to 0.2*I6, x0 to 0*l6, P0 to 0.8*I6, F k and H k The default setting is I6. The fault diagnosis probability result of BPNN is selected as the observation vector of adaptive Kalman filter 1, and the adaptive coefficient α is obtained in the prediction step. k The noise covariance matrix Q of the posterior process k Update; then in the update step according to the obtained α k Rk Adaptive adjustment is performed to obtain the state filter value. The prediction and update steps are then repeated, and the fault diagnosis probability results of the KNN and RF classification models are used as the observation vectors of adaptive Kalman filter 2 and adaptive Kalman filter 3 respectively. Finally, the state vector of adaptive Kalman filter 3 is output.

[0101] Finally, the average diagnostic accuracy results of 30 times were obtained, and the results are shown in Table 2. Table 2 is the average diagnostic accuracy results of 30 times.

[0102] Table 2 Average diagnostic accuracy results

[0103] How to use Diagnostic accuracy BPNN 89.8％ KNN 79.8％ RF 88.2％ Adaptive Kalman filter multi-model fusion 92.2％

[0104] This application takes into account that the weights of the fault diagnosis results obtained by each classification model are not consistent. Therefore, the credibility weights of each classification model are established, and the mean values ​​of the accuracy, precision and consistency indicators of each classification model are used as the credibility weights of the fault probability obtained by each model, accurately quantifying the proportion of each model in the fusion process. Then, the adaptive Kalman filter multi-model fusion method is used to use the adaptive coefficient α in the fusion process. k Update process noise covariance matrix Q k and the observation noise covariance matrix R k , that is, through the adaptive coefficient α k The key parameter Q in the adaptive Kalman filter fusion process k and R k Automatic correction solves the inaccurate Q in the Kalman filter fusion process k and R k , which leads to unreasonable use of state prediction values ​​and observation vectors, resulting in inaccurate state recursion. The present application makes more reasonable use of state prediction values ​​and observation vectors, making state recursion more accurate and fault diagnosis results more accurate.

[0105] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0106] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the above method when executed by a processor.

[0107] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0108] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRdM), magnetic random access memory (MRdM), ferroelectric random access memory (FRdM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RdM) or external cache memory, etc. By way of illustration and not limitation, RdM may be in various forms, such as static random access memory (SRdM) or dynamic random access memory (DRdM).

[0109] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0110] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for diagnosing a power transformer fault, characterized in that: The power transformer fault diagnosis method comprises: After normalizing the gas analysis data of the power transformer oil, multiple observation state data are determined; each observation state data represents a gas concentration data; Input multiple observation state data as samples into multiple classification models respectively, and determine the fault diagnosis results corresponding to each classification model; Calculating the credibility weight of each classification model according to the fault diagnosis results corresponding to each classification model, and determining the final fault diagnosis results corresponding to each classification model based on the credibility weight of each classification model; Based on the adaptive Kalman filter, the final fault diagnosis results corresponding to each classification model are fused to determine the power transformer fault diagnosis result; the power transformer fault diagnosis result includes the fault diagnosis result corresponding to each fault type.

2. The power transformer fault diagnosis method according to claim 1, characterized in that: According to the fault diagnosis results corresponding to each classification model, the credibility weight of each classification model is calculated, and based on the credibility weight of each classification model, the final fault diagnosis result corresponding to each classification model is determined, specifically including: For any classification model, calculate the accuracy index, precision index and consistency index of the classification model; Determining the credibility weight of the classification model according to the accuracy index, precision index and consistency index corresponding to the classification model; Based on the credibility weight of each classification model and the fault diagnosis result corresponding to each classification model, the final fault diagnosis result corresponding to each classification model is determined.

3. The power transformer fault diagnosis method according to claim 2, characterized in that: Determining the credibility weight of the classification model based on the accuracy index, precision index, and consistency index corresponding to the classification model, specifically including: Determine the total number of samples correctly identified for each type of fault in the classification model based on the fault diagnosis results of the classification model and the actual fault labels of the samples; Determine the accuracy index corresponding to each fault type in the classification model based on the total number of correctly identified samples corresponding to each fault type and the total number of samples; Determine the accuracy index corresponding to each fault type in the classification model based on the total number of fault types, the number of samples correctly identified for each fault type, and the number of samples identified for each fault type; Determining the consistency index corresponding to each fault type in the classification model based on the total number of fault types, the number of samples correctly identified for each fault type, and the total number of samples for each fault type; According to the mean value of the accuracy index, precision index and consistency index corresponding to each fault type in the classification model, the mean value is used as the credibility weight of the classification model.

4. The power transformer fault diagnosis method according to claim 1, characterized in that: Based on the adaptive Kalman filter, the final fault diagnosis results corresponding to each classification model are integrated to determine the power transformer fault diagnosis results, including: Sorting the final fault diagnosis results corresponding to each classification model to determine a fault diagnosis result sequence; a final fault diagnosis result in the fault diagnosis result sequence is used as an observation vector; Based on the initialized adaptive Kalman filter, determine the initial state vector and the initial state error covariance matrix; According to the fault diagnosis result sequence, each observation vector in the fault diagnosis result sequence is used as the current observation vector; Inputting the current observation vector, the initial state vector and the initial state error covariance matrix into an adaptive Kalman filter to determine a current state prediction value of the current observation vector; Determine a current innovation vector based on the current state prediction value and the current observation vector; Determine the current state error covariance matrix based on the current innovation vector; The current state prediction value and the current state error covariance matrix of the current observation vector are used as the initial state prediction value and the initial state error covariance matrix of the next observation vector, respectively, until the current state prediction value of the last observation vector is determined; the current state prediction value of the last observation vector is the fused current state prediction value; The fused current state prediction value is used as a final state vector, and a power transformer fault diagnosis result is determined based on the final state vector.

5. The power transformer fault diagnosis method according to claim 4, characterized in that: According to the current innovation vector, the current state error covariance matrix is ​​determined, specifically including: Determining a current adaptive coefficient according to the current innovation vector; The noise covariance matrix and the observation noise covariance matrix of the initial adaptive Kalman filter are adaptively adjusted according to the current adaptive coefficient to determine the current state error covariance matrix.

6. The power transformer fault diagnosis method according to claim 5, characterized in that: The current adaptive coefficient is: Among them, α k is the current adaptive coefficient; k is the number of observation vectors; is the new information vector corresponding to the k-th observation vector; T is the transpose; b is the forgetting factor; G k is a customized comprehensive evolution indicator; λ is a weighting coefficient; is the new information vector corresponding to the lth observation vector.

7. The power transformer fault diagnosis method according to claim 5, characterized in that: Adaptively adjust the noise covariance matrix and the observation noise covariance matrix of the initial adaptive Kalman filter according to the current adaptive coefficient to determine the current state error covariance matrix, specifically including: Adaptively adjust the noise covariance matrix in the adaptive Kalman filter according to the current adaptive coefficient to determine the prior state error covariance matrix; Based on the prior state error covariance matrix, adaptively adjusting the observation noise covariance matrix in the adaptive Kalman filter according to the current adaptive coefficient to determine the posterior state error covariance matrix; The posterior state error covariance matrix is ​​used as the current state error covariance matrix.

8. The power transformer fault diagnosis method according to claim 1, characterized in that: The multiple classification models include at least a BPNN classifier, a KNN classifier, and a RF classifier.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the power transformer fault diagnosis method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the power transformer fault diagnosis method according to any one of claims 1 to 8 is implemented.

Images