Transformer fault diagnosis method and system based on multi-information feature fusion

Abstract

The invention discloses a transformer fault diagnosis method and system based on multi-information feature fusion, and the method comprises the steps: collecting transformer operation information and a transformer voiceprint signal, and carrying out the filtering processing of the transformer voiceprint signal through a self-adaptive unscented Kalman filtering algorithm; a penalty factor C of the SVM is optimized through an improved snake egret optimization algorithm, and an ISBOA-SVM model is constructed; inputting the data into the constructed ISBOA-SVM model, and obtaining the probabilities of various fault types; and fusing the probability of the fault type output by each ISBOA-SVM model by adopting a D-S evidence theory, and determining the fault type. According to the method provided by the invention, the transformer fault diagnosis precision can be remarkably improved, and the generalization capability is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer fault diagnosis, and in particular to a transformer fault diagnosis method and system based on multi-information feature fusion. Background Art

[0002] Power transformers are one of the important equipment in the power system, and their safe operation ensures the safety and stability of the power system. Oil-immersed transformers are a widely used type of transformer. Therefore, the faults of oil-immersed power transformers can be accurately diagnosed, so that accurate repairs can be carried out according to the fault type, thereby ensuring the stability of the power system and reducing damage to society.

[0003] At present, the intelligent diagnosis of transformer fault types is mainly based on the gas components dissolved in the transformer oil when the transformer fault occurs. That is, the gas components dissolved in the transformer oil when the fault occurs are used as feature vectors, and a nonlinear mapping relationship between the feature vectors and the transformer fault types is established. The corresponding fault types are diagnosed by detecting and analyzing the gas components dissolved in the transformer oil when the transformer fails (Dissolved Gasses Analysis, DGA). However, this method is limited to the scope of threshold diagnosis, and often requires the content of certain characteristic gases to exceed the threshold before the diagnosis result can be obtained, so that it is difficult to diagnose the fault type in time at the early stage of the fault. In recent years, with the development of artificial intelligence research, artificial neural networks, Bayesian classifiers, support vector machines and other methods have begun to be widely used in the field of transformer faults, but these methods are generally used in use, and the fault diagnosis rate is not high. Summary of the invention

[0004] The purpose of the present invention is to provide a transformer fault diagnosis method and system based on multi-information feature fusion. Collect transformer operation information and use a sound collection device to collect transformer voiceprint signals; filter the collected signals through AUKF adaptive filtering to reduce the interference of external sounds on the transformer voiceprint signals, and then use the improved snake-bearing optimization algorithm (ISBOA) to optimize the penalty factor C of SVM; input the data into the constructed ISBOA-SVM model to obtain the probability of various fault types; finally, use the DS evidence theory to fuse the output results of each ISBOA-SVM model to determine the fault type. The method proposed in the present invention can significantly improve the accuracy of transformer fault diagnosis and has stronger generalization ability.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a transformer fault diagnosis method based on multi-information feature fusion, comprising the following steps:

[0006] Step 1: Collect transformer operation information and transformer voiceprint signals, and filter the transformer voiceprint signals through an adaptive unscented Kalman filter (AUKF) algorithm;

[0007] Step 2: Optimize the penalty factor C of SVM by improving the snake heron optimization algorithm and build the ISBOA-SVM model;

[0008] Step 3: Input the data into the constructed ISBOA-SVM model to obtain the probability of various fault types;

[0009] Step 4: Use DS evidence theory to fuse the probabilities of fault types output by each ISBOA-SVM model to determine the fault type.

[0010] Further preferably, the steps of the adaptive unscented Kalman filter algorithm include initialization, Sigma point selection, prediction and update; after each update, the adaptive unscented Kalman filter algorithm dynamically updates the process noise covariance and the observation noise covariance by observing the prediction error.

[0011] Further preferably, the penalty factor C of SVM is optimized by improving the snake heron optimization algorithm in step 2, and the process of constructing the ISBOA-SVM model is as follows:

[0012] 2.1. Set the initial population size and maximum number of iterations;

[0013] 2.2. Initialize the snake heron optimization algorithm, use the penalty factor C of SVM as the initial population, and introduce the Logistic-Sine-Cosine chaotic map to initialize the population;

[0014] 2.3. Evaluate the fitness of each individual;

[0015] 2.4, the stage of searching for prey;

[0016] 2.5. Prey consumption stage;

[0017] 2.6. In the attack prey phase, when the optimal value falls into a local optimal solution, a dynamic refraction reverse learning mechanism is introduced to disturb the population so that it jumps out of the local optimal solution;

[0018] 2.7. Escape strategy;

[0019] 2.8. Determine whether the current number of iterations reaches the preset maximum number of iterations. If so, stop iterating. If not, return to step 2.3 to continue evaluating and updating.

[0020] 2.9. Output the optimal individual, which is the penalty factor C of the optimal SVM.

[0021] Further preferably, in step 3, the data obtained in step 1 are normalized and then input into the ISBOA-SVM model.

[0022] Further preferably, the process of step 4 is as follows:

[0023] 4.1. Constructing the recognition framework:

[0024] Θ={A 1 ,A 2 ,…,A o ,…,A g};

[0025] Among them, A o The primitive called Θ represents the oth fault type, o = 1, 2, …, g; g is the total number of fault types;

[0026] 4.2. Determine the support mapping relationship between the feature evidence of the information group and each transformer state in the identification framework, and calculate the basic probability amplitude of the feature evidence for different faults; the basic probability amplitude for different faults is:

[0027] m l (A o )=α l u lo ;

[0028]

[0029] In the formula, m l (A o ) is the probability distribution function of the lth information source for the oth type of fault identification result; α l is the reliability coefficient of the lth type of information source; u lo is the membership degree of the lth information source for the output type o fault;

[0030] 4.3. Use the synthesis rule to synthesize these two sets of feature evidence into new evidence c. The basic probability amplitude synthesis rule calculation formula of evidence c is as follows:

[0031]

[0032] Among them, m(c) represents the probability after fusion, M represents the conflict degree between feature evidence A and B, and m l is the probability distribution function of the lth information source;

[0033] 4.4. Use the DS evidence theory calculation rules to fuse the support of different basic propositions in the identification framework of each feature to achieve comprehensive evaluation of transformer equipment fault diagnosis based on multi-information feature fusion;

[0034] 4.5. The basic proposition corresponding to the maximum probability value in the fusion support is taken as the final diagnosis result.

[0035] The present invention also provides a non-volatile computer storage medium, which stores computer executable instructions. The computer executable instructions can execute the above-mentioned transformer fault diagnosis method based on multi-information feature fusion.

[0036] The present invention also provides an adaptive distributed power data anomaly detection system, 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 a transformer fault diagnosis method based on multi-information feature fusion.

[0037] Beneficial effects of the present invention: The present invention utilizes the correlation between data and data, and fuses them through DS theory. On the premise of effectively reducing the conflict between the two types of evidence, the fault information provided by the two types of evidence can be more effectively complemented. Compared with a single SVM diagnosis algorithm, the accuracy of the diagnosis results of the algorithm in this paper is significantly improved, which fully verifies the effectiveness of multi-information fusion for fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of the transformer fault diagnosis method based on multi-information feature fusion.

[0039] Figure 2 Flowchart of the optimization algorithm for Snake Heron.

[0040] Figure 3 Improved comparison chart for the Snake Heron optimization algorithm. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0042] Reference Figure 1 ,The transformer fault diagnosis method based on multi-information feature fusion includes the following steps:

[0043] Step 1: Collect transformer operation information and transformer voiceprint signals, and filter the transformer voiceprint signals through an adaptive unscented Kalman filter algorithm (AUKF) to reduce the interference of external sounds on the transformer voiceprint signals; the transformer voiceprint signals are collected through a sound collection device;

[0044] Step 2: Optimize the penalty factor C of SVM by improving the snake heron optimization algorithm and build the ISBOA-SVM model;

[0045] Step 3: Input the data into the constructed ISBOA-SVM model to obtain the probability of various fault types;

[0046] Step 4: Use DS evidence theory to fuse the probabilities of fault types output by each ISBOA-SVM model to determine the fault type.

[0047] In this embodiment, the main steps of the adaptive unscented Kalman filter algorithm include initialization, Sigma point selection, prediction and update. After each update, the adaptive unscented Kalman filter algorithm will perform an adjustment step to dynamically update the process noise covariance and the observation noise covariance by observing the prediction error. The basic formula for dynamically adjusting the process noise covariance and the observation noise covariance is as follows:

[0048]

[0049] In the formula, Q w new and R w new is the updated element of the process noise covariance matrix and the observation noise covariance matrix, Q w old and R w old is the element of the process noise covariance matrix and the observation noise covariance matrix before updating, P w and P wprev are the state covariance of the current and previous steps, y w and The noise observations are divided into actual and predicted values.

[0050] This embodiment uses Logistic-Sine-Cosine chaotic mapping to improve the problem of uneven initial population distribution of the snake-heron optimization algorithm (ISBOA). Aiming at the problem of weak search ability in the later stage of the algorithm, the third-stage formula is improved to enhance the search ability. Finally, because it is easy to fall into the local optimal value, a dynamic refraction reverse learning mechanism is used to perturb the population.

[0051] The penalty factor determines the degree of penalty for misclassified samples of SVM (support vector machine), which can coordinate the classification interval and sample misclassification rate, and take into account the confidence level and empirical risk of SVM. When the penalty factor is large, SVM will increase the penalty for misclassification, requiring the algorithm to meet all constraints, and not allowing misclassification, which reduces the fitting error, but will lead to weak generalization ability of SVM, and higher classification standards will greatly prolong the training time of SVM model; when the penalty factor is small, SVM will not pay too much attention to the penalty for sample misclassification, which greatly improves the generalization ability, but the fitting error increases, which will bring greater empirical risk. Therefore, the ISBOA algorithm is used to optimize the penalty parameter C of SVM (support vector machine) and construct the ISBOA-SVM model.

[0052] Reference Figure 2 In step 2 of this embodiment, the penalty factor C of SVM is optimized by improving the snake heron optimization algorithm, and the process of constructing the ISBOA-SVM model is as follows:

[0053] 2.1. Set the initial population size and maximum number of iterations;

[0054] 2.2. Initialize the snake heron optimization algorithm, take the penalty factor C of SVM as the initial population, introduce the Logistic-Sine-Cosine chaotic map to initialize the population, and solve the uneven distribution problem caused by random initialization:

[0055]

[0056] In the formula, X represents the initialization population after Logistic-Sine-Cosine chaotic mapping, represents the initial value of the i-th snake heron in the j-th dimension, x i,j Represents the value of the i-th snake heron in the j-th dimension after Logistic-Sine-Cosine chaotic mapping, i∈1,2,…,n; j∈1,2,…,d; n is the population size, d is the dimension of the problem; each snake heron represents a parameter solution of the penalty factor C of a set of SVM, ub and lb are the upper and lower bounds of the problem respectively, Rand represents a random number between 0 and 1, and r is a random number.

[0057] 2.3. Evaluate the fitness of each individual. The quality of the individual is the performance of each set of SVM parameters. Its fitness function is:

[0058]

[0059] Where C is the penalty factor, ξ i is the relaxation factor.

[0060] 2.4. Prey hunting phase:

[0061] when It is the first stage, and the position is updated as follows:

[0062]

[0063] Among them, t represents the current number of iterations, T represents the maximum number of iterations, and X i (t) is the position of the i-th snake heron before updating, X i (t+1) represents the updated position of the i-th snake heron, denotes two random candidate solutions for the first stage iteration, R 1 A random array between 0 and 1;

[0064] Compare the two updated results. If it is better than the previous update result, use the updated value. If it is not better than the previous update result, keep the original value, as shown in the following formula:

[0065]

[0066] in, and They represent the fitness values ​​of the i-th snake heron before and after the position update, X i is the current feasible solution, is the updated feasible solution.

[0067] 2.5. Prey consumption stage:

[0068] when It is the second stage, and the position is updated as follows:

[0069] X i (t+1)=X best +exp((t / T) 4 )×(RB-0.5)×(X best -X i (t));

[0070]

[0071] Among them, RB = randn (1, Dim) represents an array of dimension 1 × Dim randomly generated from a standard normal distribution (mean value is 0, standard deviation is 1), X best Indicates the current optimal value;

[0072] 2.6. Attacking prey stage:

[0073] Aiming at the shortcoming of the original algorithm's weak late search capability, it is improved to achieve the purpose of improving the search capability.

[0074] when It is the third stage, and the position is updated as follows:

[0075]

[0076] C 1 =2a×r 1 -a;

[0077] C 2 =2r 2 ;

[0078]

[0079] Among them, r 1 and r 2 is a random number between 0 and 1, X rand is a random individual, · represents the dot product, RL is the Levy flight operator, a is a constant that changes with the number of iterations, C 1 is and a, r 2 The related constant, C 2 for and r 2 Related constants;

[0080] When the algorithm enters the third stage When the optimal value falls into the local optimal solution, a dynamic refraction reverse learning mechanism is introduced on the basis of the original algorithm to disturb the population so that it jumps out of the local optimal solution. The formula is:

[0081]

[0082] In the formula, is the position of the ith individual after dynamic refraction reverse learning, and k is the adjustment parameter.

[0083] 2.7. Escape strategy:

[0084]

[0085] Where r = 0.5, R 2 represents an array randomly generated from a normal distribution, X random represents a random candidate solution for the current iteration, and K represents a random selection of an integer 1 or 2.

[0086] 2.8. Determine whether the current number of iterations reaches the preset maximum number of iterations. If so, stop iterating. If not, return to step 2.3 to continue evaluating and updating.

[0087] 2.9. Output the optimal individual, which is the penalty factor C of the optimal SVM.

[0088] The convergence curves of the improved snake heron optimization algorithm before and after improvement are as follows Figure 3 As shown in the figure, the convergence speed of the improved snake heron optimization algorithm (ISBOA) is faster than that of the original snake heron optimization algorithm (SBOA).

[0089] In step 3, the data is normalized and input into the constructed ISBOA-SVM model to complete the "one-to-one" classification of fault types and obtain the probability of various fault types. The input data matrix is:

[0090]

[0091] In the formula, I 1,t represents the voiceprint signal at time t, I 2,t Represents the first transformer operation information at time t, I s,t Represents the s-1th transformer operation information at time t.

[0092] Normalize the data matrix to eliminate the dimension and magnitude effects, and change the weight of the variables in the analysis to solve the problem of different measurements. The formula is as follows:

[0093]

[0094] In the formula, I nor is the normalized value, I i is the value before normalization, I max is the maximum value of the data, I min is the minimum value of the data.

[0095] Specifically, in step 4, the DS evidence theory is used to fuse the probabilities of the fault types output by each ISBOA-SVM model, and the process of determining the fault type is as follows:

[0096] 4.1. Constructing the recognition framework:

[0097] Θ={A 1 ,A 2 ,…,A o ,…,A g};

[0098] Among them, A o The primitive called Θ represents the oth fault type, o = 1, 2, …, g; g is the total number of fault types;

[0099] 4.2. Determine the support mapping relationship between the feature evidence of the information group and each transformer state in the identification framework, and calculate the basic probability amplitude of the feature evidence for different faults; the basic probability amplitude for different faults is:

[0100] m l (A o )=α l ulo ;

[0101]

[0102] In the formula, m l (A o ) is the probability distribution function of the lth information source for the oth type of fault identification result; α l is the reliability coefficient of the l-th information source, and the reliability coefficient is replaced by the test accuracy of the l-th information source alone under the ISBOA-SVM model; u lo is the membership degree of the lth information source for the output type o fault.

[0103] 4.3. Use the synthesis rule to synthesize these two sets of feature evidence into new evidence c. The basic probability amplitude of evidence c is calculated by the synthesis rule formula as follows:

[0104]

[0105] Where m(c) represents the probability after fusion, and M represents the degree of conflict between feature evidence A and B. When M = 1, it means that the two feature evidences are completely conflicting and cannot be synthesized; when M approaches 1, it means that the two feature evidences are highly conflicting, which may lead to inaccurate synthesis results. l is the probability distribution function of the lth information source;

[0106] 4.4. Use the DS evidence theory calculation rules to fuse the support of different basic propositions in the identification framework of each feature to achieve comprehensive evaluation of transformer equipment fault diagnosis based on multi-information feature fusion;

[0107] 4.5. The basic proposition corresponding to the maximum probability value in the fusion support is taken as the final diagnosis result.

[0108] In another embodiment, a non-volatile computer storage medium is provided, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the transformer fault diagnosis method based on multi-information feature fusion in the above embodiment.

[0109] The present embodiment provides an adaptive distributed power data anomaly detection system, 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 a transformer fault diagnosis method based on multi-information feature fusion.

[0110] Finally, it should be noted that the above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A transformer fault diagnosis method based on multi-information feature fusion, characterized in that: The following steps are involved: Step 1: Collect transformer operation information and transformer voiceprint signals, and filter the transformer voiceprint signals through an adaptive unscented Kalman filter algorithm; Step 2: Optimize the penalty factor C of SVM by improving the snake heron optimization algorithm and build the ISBOA-SVM model; Step 3: Input the data into the constructed ISBOA-SVM model to obtain the probability of various fault types; Step 4: Use DS evidence theory to fuse the probabilities of fault types output by each ISBOA-SVM model to determine the fault type.

2. The transformer fault diagnosis method based on multi-information feature fusion according to claim 1 is characterized in that: The steps of the adaptive unscented Kalman filter algorithm include initialization, Sigma point selection, prediction and update; The adaptive unscented Kalman filter algorithm dynamically updates the process noise covariance and the observation noise covariance by observing the prediction error after each update.

3. The transformer fault diagnosis method based on multi-information feature fusion according to claim 1 is characterized in that: In step 2, the penalty factor C of SVM is optimized by improving the snake heron optimization algorithm, and the process of constructing the ISBOA-SVM model is as follows: 2.

1. Set the initial population size and maximum number of iterations; 2.

2. Initialize the snake heron optimization algorithm, use the penalty factor C of SVM as the initial population, and introduce the Logistic-Sine-Cosine chaotic map to initialize the population; 2.

3. Evaluate the fitness of each individual; 2.4, the stage of searching for prey; 2.

5. Prey consumption stage; 2.

6. In the attack prey phase, when the optimal value falls into a local optimal solution, a dynamic refraction reverse learning mechanism is introduced to disturb the population so that it jumps out of the local optimal solution; 2.

7. Escape strategy; 2.

8. Determine whether the current number of iterations reaches the preset maximum number of iterations. If so, stop iterating. If not, return to step 2.3 to continue evaluating and updating. 2.

9. Output the optimal individual, which is the penalty factor C of the optimal SVM.

4. The transformer fault diagnosis method based on multi-information feature fusion according to claim 1 is characterized in that: In step 3, the data obtained in step 1 are normalized and then input into the ISBOA-SVM model.

5. The transformer fault diagnosis method based on multi-information feature fusion according to claim 1 is characterized in that: The process of step 4 is as follows: 4.

1. Constructing the recognition framework: Θ={A1,A2,,A o ,…,A g }; Among them, A o The primitive called Θ represents the oth fault type, o = 1, 2, …, g; g is the total number of fault types; 4.

2. Determine the support mapping relationship between the feature evidence of the information group and each transformer state in the identification framework, and calculate the basic probability amplitude of the feature evidence for different faults; the basic probability amplitude for different faults is: m l (A o )=a l you lo ; In the formula, m l (A o ) is the probability distribution function of the lth information source for the oth type of fault identification result; α l is the reliability coefficient of the lth type of information source; u lo is the membership degree of the lth information source for the output type o fault; 4.

3. Use the synthesis rule to synthesize these two sets of feature evidence into new evidence c. The basic probability amplitude synthesis rule calculation formula of evidence c is as follows: Among them, m(c) represents the probability after fusion, M represents the conflict degree between feature evidence A and B, and m l is the probability distribution function of the lth information source; 4.

4. Use the DS evidence theory calculation rules to fuse the support of different basic propositions in the identification framework of each feature to achieve comprehensive evaluation of transformer equipment fault diagnosis based on multi-information feature fusion; 4.

5. The basic proposition corresponding to the maximum probability value in the fusion support is taken as the final diagnosis result.

6. The transformer fault diagnosis method based on multi-information feature fusion according to claim 1 is characterized in that: when It is the attack prey stage, and the position is updated as follows: C1=2a×r1-a; C2＝2r2； Among them, t represents the current iteration number, T represents the maximum iteration number, Rand represents a random number between 0 and 1, and X i (t) is the position of the i-th snake heron before updating, X i (t+1) represents the updated position of the i-th snake heron, r1 and r2 are random numbers between 0 and 1, and X rand is a random individual, representing point product, RL is the Levy flight operator, a is a constant that changes with the number of iterations, C1 is a constant related to a and r2, and C2 is a constant related to r2; and They represent the fitness values ​​of the i-th snake heron before and after the position update, X i is the current feasible solution, X i new is the updated feasible solution; When the optimal value falls into a local optimal solution, a dynamic refraction reverse learning mechanism is introduced to perturb the population so that it jumps out of the local optimal solution. The formula is: In the formula, X i * is the position of the ith individual after dynamic refraction reverse learning; ub and lb are the upper and lower bounds of the problem respectively, and k is the adjustment parameter.

7. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions execute the transformer fault diagnosis method based on multi-information feature fusion as described in any one of claims 1-6.

8. An adaptive distributed power data anomaly detection system, 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, characterized in that the instructions are executed by the at least one processor so that the at least one processor executes the transformer fault diagnosis method based on multi-information feature fusion as described in any one of claims 1-6.

Images