Transformer fault diagnosis method based on improved neural network

Abstract

The invention discloses a transformer fault diagnosis method based on an improved neural network, and provides a fault diagnosis technology based on an improved algorithm aiming at the problems of low efficiency and strong diagnosis result subjectivity of traditional transformer fault diagnosis. The method comprises the following steps: firstly, preprocessing voiceprint data of a transformer, further processing preprocessed voiceprint features through a fuzzy clustering C-means clustering algorithm, and labeling the voiceprint features to obtain a data set; and then constructing a neural network model optimized based on the improved He-Map algorithm, optimizing parameters of the neural network through the He-Map algorithm, and improving search and convergence performance of the model. According to the method, the efficiency and accuracy of fault diagnosis are improved, the generalization ability of the model is enhanced, and powerful technical support is provided for maintenance and fault prevention of the transformer.

Description

Technical Field

[0001] The invention belongs to the technical field of partial discharge detection of electric power equipment, and in particular relates to a transformer fault diagnosis method based on an improved neural network. Background Art

[0002] In the power transmission and distribution system, the transformer is a key device for realizing voltage level conversion. Its stable and reliable operation is crucial to the safe and efficient operation of the entire power system. However, since the transformer bears heavy load changes and voltage regulation tasks in the power network, its operating environment is complex and changeable, and it is easily affected by various factors and fails. The traditional transformer management method mainly relies on emergency repairs after failures and regular manual inspections. This method is not only inefficient, but also highly dependent on the experience and judgment of technicians. There is a risk of misjudgment and it is difficult to ensure that the transformer is maintained at the best time, which may aggravate the degree of failure and affect the overall stability and reliability of the power system.

[0003] Traditional transformer fault diagnosis relies on the experience of technicians and predicts faults through preventive data analysis and experimental deduction. However, with the development of technology, fault diagnosis methods based on intelligent algorithms are gradually emerging. This method uses pattern recognition and intelligent algorithms, combined with computer automation processing, to more quickly and accurately identify fault types and even achieve fault warnings, significantly improving diagnostic efficiency and accuracy.

[0004] However, in the process of using intelligent algorithms to obtain fault types, the selection of algorithm or model parameters greatly affects the detection results. For example, a Chinese patent with the patent publication number CN117970783B discloses a CNC high-speed drilling and milling machine control method based on an improved Hippo algorithm. However, the introduction of the improved Hippo algorithm with a triangle walk strategy improves the global search capability by adding additional calculation steps and improving the exploration capability of the search space, but this inevitably increases the computational complexity of the Hippo algorithm, resulting in an extension of the algorithm execution time. This is not conducive to the acquisition of real-time fault diagnosis results for transformers. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a transformer fault diagnosis method based on an improved neural network, comprising the following steps: Step S1: preprocessing the transformer voiceprint data, further processing the preprocessed voiceprint features through the fuzzy clustering C-means clustering algorithm, marking the voiceprint features and obtaining a data set; Step S2: construct a BP neural network model, import the voiceprint data set in the data set in step S1 into the BP neural network model for training, and during the training process, optimize the learning rate of the BP neural network model through the Hippo algorithm; Step S3: Acquire the voiceprint features of the transformer in real time, import the real-time voiceprint features into the trained BP neural network model, and obtain the predicted fault type; In step S2, the learning rate of the BP neural network model is optimized by the Hippo algorithm, specifically: Step S21: Initialize the population and the predator position; Step S22: Entering the river; Step S23: patrolling the territory; Step S24: Escape from the predator, during which a triangle wandering strategy is introduced; Step S25: Determine whether the maximum number of iterations has been reached. If so, calculate the fitness of the hippopotamus at the current number of iterations, and select the hippopotamus with the minimum fitness as the optimal learning rate. If not, continue with steps S22 to S25.

[0006] Furthermore, step S1 is specifically as follows: Step S11: Acquire the voiceprint data of the transformer, and extract voiceprint features therefrom by using the Fourier transform method; clean the extracted voiceprint features to remove noise and outliers; Step S12: introducing the fuzzy C-means clustering algorithm to classify the voiceprint data; Step S13: Formalize the samples of transformer voiceprint data into a pattern matrix and set the objective function. The smaller the function value, the better the clustering effect. It is expressed as:

[0007] ; in, Indicates the pattern matrix Sample pair The membership degree of the cluster centers is is the fuzzy index, which is used to control the fuzziness of the membership. , For the The samples and The distance between cluster centers, is the sample size; The Lagrange multiplier method is introduced to iteratively optimize the membership matrix. If the change of the membership matrix is ​​less than or equal to the threshold, the iteration is stopped; it is expressed as: ; in, is the membership of the i-th sample point to the k-th cluster center, is the sample in the pattern matrix To the kth cluster center The distance is the pattern matrix To the jth cluster center distance, represents the preset weight, , Indicates the number of clusters.

[0008] Furthermore, the fault types in step S1 include: no fault, low-risk fault, medium-risk fault and high-risk fault.

[0009] Furthermore, step S21 is specifically as follows: The position of the first hippopotamus is initialized by a pseudo-random sequence, and then the positions of the remaining hippopotamus are initialized by chaotic mapping. The initialization formula is expressed as: ; ; in, represents the initial position of the i-th hippopotamus, is the initial position of the i+1th hippopotamus, , represents the total number of hippos in the hippo population, Represents the lower bound of the hippopotamus position, Indicates the upper limit of the hippopotamus position, represents a random number between [0,1]. Indicates a random number between (0, 0.5); A nonlinear convergence factor is introduced to update the predator position and improve the global search capability; it is expressed as: ; ; in, represents the position of the predator in the t+1th iteration, is the position of the predator in the tth iteration, is the nonlinear convergence factor, is the nonlinear adjustment coefficient, t is the current iteration number, is the maximum number of iterations; Indicates the current predator position; Indicates the current hippo position.

[0010] Furthermore, step S22 is specifically as follows: Simulate the behavior of half of the hippopotamus in the hippopotamus population entering the river; expressed as: ; ; in, Indicated in The updated position of the ith male hippopotamus during the first entry into the river, Indicates The position of the i-th hippopotamus before the update during the first entry into the river, represents a random number between [0,1]. Indicates The best position of the hippopotamus during the iteration, and are all random vectors with values ​​between [1,2]. For the The updated position of the i-th female hippopotamus during the iteration, is a random vector with values ​​between [0,1]. is the average value of the position of each hippo in the current hippo population; Determine the fitness function, expressed as: ; in, represents the fitness, e(t) is the deviation between the predicted fault type and the actual detection result, Indicates the current iteration number, Indicates the number of iterations of the calculation; Get the fitness of each hippopotamus position after the update, and select the position with the smallest fitness as the latest position of the hippopotamus in this round of evolution through greedy selection; expressed as: ; ; in, Indicated in The latest position of the i-th hippopotamus during the iteration, represents the fitness function, Indicates The fitness corresponding to the hippopotamus position during the iteration, Indicated in The fitness corresponding to the updated position of the i-th male hippopotamus during the first entry into the river; Indicated in The fitness of the updated position of the i-th female hippopotamus during the iteration.

[0011] Further, step S23 is specifically as follows: Simulate the hippopotamus patrolling in its own territory, so that the predator can be aware of the behavior of being in the hippopotamus territory; expressed as: ; in, Indicated in The updated position of the i-th hippopotamus during the patrol, RL represents the random vector with Rhine flight, Indicated in The updated position of the predator during the patrol. Represents a random number between [2,4], Represents a random number between [1,1.5], Represents a random number between [0,1], represents a random number between [2,3], represents a random vector with values ​​between [0,1], represents a random vector with values ​​between [0,1]. Indicated in The fitness corresponding to the predator's updated position during the patrol.

[0012] Further, step S24 is specifically as follows: Simulate the stage where the hippopotamus escapes from predators. When the hippopotamus cannot resist the predator, it will choose to leave the area and enter the nearest river or pond area to avoid being harmed by the predator; expressed as: ; in, Indicates The updated position of the hippopotamus during the escape process. represents a random vector with values ​​between [0,1]. Represents the lower bound of the hippopotamus position, Represents the upper limit value at the hippopotamus position, Represents random numbers that conform to the normal distribution; Introduce the triangle walking strategy and update the individual position of the hippopotamus, expressed as: ; in, Indicates the introduction of the triangle walk strategy The position of the i-th hippopotamus after updating in the escape process, Indicates the introduction of the triangle walk strategy The position of a randomly selected hippopotamus during the escape process, represents a random number between [0,1]. Represents a random number between [0,1]; In this process, As the gradient, the Adagrad gradient descent algorithm is introduced to optimize the vertex selection strategy, dynamically adjust the step size when selecting random vertices, and reduce the complexity of the triangle walk strategy calculation, which is expressed as: ; in, is the initial learning rate, is the sum of squared gradients of the first n-1 times, is the sum of squared gradients of the previous n times, s is initially 0, , for ; After optimization, the formula for the random vertex position corresponding to the gradient is: ; in, is the position of the random vertex at the nth selection, The position of the random vertex when selecting for the n+1th time.

[0013] The positive and progressive effects of the present invention are:

[0014] The present invention learns the voiceprint data of the transformer and uses the improved Hippo algorithm to optimize the BP neural network model to realize online detection of transformer fault types, thereby improving the search and convergence performance of the model, further improving the efficiency and accuracy of fault diagnosis, and enhancing the generalization ability of the model, providing strong technical support for transformer maintenance and fault prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention is a flowchart of the steps of a transformer fault diagnosis method based on an improved neural network. DETAILED DESCRIPTION

[0016] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0017] Reference Figure 1 In one example, a transformer fault diagnosis method based on an improved neural network includes the following steps: Step S1: preprocessing the transformer voiceprint data, further processing the preprocessed voiceprint features through the fuzzy clustering C-means clustering algorithm, marking the voiceprint features and obtaining a data set; Step S2: construct a BP neural network model, import the voiceprint data set in the data set in step S1 into the BP neural network model for training, and during the training process, optimize the learning rate of the BP neural network model through the Hippo algorithm; Step S3: Acquire the voiceprint features of the transformer in real time, import the real-time voiceprint features into the trained BP neural network model, and obtain the predicted fault type; In step S2, the learning rate of the BP neural network model is optimized by the Hippo algorithm, specifically: Step S21: Initialize the population and the predator position; Step S22: Entering the river; Step S23: patrolling the territory; Step S24: Escape from the predator, during which a triangle wandering strategy is introduced; Step S25: Determine whether the maximum number of iterations has been reached. If so, calculate the fitness of the hippopotamus at the current number of iterations, and select the hippopotamus with the minimum fitness as the optimal learning rate. If not, continue with steps S22 to S25.

[0018] Furthermore, in one example, step S1 is specifically as follows: Step S11: Acquire the voiceprint data of the transformer, and extract voiceprint features therefrom by using the Fourier transform method; clean the extracted voiceprint features to remove noise and outliers; Step S12: introducing the fuzzy C-means clustering algorithm to classify the voiceprint data; Step S13: Formalize the samples of transformer voiceprint data into a pattern matrix and set the objective function. The smaller the function value, the better the clustering effect. It is expressed as:

[0019] ; in, Indicates the pattern matrix Sample pair The membership degree of the cluster centers is is the fuzzy index, which is used to control the fuzziness of the membership. , For the The samples and The distance between cluster centers, is the sample size; The Lagrange multiplier method is introduced to iteratively optimize the membership matrix. If the change of the membership matrix is ​​less than or equal to the threshold, the iteration is stopped; it is expressed as: ; in, is the membership of the i-th sample point to the k-th cluster center, is the sample in the pattern matrix To the kth cluster center The distance is the pattern matrix To the jth cluster center distance, represents the preset weight, , Indicates the number of clusters.

[0020] Further, in one example, the fault types in step S1 include: no fault, low-risk fault, medium-risk fault, and high-risk fault.

[0021] Furthermore, in one example, step S21 is specifically as follows: The position of the first hippopotamus is initialized by a pseudo-random sequence, and then the positions of the remaining hippopotamus are initialized by chaotic mapping. The initialization formula is expressed as: ; ; in, represents the initial position of the i-th hippopotamus, is the initial position of the i+1th hippopotamus, , represents the total number of hippos in the hippo population, Represents the lower bound of the hippopotamus position, Indicates the upper limit of the hippopotamus position, represents a random number between [0,1]. Indicates a random number between (0, 0.5); A nonlinear convergence factor is introduced to update the predator position and improve the global search capability; it is expressed as: ; ; in, represents the position of the predator in the t+1th iteration, is the position of the predator in the tth iteration, is the nonlinear convergence factor, is the nonlinear adjustment coefficient, t is the current iteration number, is the maximum number of iterations; Indicates the current predator position; Indicates the current hippo position.

[0022] Furthermore, in one example, step S22 is specifically as follows: Simulate the behavior of half of the hippopotamus in the hippopotamus population entering the river; expressed as: ; ; in, Indicated in The updated position of the ith male hippopotamus during the first entry into the river, Indicates The position of the i-th hippopotamus before the update during the first entry into the river, represents a random number between [0,1]. Indicates The best position of the hippopotamus during the iteration, and are all random vectors with values ​​between [1,2]. For the The updated position of the i-th female hippopotamus during the iteration, is a random vector with values ​​between [0,1]. is the average value of the position of each hippo in the current hippo population; Determine the fitness function, expressed as: ; in, represents the fitness, e(t) is the deviation between the predicted fault type and the actual detection result, Indicates the current iteration number, Indicates the number of iterations of the calculation; Get the fitness of each hippopotamus position after the update, and select the position with the smallest fitness as the latest position of the hippopotamus in this round of evolution through greedy selection; expressed as: ; ; in, Indicated in The latest position of the i-th hippopotamus during the iteration, represents the fitness function, Indicates The fitness corresponding to the hippopotamus position during the iteration, Indicated in The fitness corresponding to the updated position of the i-th male hippopotamus during the first entry into the river; Indicated in The fitness of the updated position of the i-th female hippopotamus during the iteration.

[0023] Further, in one example, step S23 is specifically as follows: Simulate the hippopotamus patrolling in its own territory, so that the predator can be aware of the behavior of being in the hippopotamus territory; expressed as: ; in, Indicated in The updated position of the i-th hippopotamus during the patrol, RL represents the random vector with Rhine flight, Indicated in The updated position of the predator during the patrol. Represents a random number between [2,4], Represents a random number between [1,1.5], Represents a random number between [0,1], represents a random number between [2,3], represents a random vector with values ​​between [0,1], represents a random vector with values ​​between [0,1]. Indicated in The fitness corresponding to the predator's updated position during the patrol.

[0024] Further, in one example, step S24 is specifically as follows: Simulate the stage where the hippopotamus escapes from predators. When the hippopotamus cannot resist the predator, it will choose to leave the area and enter the nearest river or pond area to avoid being harmed by the predator; expressed as: ; in, Indicates The updated position of the hippopotamus during the escape process. represents a random vector with values ​​between [0,1]. Represents the lower bound of the hippopotamus position, Represents the upper limit value at the hippopotamus position, Represents random numbers that conform to the normal distribution; Introduce the triangle walking strategy and update the individual position of the hippopotamus, expressed as: ; in, Indicates the introduction of the triangle walk strategy The position of the i-th hippopotamus after updating in the escape process, Indicates the introduction of the triangle walk strategy The position of a randomly selected hippopotamus during the escape process, represents a random number between [0,1]. Represents a random number between [0,1]; In this process, As the gradient, the Adagrad gradient descent algorithm is introduced to optimize the vertex selection strategy, dynamically adjust the step size when selecting random vertices, and reduce the complexity of the triangle walk strategy calculation, which is expressed as: ; in, is the initial learning rate, is the sum of squared gradients of the first n-1 times, is the sum of squared gradients of the previous n times, s is initially 0, , for ; After optimization, the formula for the random vertex position corresponding to the gradient is: ; in, is the position of the random vertex at the nth selection, The position of the random vertex when selecting for the n+1th time.

[0025] The present invention is described in detail above in conjunction with the embodiments of the accompanying drawings. A person skilled in the art can make various variations of the present invention according to the above description. Therefore, some details in the embodiments should not be construed as limiting the present invention, and the present invention shall be protected by the scope defined by the attached claims.

Claims

1. A transformer fault diagnosis method based on improved neural network, characterized in that: The following steps are involved: Step S1: preprocessing the transformer voiceprint data, further processing the preprocessed voiceprint features through the fuzzy clustering C-means clustering algorithm, marking the voiceprint features and obtaining a data set; Step S2: construct a BP neural network model, import the voiceprint data set in the data set in step S1 into the BP neural network model for training, and during the training process, optimize the learning rate of the BP neural network model through the Hippo algorithm; Step S3: Acquire the voiceprint features of the transformer in real time, import the real-time voiceprint features into the trained BP neural network model, and obtain the predicted fault type; In step S2, the learning rate of the BP neural network model is optimized by the Hippo algorithm, specifically: Step S21: Initialize the population and the predator position; Step S22: Entering the river; Step S23: patrolling the territory; Step S24: Escape from the predator, during which a triangle wandering strategy is introduced; Step S25: Determine whether the maximum number of iterations has been reached. If so, calculate the fitness of the hippopotamus at the current number of iterations, and select the hippopotamus with the minimum fitness as the optimal learning rate. If not, continue with steps S22 to S25.

2. A transformer fault diagnosis method based on an improved neural network as claimed in claim 1, characterized in that: Step S1 is specifically as follows: Step S11: Acquire the voiceprint data of the transformer, and extract voiceprint features therefrom by using the Fourier transform method; clean the extracted voiceprint features to remove noise and outliers; Step S12: introducing the fuzzy C-means clustering algorithm to classify the voiceprint data; Step S13: Formalize the samples of transformer voiceprint data into a pattern matrix and set the objective function. The smaller the function value, the better the clustering effect. The objective function It is expressed as: ； in, Indicates the pattern matrix Sample pair The membership degree of the cluster centers is is the fuzzy index, which is used to control the fuzziness of the membership. , For the The samples and The distance between cluster centers, is the sample size; The Lagrange multiplier method is introduced to iteratively optimize the membership matrix. If the change of the membership matrix is ​​less than or equal to the threshold, the iteration is stopped; it is expressed as: ； in, is the membership of the i-th sample point to the k-th cluster center, is the sample in the pattern matrix To the kth cluster center The distance is the pattern matrix To the jth cluster center distance, represents the preset weight, , Indicates the number of clusters.

3. A transformer fault diagnosis method based on an improved neural network as claimed in claim 1, characterized in that: The fault types in step S1 include: no fault, low risk fault, medium risk fault and high risk fault.

4. A transformer fault diagnosis method based on an improved neural network as claimed in claim 3, characterized in that: Step S21 is specifically as follows: The position of the first hippopotamus is initialized by a pseudo-random sequence, and then the positions of the remaining hippopotamus are initialized by chaotic mapping. The initialization formula is expressed as: ； ； in, represents the initial position of the i-th hippopotamus, is the initial position of the i+1th hippopotamus, , represents the total number of hippos in the hippo population, represents the lower bound of the hippopotamus position, Indicates the upper limit of the hippopotamus position, represents a random number between [0,1]. Indicates a random number between (0, 0.5); A nonlinear convergence factor is introduced to update the predator position and improve the global search capability; it is expressed as: ； ； in, represents the updated position of the predator in the t+1th iteration, is the updated position of the predator in the tth iteration, is the nonlinear convergence factor, is the nonlinear adjustment coefficient, t is the current iteration number, is the maximum number of iterations; Indicates the current predator position; Indicates the current hippo position.

5. A transformer fault diagnosis method based on an improved neural network as claimed in claim 4, characterized in that: Step S22 is specifically as follows: Simulate the behavior of half of the hippopotamus in the hippopotamus population entering the river; expressed as: ； ； in, Indicated in The updated position of the ith male hippopotamus during the first entry into the river, Indicates The position of the i-th hippopotamus before the update during the first entry into the river, represents a random number between [0,1]. Indicates The best position of the hippopotamus during the iteration, and are all random vectors with values ​​between [1,2]. For the The updated position of the i-th female hippopotamus during the iteration, is a random vector with values ​​between [0,1]. is the average value of the position of each hippo in the current hippo population; Determine the fitness function, expressed as: ； in, represents the fitness, e(t) is the deviation between the predicted fault type and the actual detection result, Indicates the current iteration number, Indicates the number of iterations of the calculation; Get the fitness of each hippopotamus position after the update, and select the position with the smallest fitness as the latest position of the hippopotamus in this round of evolution through greedy selection; expressed as: ； ； in, Indicated in The latest position of the i-th hippopotamus during the iteration, represents the fitness function, Indicates The fitness corresponding to the hippopotamus position during the iteration, Indicated in The fitness corresponding to the updated position of the i-th male hippopotamus during the first entry into the river; Indicated in The fitness of the updated position of the i-th female hippopotamus during the iteration.

6. A transformer fault diagnosis method based on improved neural network as claimed in claim 5, characterized in that: Step S23 is specifically as follows: Simulate the hippopotamus patrolling in its own territory, so that the predator can be aware of the behavior of being in the hippopotamus territory; expressed as: ； in, Indicated in The updated position of the i-th hippopotamus during the patrol, RL represents the random vector with Rhine flight, Indicated in The updated position of the predator during the patrol. Represents a random number between [2,4], Represents a random number between [1,1.5], Represents a random number between [0,1], represents a random number between [2,3], represents a random vector with values ​​between [0,1], represents a random vector with values ​​between [0,1]. Indicated in The fitness corresponding to the predator's updated position during the patrol.

7. A transformer fault diagnosis method based on improved neural network as claimed in claim 6, characterized in that: Step S24 is specifically as follows: Simulate the stage where the hippopotamus escapes from predators. When the hippopotamus cannot resist the predator, it will choose to leave the river or pond area where the hippopotamus is currently located and enter the nearest river or pond area to avoid being harmed by the predator; expressed as: ； in, Indicates The updated position of the hippopotamus during the escape process. represents a random vector with values ​​between [0,1]. represents the lower bound of the hippopotamus position, Represents the upper limit value at the hippopotamus position, Represents random numbers that conform to the normal distribution; Introduce the triangle walking strategy and update the individual position of the hippopotamus, expressed as: ； in, Indicates the introduction of the triangle walk strategy The position of the i-th hippopotamus after updating in the escape process, Indicates the introduction of the triangle walk strategy The position of a randomly selected hippopotamus during the escape process, represents a random number between [0,1]. Represents a random number between [0,1]; In this process, As the gradient, the Adagrad gradient descent algorithm is introduced to optimize the vertex selection strategy, dynamically adjust the step size when selecting random vertices, and reduce the complexity of the triangle walk strategy calculation, which is expressed as: ； in, is the initial learning rate, is the sum of squared gradients of the first n-1 times, is the sum of squared gradients of the previous n times, s is initially 0, , for ; After optimization, the formula for the random vertex position corresponding to the gradient is: ； in, is the position of the random vertex at the nth selection, The position of the random vertex when selecting for the n+1th time.

Images