Power transformer fault diagnosis method and system based on improved ant algorithm

Abstract

The invention relates to the technical field of power diagnosis, in particular to a power transformer fault diagnosis method and system based on an improved ant algorithm, and the method comprises the steps: collecting the operation state data of a power transformer, and carrying out the signal processing of the operation state data; performing data mining and feature extraction on the data after signal processing to obtain the features of the fault state of the power transformer; based on an ant algorithm and an elite strategy mechanism, the pheromone cumulative effect of the high-quality path is enhanced, and positive feedback is applied to the optimal path in the current iteration; and after each iteration is completed, aiming at the currently found optimal solution, adopting a local search algorithm to further optimize the optimal solution, and updating the current optimal solution to obtain a fault diagnosis result. According to the invention, the problems of low response speed and low diagnosis accuracy of an existing fault diagnosis method are effectively solved, and the efficiency and precision of power transformer fault diagnosis are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power diagnosis, and in particular to a method and system for diagnosing faults of electric power transformers based on an improved ant algorithm. Background Art

[0002] Power transformers are key equipment in the power system, and the stability of their operating status is directly related to the safety of the power grid. However, traditional transformer fault diagnosis methods mostly rely on periodic maintenance and offline testing. These methods have obvious deficiencies in time efficiency and response speed, and it is difficult to meet the needs of modern power systems for efficient and real-time diagnosis.

[0003] In this context, intelligent optimization algorithms are gradually introduced into the field of fault diagnosis. Ant Colony Optimization (ACO) is an intelligent algorithm that achieves path optimization by simulating the foraging behavior of ants. It has strong global search capabilities and complex problem-solving capabilities. However, traditional ant colony algorithms have exposed some limitations in practical applications, such as slow convergence speed and insufficient solution quality, which limits their application effect in complex problems. Summary of the invention

[0004] The present invention provides a power transformer fault diagnosis method and system based on an improved ant algorithm, thereby effectively solving the problems pointed out in the background technology.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A power transformer fault diagnosis method based on an improved ant algorithm, comprising:

[0007] Collecting operating status data of the power transformer and performing signal processing on the operating status data;

[0008] Perform data mining and feature extraction on the signal processed data to obtain the characteristics of the power transformer fault state;

[0009] Based on the ant algorithm and elite strategy mechanism, the pheromone accumulation effect of high-quality paths is strengthened, and positive feedback is applied to the optimal path in the current iteration;

[0010] After each iteration is completed, a local search algorithm is used to further optimize the currently found optimal solution, and the current optimal solution is updated to obtain a fault diagnosis result.

[0011] Furthermore, the ant algorithm is improved based on the elite strategy mechanism, including:

[0012] The optimal path found during the current algorithm iteration is recorded as BFS;

[0013] A positive feedback value is set based on the original pheromone increment. The positive feedback is an incentive for ants to take the optimal path. The formula is: Among them, stimulate is the positive feedback value, m is the total number of ants, L bsf The length of the currently known optimal path, n is the incentive weight hyperparameter;

[0014] The positive feedback value is added to the pheromone increment formula to obtain the improved pheromone increment update formula. Δτ ij is the total increment of pheromone on path (i, j), represents the sum of the pheromone increments of all ants on the path (i, j) in this iteration, k is the current ant number, It represents the pheromone increment of the kth ant on the path (i, j) in this iteration.

[0015] Furthermore, signal processing is performed on the operating status data, including: denoising filtering and normalization operations.

[0016] Furthermore, data mining and feature extraction are performed on the signal processed data to obtain the characteristics of the power transformer fault state, including:

[0017] Perform frequency domain analysis on the signal through Fourier transform, extract the frequency component and obtain the abnormal signal;

[0018] Use wavelet transform to perform time-frequency analysis, capture the changing characteristics of the signal over time, and reveal local fault information;

[0019] Automatically extract high-dimensional features based on deep learning methods and learn complex failure modes;

[0020] The results of frequency domain, time domain and deep learning extraction are comprehensively analyzed to screen out the key features that characterize the fault status of the power transformer.

[0021] Furthermore, high-dimensional features are automatically extracted based on deep learning methods to learn complex failure modes, including:

[0022] Construct a convolutional neural network to extract local spatial features of the signal and capture the deep patterns of the data through multi-layer convolution and pooling operations;

[0023] Use spatiotemporal convolutional networks to model the time series characteristics of signals and analyze the dynamic changes of data in the time dimension;

[0024] Combined with the autoencoder for feature dimensionality reduction and reconstruction, the most representative key features are extracted from high-dimensional data, and finally a deep learning feature vector representing the complex fault state is generated.

[0025] Furthermore, after each iteration is completed, a local search algorithm is used to further optimize the currently found optimal solution, and the current optimal solution is updated to obtain a fault diagnosis result, including:

[0026] By using a local search algorithm, the optimal solution found in the current iteration is adjusted and optimized in a local range;

[0027] Improving the quality of the optimal solution by adjusting the path sequence and structure;

[0028] The solution optimized by local search is compared with the current optimal solution. If the optimized solution is better than the current optimal solution, the current optimal solution is updated with the optimized solution, and finally the fault diagnosis result is generated.

[0029] Furthermore, the method also includes: adjusting the pheromone concentration by using a dynamic pheromone volatilization mechanism, maintaining a low volatilization rate in the early stage of algorithm iteration by setting a dynamic change rule of the volatilization coefficient, and increasing the volatilization rate in the later stage of the algorithm iteration to accelerate convergence.

[0030] Furthermore, the method also includes: performing cluster analysis on characteristic data of multiple types of fault states, and classifying similar fault modes into the same category.

[0031] A power transformer fault diagnosis system based on an improved ant algorithm, the system comprising:

[0032] A data acquisition module, which collects the operating status data of the power transformer and performs signal processing on the operating status data;

[0033] The feature extraction module performs data mining and feature extraction on the data after signal processing to obtain the characteristics of the fault state of the power transformer;

[0034] Iterative optimization module, based on ant algorithm and elite strategy mechanism, strengthens the pheromone accumulation effect of high-quality paths and applies positive feedback to the optimal path in the current iteration;

[0035] The result acquisition module, after each iteration, uses a local search algorithm to further optimize the currently found optimal solution, and updates the current optimal solution to obtain a fault diagnosis result.

[0036] Furthermore, the result acquisition module includes:

[0037] A local optimization unit, which adjusts and optimizes the optimal solution found in the current iteration in a local range through a local search algorithm;

[0038] A quality improvement unit, which improves the quality of the optimal solution by adjusting the path sequence and structure;

[0039] The optimal solution comparison unit compares the solution optimized by the local search with the current optimal solution. If the optimized solution is better than the current optimal solution, the current optimal solution is updated with the optimized solution, and finally the fault diagnosis result is generated.

[0040] The technical solution of the present invention can achieve the following technical effects:

[0041] It effectively solves the problems of slow response speed and low diagnostic accuracy of existing fault diagnosis methods, and significantly improves the efficiency and accuracy of power transformer fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a flowchart of a method for fault diagnosis of power transformer based on improved ant algorithm;

[0044] Figure 2 A schematic diagram of a process for obtaining characteristics of a power transformer fault state;

[0045] Figure 3 A schematic diagram of the process of automatically extracting high-dimensional features and learning complex failure modes based on deep learning methods;

[0046] Figure 4 A flow chart for optimizing the optimal solution and obtaining fault diagnosis results. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0049] Embodiment 1

[0050] like Figure 1 As shown, the present invention provides a method for diagnosing power transformer faults based on an improved ant algorithm, the method comprising:

[0051] S1: Collecting the operating status data of the power transformer and performing signal processing on the operating status data;

[0052] Specifically, through the sensor equipment arranged on the power transformer, key operating parameters including current, voltage, oil temperature, vibration signals, etc. are collected in real time to ensure comprehensive monitoring of the transformer's operating status. The collected raw data is pre-processed and converted into a unified numerical range to facilitate subsequent analysis; then, the signal is processed. After the signal processing is completed, the data is checked for integrity and consistency, and abnormal or lost data is eliminated to ensure that the final data is accurate and reliable, providing a high-quality input basis for feature extraction and fault diagnosis.

[0053] S2: Perform data mining and feature extraction on the signal processed data to obtain the characteristics of the fault state of the power transformer;

[0054] Specifically, by deeply analyzing and mining the signal-processed data, key features that can characterize the operating status and potential fault modes of the power transformer are extracted. Through the extraction of these features, the complex raw data is converted into more representative high-dimensional features, which facilitates the subsequent diagnostic model to accurately identify and classify the fault status, and ultimately improves the accuracy and efficiency of power transformer fault diagnosis. The core goal of this step is to convert a large amount of operating status data into useful information that can be directly used for diagnosis, laying a data foundation for realizing intelligent fault diagnosis.

[0055] S3: Based on the ant algorithm and elite strategy mechanism, the pheromone accumulation effect of high-quality paths is strengthened, and positive feedback is applied to the optimal path in the current iteration;

[0056] Specifically, by improving the path selection mechanism of the ant algorithm and strengthening the pheromone accumulation effect of the optimal path, more ants tend to choose high-quality paths, thereby accelerating the convergence of the algorithm; at the same time, by applying positive feedback to the optimal path in the current iteration, the pheromone concentration of the optimal path can be further improved, ensuring that the algorithm can more efficiently discover potential optimal solutions during global search.

[0057] S4: After each iteration is completed, a local search algorithm is used to further optimize the optimal solution currently found, and the current optimal solution is updated to obtain a fault diagnosis result.

[0058] Specifically, a local search algorithm is used to perform more refined optimization in the neighborhood of the optimal solution to further improve the quality of the solution and avoid the traditional ant algorithm from falling into the local optimum due to insufficient search accuracy. Local search discovers potential better solutions by fine-tuning and optimizing the current optimal path, thereby improving the accuracy and reliability of the final diagnostic results.

[0059] The present invention effectively solves the problems of slow response speed and low diagnostic accuracy of existing fault diagnosis methods, and significantly improves the efficiency and accuracy of power transformer fault diagnosis.

[0060] As a preferred embodiment of the above, the ant algorithm is improved based on the elite strategy mechanism, including:

[0061] The optimal path found during the current algorithm iteration is recorded as BFS;

[0062] A positive feedback value is set based on the original pheromone increment. Positive feedback is an incentive for ants to take the optimal path. The formula is: Among them, stimulate is the positive feedback value, m is the total number of ants, L bsf The length of the currently known optimal path, n is the incentive weight hyperparameter;

[0063] The positive feedback value is added to the pheromone increment formula to obtain the improved pheromone increment update formula. Δτ ij is the total increment of pheromone on path (i, j), represents the sum of the pheromone increments of all ants on the path (i, j) in this iteration, k is the current ant number, It represents the pheromone increment of the kth ant on the path (i, j) in this iteration.

[0064] Specifically, in each iteration, the currently found optimal path is recorded and recorded as BFS (Base SoFar). The optimal path BFS is the path with the shortest total length among all the paths found by ants in the current iteration, representing the best solution that the algorithm can currently find. The purpose of marking BFS is to use it as a key path to strengthen the distribution of pheromones, and to increase the probability of ants choosing this path through a positive feedback mechanism, thereby further optimizing the quality of the solution and the convergence efficiency; based on the pheromone update mechanism of the traditional ant algorithm, a positive feedback value stimulate is set for the path belonging to the optimal path BFS to incentivize ants that choose the optimal path. The core of this step is to strengthen the pheromone concentration of the optimal path through a positive feedback mechanism, so that subsequent ants are more inclined to choose this path. If it belongs to the optimal path, positive feedback is given, otherwise, the positive feedback value is 0; in the traditional ant algorithm, the pheromone increment Δτ of the path (i, j) is ij is the sum of pheromone contributions of all ants on the path. Based on this, this embodiment superimposes the positive feedback value stimulate into the pheromone increment formula to obtain an improved pheromone update formula. By introducing the positive feedback value stimulate into the pheromone update formula, the pheromone concentration of the optimal path is further enhanced. The improved pheromone update mechanism significantly enhances the ant algorithm's ability to focus on high-quality paths, making the algorithm more inclined to explore and optimize the optimal path; at the same time, due to the dynamic adjustment of the positive feedback value, the algorithm can strengthen the distribution of pheromones in real time according to the currently found optimal path, thereby accelerating the convergence speed, reducing invalid searches, and improving the efficiency of the algorithm in complex optimization problems and the quality of solutions.

[0065] As a preferred embodiment of the above embodiment, signal processing is performed on the operating status data, including: denoising filtering and normalization operations.

[0066] Specifically, the collected operating status data is first subjected to denoising and filtering, and external environmental noise and non-fault signal interference are removed using low-pass filtering, high-pass filtering or band-pass filtering techniques. For example, low-pass filtering can be used to eliminate high-frequency noise, while band-pass filtering is used to extract signals within a frequency range associated with specific fault characteristics, thereby retaining signal components that are useful for fault diagnosis. Subsequently, the filtered data is normalized to map data of different dimensions or amplitudes to a unified numerical range (such as [0,1] or [-1,1]). Normalization uses minimum-maximum normalization or Z-score normalization methods to eliminate the impact of amplitude differences between different signals on the analysis, thereby improving data comparability and processing efficiency. De-noising filtering and normalization operations effectively improve data quality, laying a solid foundation for subsequent feature extraction and fault diagnosis.

[0067] As a preferred embodiment of the above, Figure 2As shown, step S2, data mining and feature extraction are performed on the signal processed data to obtain the features of the power transformer fault state, including:

[0068] S21: Perform frequency domain analysis on the signal through Fourier transform, extract the frequency component, and obtain the abnormal signal;

[0069] S22: Use wavelet transform to perform time-frequency analysis to capture the changing characteristics of the signal over time and reveal local fault information;

[0070] S23: Automatically extract high-dimensional features based on deep learning methods and learn complex failure modes;

[0071] S24: Comprehensively analyze the results of frequency domain, time domain and deep learning extraction to screen out the key features that characterize the fault status of the power transformer.

[0072] Specifically, first, the signal is analyzed in the frequency domain through Fourier transform, and the signal is converted from the time domain to the frequency domain. By extracting the frequency components of the signal, the abnormal frequency characteristics related to the power transformer fault are identified. Frequency domain analysis can effectively detect spectrum anomalies caused by faults, such as increased harmonics or changes in spectrum peaks, and preliminarily reveal the frequency distribution characteristics of the fault signal; then, the signal is analyzed in time and frequency using wavelet transform. Through wavelet transform, the local characteristics of the signal changing with time are captured, thereby revealing the dynamic characteristics of the fault signal in time and frequency. Wavelet transform has obvious advantages in processing non-stationary signals, and can locate the time point of the fault signal and its corresponding frequency characteristics to further mine local fault information; then, a deep neural network is constructed to extract features from the preprocessed signal data. The deep learning method can learn complex patterns in signals through a multi-layer network structure, capture potential fault features that are difficult to detect with traditional signal processing methods, and provide a deeper feature representation for the fault state; finally, a comprehensive analysis of the results of frequency domain, time domain and deep learning extraction is performed to screen out key features that can characterize the fault state of the power transformer. By cross-validating and screening features of different dimensions, redundant or irrelevant information is removed, and the features with the most diagnostic value are retained, providing accurate input for the training and application of subsequent fault diagnosis models.

[0073] As a preferred embodiment of the above, Figure 3 As shown, step S23, automatically extracting high-dimensional features based on deep learning methods and learning complex failure modes, includes:

[0074] S231: Construct a convolutional neural network to extract local spatial features of the signal, and capture the deep patterns of the data through multi-layer convolution and pooling operations;

[0075] S232: Use spatiotemporal convolutional networks to model the time series characteristics of signals and analyze the dynamic changes of data in the time dimension;

[0076] S233: Combined with the autoencoder to perform feature dimensionality reduction and reconstruction, the most representative key features are extracted from high-dimensional data, and finally a deep learning feature vector is generated to characterize the complex fault state.

[0077] Specifically, first, a convolutional neural network (CNN) is constructed to extract the local spatial features of the signal. By designing multi-layer convolution kernels, the input signal is convolved layer by layer to capture the local pattern features in the signal. At the same time, the feature dimension is reduced through pooling operations. The convolution operation can extract the key edge features and local patterns in the signal, while the pooling operation downsamples the signal, thereby reducing the computational complexity while retaining important features and avoiding overfitting. Finally, the deep features generated by CNN can characterize the spatial distribution characteristics of the signal. Then, a spatiotemporal convolutional network (STCN) is used to perform time series modeling on the signal, combining the time dimension of the input signal with the spatial features, and using the spatiotemporal convolutional layer to simultaneously capture the dynamic change characteristics and spatial characteristics of the signal. The time dimension convolution of the convolutional network can extract the changing law of the signal in the time series, while the spatial convolution can retain the local features. Through this network, the coupling characteristics of the signal in time and space are accurately extracted, thereby revealing the dynamic evolution law of complex fault signals; then, the extracted high-dimensional features are reduced and reconstructed in combination with the autoencoder. The signal is mapped to the low-dimensional feature space after the encoder part, and the key features are extracted while removing redundant information. The decoder reconstructs the low-dimensional features to verify the accuracy and completeness of the extracted features. This step strengthens the feature extraction capability of complex fault modes through dimensionality reduction operations, and further optimizes the dimension of the feature data, which is convenient for efficient processing of subsequent diagnostic models.

[0078] As a preferred embodiment of the above, Figure 4 As shown, in step S4, after each iteration, a local search algorithm is used to further optimize the optimal solution currently found, and the current optimal solution is updated to obtain a fault diagnosis result, including:

[0079] S41: adjusting and optimizing the optimal solution found in the current iteration locally through a local search algorithm;

[0080] S42: Improve the quality of the optimal solution by adjusting the path order and structure;

[0081] S43: Compare the solution optimized by the local search with the current optimal solution. If the optimized solution is better than the current optimal solution, update the current optimal solution with the optimized solution, and finally generate a fault diagnosis result.

[0082] Specifically, the local search algorithm can use classic optimization methods such as 2-opt, 3-opt or greedy algorithm. These methods explore possible better solutions by adjusting path nodes or parameter configurations in a local range. The local search is based on the current optimal solution and searches its neighborhood to find a better solution.

[0083] Take the 2-opt algorithm as an example. This is a local search algorithm commonly used for path optimization, which can significantly improve the quality of the current solution. In power transformer fault diagnosis, the 2-opt algorithm can be used to optimize the order of feature analysis or parameter adjustment to improve the efficiency and accuracy of the diagnosis model. The principle of the 2-opt algorithm is to reduce the length of the path or improve the quality of the solution by selecting two nodes in the path and reversing the path segment between them. Specifically, the algorithm first starts with the current optimal solution, such as the path P = {A, B, C, D, E}, and then randomly selects two nodes, such as B and D, and reverses the path segment {C, D} between them to generate a new path P′ = {A, D, C, B, E}; then, evaluate the objective function value of the new path P′ (such as the total length of the path or the fault feature matching degree). If the new path is better, it replaces the current solution. By repeatedly performing this operation, the 2-opt algorithm can continuously optimize the current solution until no better solution can be found or the upper limit of the number of iterations is reached. In the power transformer fault diagnosis scenario, for example, when optimizing the feature analysis order, the 2-opt algorithm can adjust the analysis order of fault features from P = {F1, F2, F3, F4} to P′ = {F1, F3, F2, F4}, thereby reducing computational overhead or improving diagnostic accuracy. The 2-opt algorithm, with its simplicity and efficiency, can effectively avoid the dilemma of redundant paths or non-optimal solutions in the local search process, providing a better optimization strategy for the diagnostic model.

[0084] Next, the quality of the optimal solution is further improved by adjusting the path order and structure. For example, in the path optimization problem, local search can adjust the order of nodes in the path, or insert, delete or replace certain nodes in the path, thereby reducing the path length or improving the fitness of the solution. In the power transformer fault diagnosis scenario, these adjustments may correspond to more accurate matching of fault characteristics or optimization of model parameters. Subsequently, the solution optimized by local search is compared with the current optimal solution. If the optimized solution is better than the current optimal solution in quality, the optimized solution is used to update the current optimal solution. If there is no improvement, the current optimal solution is retained. This process ensures the effectiveness of iterative optimization while avoiding unnecessary update operations, thereby improving search efficiency.

[0085] As a preferred embodiment of the above embodiment, the method also includes: adjusting the pheromone concentration by using a dynamic pheromone volatilization mechanism, maintaining a low volatilization rate in the early stage of the algorithm iteration by setting a dynamic change rule of the volatilization coefficient, and increasing the volatilization rate in the later stage of the algorithm iteration to accelerate convergence.

[0086] Specifically, in the initial stage of the algorithm, in order to ensure the exploration of more solution spaces, a lower pheromone evaporation rate is set, and the retention time of pheromones on the path is prolonged to encourage ants to make more attempts on unexplored paths, thereby enhancing the algorithm's global search ability and avoiding falling into local optimal solutions; as the iterations proceed, the pheromone evaporation rate is gradually increased, and the pheromone concentration on low-quality paths is reduced by accelerating the evaporation mechanism, further strengthening the pheromone accumulation effect on the optimal path, promoting the ant colony to concentrate on a better path, and accelerating the convergence of the algorithm. This dynamic adjustment mechanism achieves a balance between global exploration and rapid convergence by adjusting the rules of the pheromone evaporation rate at different stages, improving the performance of the algorithm in complex optimization problems, and providing higher-quality optimization results for the fault diagnosis model.

[0087] As a preferred embodiment of the above embodiment, the method further includes: performing cluster analysis on characteristic data of various types of fault states, and classifying similar fault modes into the same category.

[0088] Specifically, the high-dimensional feature data obtained in the signal processing and feature extraction steps are first used to group the fault feature data through a clustering algorithm (such as K-means clustering, hierarchical clustering, or density clustering). The clustering algorithm classifies data with similar feature patterns into the same category by measuring the similarity between feature data. For example, for data in different fault states, multiple fault modes are dynamically divided by calculating the Euclidean distance or other similarity indicators in the feature space; then, the clustering results are evaluated and verified to ensure the accuracy and consistency of the classification, thereby providing a reliable basis for subsequent diagnostic decisions.

[0089] Embodiment 2

[0090] Based on the same inventive concept as the power transformer fault diagnosis method based on the improved ant algorithm in the aforementioned embodiment, the present invention also provides a power transformer fault diagnosis system based on the improved ant algorithm, the system comprising:

[0091] A data acquisition module collects the operating status data of the power transformer and performs signal processing on the operating status data;

[0092] The feature extraction module performs data mining and feature extraction on the data after signal processing to obtain the characteristics of the fault state of the power transformer;

[0093] Iterative optimization module, based on ant algorithm and elite strategy mechanism, strengthens the pheromone accumulation effect of high-quality paths and applies positive feedback to the optimal path in the current iteration;

[0094] The result acquisition module, after each iteration, uses a local search algorithm to further optimize the optimal solution currently found, and updates the current optimal solution to obtain the fault diagnosis result.

[0095] The above-mentioned evaluation system in the present invention can effectively implement the power transformer fault diagnosis method based on the improved ant algorithm, and the technical effects that can be achieved are as described in the above-mentioned embodiments, which will not be repeated here.

[0096] As a preferred embodiment of the above, the result acquisition module includes:

[0097] The local optimization unit adjusts and optimizes the optimal solution found in the current iteration locally through the local search algorithm;

[0098] The quality improvement unit improves the quality of the optimal solution by adjusting the path sequence and structure;

[0099] The optimal solution comparison unit compares the solution optimized by local search with the current optimal solution. If the optimized solution is better than the current optimal solution, the optimized solution is used to update the current optimal solution, and finally a fault diagnosis result is generated.

[0100] Similarly, the above-mentioned optimization schemes for the system can also respectively achieve the optimization effects corresponding to the method in Example 1, which will not be repeated here.

[0101] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present application as defined therein, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for power transformer fault diagnosis based on improved ant algorithm, characterized in that: include: Collecting operating status data of the power transformer and performing signal processing on the operating status data; Perform data mining and feature extraction on the signal processed data to obtain the characteristics of the power transformer fault state; Based on the ant algorithm and elite strategy mechanism, the pheromone accumulation effect of high-quality paths is strengthened, and positive feedback is applied to the optimal path in the current iteration; After each iteration is completed, a local search algorithm is used to further optimize the currently found optimal solution, and the current optimal solution is updated to obtain a fault diagnosis result.

2. The power transformer fault diagnosis method based on the improved ant algorithm according to claim 1 is characterized in that: The ant algorithm is improved based on the elite strategy mechanism, including: The optimal path found during the current algorithm iteration is recorded as BFS; A positive feedback value is set based on the original pheromone increment. The positive feedback is an incentive for ants to take the optimal path. The formula is: Among them, stimulate is the positive feedback value, m is the total number of ants, L bsf The length of the currently known optimal path, n is the incentive weight hyperparameter; The positive feedback value is added to the pheromone increment formula to obtain the improved pheromone increment update formula. Δτ ij is the total increment of pheromone on path (i, j), represents the sum of the pheromone increments of all ants on the path (i, j) in this iteration, k is the current ant number, It represents the pheromone increment of the kth ant on the path (i, j) in this iteration.

3. The power transformer fault diagnosis method based on the improved ant algorithm according to claim 1 is characterized in that: The operating status data is subjected to signal processing, including: denoising filtering and normalization operations.

4. The power transformer fault diagnosis method based on the improved ant algorithm according to claim 1 is characterized in that: Data mining and feature extraction are performed on the signal processed data to obtain the characteristics of the power transformer fault state, including: Perform frequency domain analysis on the signal through Fourier transform, extract the frequency component and obtain the abnormal signal; Use wavelet transform to perform time-frequency analysis, capture the changing characteristics of the signal over time, and reveal local fault information; Automatically extract high-dimensional features based on deep learning methods and learn complex failure modes; The results of frequency domain, time domain and deep learning extraction are comprehensively analyzed to screen out the key features that characterize the fault status of the power transformer.

5. The power transformer fault diagnosis method based on improved ant algorithm according to claim 1 is characterized in that: Automatically extract high-dimensional features based on deep learning methods and learn complex failure modes, including: Construct a convolutional neural network to extract local spatial features of the signal and capture the deep patterns of the data through multi-layer convolution and pooling operations; Use spatiotemporal convolutional networks to model the time series characteristics of signals and analyze the dynamic changes of data in the time dimension; Combined with the autoencoder for feature dimensionality reduction and reconstruction, the most representative key features are extracted from high-dimensional data, and finally a deep learning feature vector representing the complex fault state is generated.

6. The power transformer fault diagnosis method based on improved ant algorithm according to claim 1 is characterized in that: After each iteration, the optimal solution currently found is further optimized using a local search algorithm, and the current optimal solution is updated to obtain a fault diagnosis result, including: By using a local search algorithm, the optimal solution found in the current iteration is adjusted and optimized in a local range; Improving the quality of the optimal solution by adjusting the path sequence and structure; The solution optimized by local search is compared with the current optimal solution. If the optimized solution is better than the current optimal solution, the current optimal solution is updated with the optimized solution, and finally the fault diagnosis result is generated.

7. The power transformer fault diagnosis method based on improved ant algorithm according to claim 1 is characterized in that: The method also includes: adjusting the pheromone concentration by using a dynamic pheromone volatilization mechanism, maintaining a low volatilization rate in the early stage of algorithm iteration by setting a dynamic change rule of the volatilization coefficient, and increasing the volatilization rate in the late stage of algorithm iteration to accelerate convergence.

8. The power transformer fault diagnosis method based on improved ant algorithm according to claim 1 is characterized in that: The method further includes: performing cluster analysis on characteristic data of multiple types of fault states, and classifying similar fault modes into the same category.

9. A power transformer fault diagnosis system based on improved ant algorithm, characterized in that: The system comprises: A data acquisition module, which collects the operating status data of the power transformer and performs signal processing on the operating status data; The feature extraction module performs data mining and feature extraction on the data after signal processing to obtain the characteristics of the fault state of the power transformer; Iterative optimization module, based on ant algorithm and elite strategy mechanism, strengthens the pheromone accumulation effect of high-quality paths and applies positive feedback to the optimal path in the current iteration; The result acquisition module, after each iteration, uses a local search algorithm to further optimize the currently found optimal solution, and updates the current optimal solution to obtain a fault diagnosis result.

10. The power transformer fault diagnosis system based on improved ant algorithm according to claim 9, characterized in that: The result acquisition module comprises: A local optimization unit adjusts and optimizes the optimal solution found in the current iteration in a local range through a local search algorithm; a quality improvement unit improves the quality of the optimal solution by adjusting the path sequence and structure; The optimal solution comparison unit compares the solution optimized by the local search with the current optimal solution. If the optimized solution is better than the current optimal solution, the current optimal solution is updated with the optimized solution, and finally the fault diagnosis result is generated.

Images