Winding deformation classification method, system and equipment based on dwarf weasel algorithm and medium
By constructing a multi-source fault sample set and optimizing the dwarf mongoose algorithm, combined with a support vector machine model, the problems of real-time performance and accuracy in winding deformation classification were solved, achieving efficient and reliable winding state characterization and fault type identification.
Patent Information
- Application Number
- CN202511671598.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-02-06
AI Technical Summary
Existing winding deformation classification methods are difficult to meet the real-time diagnostic requirements in practical engineering, cannot fully explore the potential optimal region in the parameter space, exhibit search stagnation in the later stages of iteration, and are difficult to comprehensively characterize the frequency response characteristics changes under different fault modes.
A multi-source fault sample set was constructed, frequency response data of transformer windings were collected through multi-source data acquisition methods, swarm intelligence optimization was performed using the dwarf mongoose algorithm, parameter search was performed based on the population division of labor and cooperation mechanism, classification and discrimination were performed using the support vector machine model, and performance was verified through multi-index evaluation.
It achieves efficient and comprehensive characterization of transformer winding conditions, significantly shortens optimization time, improves classification accuracy and efficiency, and ensures the reliability and robustness of classification results.
Smart Images
Figure CN121479484A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, specifically to a winding deformation classification method, system, equipment, and medium based on the dwarf mongoose algorithm. Background Technology
[0002] Frequency response analysis, as the mainstream method for diagnosing the mechanical condition of windings, identifies potential faults by analyzing the response characteristics of the windings at different frequencies. In recent years, machine learning algorithms such as support vector machines have been widely used in winding deformation classification tasks, and the classification performance largely depends on the rationality of hyperparameter settings. To address this, parameter optimization strategies such as grid search and particle swarm optimization have been introduced, attempting to improve diagnostic accuracy through systematic parameter tuning. Grid search searches for the optimal solution by traversing a predefined parameter space, while particle swarm optimization simulates the cooperative foraging behavior of bird flocks to achieve parameter search. These methods, to some extent, improve the subjectivity and limitations of traditional manual parameter tuning, promoting the development of intelligent winding fault diagnosis.
[0003] However, existing parameter optimization methods still have significant limitations in engineering practice: while grid search can guarantee the integrity of the search within a given range, its computational efficiency increases exponentially with the increase of parameter dimension, making it difficult to meet the real-time requirements of diagnosis in actual engineering; although particle swarm optimization has good global search capabilities, its information exchange mechanism between individuals is relatively simple, making it prone to premature convergence to local optima, especially when dealing with high-dimensional, nonlinear winding feature data, it often fails to fully explore the potential optimal regions in the parameter space; in addition, existing methods generally lack effective population diversity maintenance mechanisms, which can easily lead to search stagnation in the later stages of iteration; on the other hand, traditional diagnostic models do not fully explore the winding state characteristics, often relying on a single type of statistical feature, making it difficult to comprehensively characterize the frequency response characteristics changes under different fault modes; in terms of model validation, existing research focuses more on improving classification accuracy, lacking a systematic evaluation of the comprehensive performance of the algorithm, such as convergence speed and computational resource consumption, which limits its widespread application in actual operation and maintenance scenarios. Summary of the Invention
[0004] In view of the above-mentioned existing problems, the present invention provides a winding deformation classification method, system, device and medium based on the dwarf mongoose algorithm, in order to solve the problems that the existing technology is difficult to meet the real-time requirements of diagnosis in actual engineering, cannot fully explore the potential optimal region in the parameter space, has search stagnation in the later stage of iteration, and is difficult to comprehensively characterize the frequency response characteristics changes under different fault modes.
[0005] To address the aforementioned technical problems, a winding deformation classification method based on the dwarf mongoose algorithm is proposed, including: A multi-source fault sample set is constructed, and frequency response data of transformer windings are collected through multi-source data acquisition methods. Multi-source numerical features are extracted from the frequency response data. A swarm intelligence optimization algorithm is used to adaptively optimize the hyperparameters of the classification model globally, and parameter search is performed based on a swarm division of labor and cooperation mechanism. The optimized classification model is used to classify and distinguish the deformation state of the transformer windings, and the performance advantages are evaluated through a comparative verification process.
[0006] As a preferred embodiment of the winding deformation classification method based on the dwarf mongoose algorithm described in this invention, the construction of the multi-source fault sample set includes: synchronously collecting data through multi-source data acquisition methods, and setting fault severity levels for different fault types. The collected raw data were standardized and preprocessed to eliminate dimensional differences, a sample database was established, and the processed data was divided into training and test sets.
[0007] As a preferred embodiment of the winding deformation classification method based on the dwarf mongoose algorithm described in this invention, the extraction of multi-source numerical features includes: calculating the statistical deviation index between the frequency response curve and the reference curve based on the frequency response curve, and extracting feature parameters characterizing the changes in winding state from multiple dimensions in the time domain and frequency domain. The extracted initial features are normalized to eliminate the magnitude differences between features. The most representative feature subset is selected through feature correlation analysis, and a standardized feature vector sequence is obtained.
[0008] As a preferred embodiment of the winding deformation classification method based on the dwarf mongoose algorithm described in this invention, the adaptive global optimization includes initializing algorithm parameters and setting the search space of hyperparameters, and establishing a collaborative search strategy among individuals based on a population division of labor and cooperation mechanism. By simulating the social behavior of biological groups to balance global exploration and local development, and dynamically adjusting the search direction based on fitness assessment results, the optimal parameter combination is output and the classification model configuration is updated.
[0009] As a preferred embodiment of the winding deformation classification method based on the mongoose algorithm described in this invention, the population division of labor and cooperation mechanism includes: using the mongoose optimization algorithm and setting the population size, maximum number of iterations, leader ratio, sentinel ratio, and migration probability threshold. The position of each individual is defined as a two-dimensional vector, and the classification accuracy of the support vector machine on the validation set is used as the fitness function. In each iteration, the individual with the highest fitness is selected as the leader, and the position is updated based on the globally optimal individual and the group center position. Followers perform local searches based on the leader's position and their own experience, set scouts to randomly reset their positions, and trigger the offspring migration mechanism when the optimal fitness improvement rate is lower than the threshold for several consecutive generations. They check the individual position boundaries and truncate them, and output the optimal parameter combination. The classification model includes: using a support vector machine model, using radial basis functions as kernel functions, training the model based on optimized penalty factors and kernel parameters, solving the dual form of a convex quadratic programming problem on training samples to obtain Lagrange multipliers, and when the Lagrange multipliers are greater than zero, calculating bias terms for support vectors and constructing a decision function for classification prediction. The formula for the support vector machine model is expressed as: in, , and For Lagrange multipliers, Let be the objective function of the dual problem. The total number of training samples. For the sample Category tags, For the sample Category tags, and The feature vector of the training samples is the numerical feature extracted from the frequency response curve. For transpose operation, As a limiting condition, For sample index; The kernel function is represented as: in, For the sample and kernel function values between, For kernel parameters, For Euclidean distance.
[0010] As a preferred embodiment of the winding deformation classification method based on the dwarf mongoose algorithm described in this invention, the classification includes: inputting the extracted numerical features into the optimized support vector machine model, using the 10-fold cross-validation method, repeating it 10 times, using accuracy, precision, recall and F1 score as performance evaluation indicators, training the classification model based on the evaluation results, and classifying the fault types of axial displacement, radial concavity, radial convexity and radial misalignment. Accuracy is expressed as: Precision is expressed as: Recall rate is expressed as: The F1 value is expressed as: in, For accuracy, The true cases are the number of samples that actually belong to category A and are correctly predicted as category A by the model. False positives are the number of samples that do not actually belong to category A but are incorrectly predicted as category A by the model. False negatives are the number of samples that actually belong to category A but are incorrectly predicted by the model to belong to one of the remaining categories. A true counterexample is the number of samples that do not actually belong to category A but are correctly predicted by the model as not belonging to category A. For accuracy, For recall rate, This is the F1 value.
[0011] As a preferred embodiment of the winding deformation classification method based on the dwarf mongoose algorithm described in this invention, the comparative verification includes comparing the dwarf mongoose optimization algorithm with the particle swarm optimization algorithm and the grid search algorithm, testing with the same three sets of sample data, statistically analyzing the performance of each algorithm in three indicators: accuracy, computation time, and convergence iterations, and displaying the comparison results through visualization charts to verify the advantages of the dwarf mongoose optimization algorithm in terms of classification accuracy and convergence speed.
[0012] The beneficial effects of this preferred technical solution are that through multiple rounds of cross-validation and multi-index evaluation, the rigor and comprehensiveness of model validation are significantly improved. The use of composite indicators balances precision and recall, avoids the bias of single indicators, and makes the diagnostic results more in line with the actual needs of engineering.
[0013] As a preferred embodiment of the winding deformation classification system based on the dwarf mongoose algorithm described in this invention, it is characterized by including a data acquisition and interface module, a data processing and feature extraction module, an intelligent optimization module, a model training and diagnosis module, and a human-computer interaction and visualization module.
[0014] The data acquisition and interface module is used to establish a communication connection with the frequency response analyzer to receive and acquire the raw frequency response data of the transformer winding.
[0015] The data processing and feature extraction module is used to receive the original frequency response curve and automatically extract key numerical features characterizing the winding state through signal processing and feature engineering methods.
[0016] The intelligent optimization module is used to intelligently optimize the penalty factor and kernel function parameters of the support vector machine by simulating the division of labor and cooperation mechanism of the dwarf mongoose population, and output the optimal parameter combination with the classification accuracy as the fitness function.
[0017] The model training and diagnosis module is used to configure the support vector machine model with optimized hyperparameters, train the model using radial basis kernel functions, evaluate the model performance through 10-fold cross-validation, and automatically classify and diagnose new transformer winding states based on the trained model, identifying fault types and severity such as axial displacement, radial concavity, radial convexity, and radial misalignment.
[0018] The human-computer interaction and visualization module is used to provide users with a user-friendly graphical interface and a results display platform.
[0019] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method for classifying winding deformation based on the mongoose algorithm.
[0020] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for classifying winding deformation based on the mongoose algorithm.
[0021] The beneficial effects of this invention are as follows: By constructing a multi-source fault sample set and extracting multi-source numerical features, this invention achieves a comprehensive characterization and in-depth mining of transformer winding state information, providing high-quality, high-information-density input data for the classification model; by utilizing the mongoose optimization algorithm and population division of labor and cooperation mechanism, it achieves adaptive global optimization of support vector machine hyperparameters, balancing global exploration and local development, not only finding the optimal model configuration but also significantly shortening the optimization time, achieving a dual breakthrough in classification accuracy and efficiency; by combining multi-dimensional performance evaluation indicators to classify and verify the optimized model, the reliability and robustness of the classification results are ensured. Attached Figure Description
[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 The above is a flowchart of the winding deformation classification method based on the dwarf mongoose algorithm provided in one embodiment of the present invention.
[0024] Figure 2This is a schematic diagram illustrating four fault types of transformer windings according to a winding deformation classification method based on the dwarf mongoose algorithm provided in an embodiment of the present invention.
[0025] Figure 3 This is a schematic diagram of the search mechanism for optimizing support vector machine parameters using the dwarf mongoose algorithm in a winding deformation classification method based on the dwarf mongoose algorithm, as provided in an embodiment of the present invention.
[0026] Figure 4 The flowchart shows a system scheme for a winding deformation classification system based on the dwarf mongoose algorithm, provided in one embodiment of the present invention. Detailed Implementation
[0027] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0028] Example 1, referring to Figure 1 As an embodiment of the present invention, a winding deformation classification method based on the dwarf mongoose algorithm is provided, comprising: S100: Construct a multi-source fault sample set, collect frequency response data of transformer windings through multi-source data acquisition methods, and extract multi-source numerical features from the frequency response data.
[0029] S200: Adaptive global optimization of hyperparameters of classification models is performed using swarm intelligence optimization algorithms, and parameter search is performed based on a population division of labor and cooperation mechanism.
[0030] S300: Uses an optimized classification model to classify and identify the deformation state of transformer windings, and evaluates performance advantages through a comparative verification process.
[0031] It should be noted that this invention achieves a multi-dimensional and comprehensive characterization of winding states by constructing a multi-source fault sample set, and adopts an innovative dwarf mongoose optimization algorithm to efficiently balance global exploration and local development, avoiding the defect of being easily trapped in local optima, and significantly improving the optimization speed and accuracy; combined with multi-round cross-validation and multi-index evaluation, the optimized support vector machine model improves classification accuracy and generalization ability.
[0032] Example 2, refer to Figures 1-3 This is a second embodiment of the present invention, which provides a winding deformation classification method based on the dwarf mongoose algorithm, including: In this embodiment of the application, in step S100, the construction of the multi-source fault sample set includes steps S101 to S104: S101: The first group is a small transformer model that was independently wound. Frequency response curves were obtained by setting multiple levels of fault severity, with a total of 560 samples.
[0033] S102: The second group is simulation data, which simulates faults by changing key physical dimensions in simulation software, totaling 1020 samples.
[0034] S103: The third group contains on-site testing data for 110kV transformers, totaling 180 samples.
[0035] S104: The four types of faults include axial displacement, radial concavity, radial convexity, and radial misalignment. Each type is divided into three levels: mild, moderate, and severe, and includes healthy control samples for model supervised training and graded evaluation.
[0036] It should be noted that the axial displacement grading standards include: mild: axial displacement of the winding ≤ 1.5% of the winding height; moderate: axial displacement of the winding 1.5%-3% of the winding height; severe: axial displacement of the winding > 3% of the winding height. The radial concavity grading standards include: mild: radial concavity deformation of the winding ≤ 2% of the winding diameter; moderate: radial concavity deformation of the winding 2%-4% of the winding diameter; severe: radial concavity deformation of the winding > 4% of the winding diameter. The radial convexity grading standards include: mild: radial convexity deformation of the winding ≤ 2% of the winding diameter; moderate: radial convexity deformation of the winding 2%-4% of the winding diameter; severe: radial convexity deformation of the winding > 4% of the winding diameter. The radial misalignment grading standards include: mild: radial misalignment of the winding ≤ 1.5% of the winding diameter; moderate: radial misalignment of the winding 1.5%-3% of the winding diameter; severe: radial misalignment of the winding > 3% of the winding diameter.
[0037] In an optional implementation, in step S100, the construction of the multi-source fault sample set further includes selecting operating transformers of three voltage levels: 10kV, 35kV, and 110kV, artificially setting standard faults and collecting frequency response data during maintenance, and collecting historical fault data of the transformers to construct a fault sample library covering multiple capacities and multiple years of operation.
[0038] In another optional implementation, in step S100, the construction of the multi-source fault sample set may further include generating basic samples through electromagnetic-mechanical multiphysics coupling simulation, and performing physical verification in the laboratory using a transformer model with adjustable faults, retaining only sample data with high matching degree between simulation and actual measurement results to ensure the reliability of the data.
[0039] In this embodiment of the application, step S100, the extraction of multi-source numerical features includes steps S111~S112: S111: Nine statistical features are extracted from the frequency response curve, including Pearson correlation coefficient (CC), Euclidean distance (ED), sum of squared errors (SSE), root mean square error (RMSE), square ratio error (SSRE), standard deviation (SD), comparative standard deviation (CSD), cross correlation coefficient (CCF), and absolute logarithmic error (ALSE). in, This represents the frequency response amplitude of the transformer winding at frequency point n under reference conditions (healthy or normal). This represents the frequency response amplitude of the transformer winding under the measured state at frequency point n. This represents the total number of frequency points collected across the entire analysis band. The index is the sequence number of the frequency point. The reference curve X is the average amplitude across all frequency points. The average amplitude of the curve Y to be measured at all frequency points. For the sum of squared errors, The error is the square ratio. This is the cross-correlation coefficient. To compare standard deviations, This is the absolute sum of logarithmic errors.
[0040] S112: By calculating the differences in indicators between the fault curve and the health reference curve, an eigenvector representing the winding state is formed. .
[0041] In an optional implementation, in step S100, the extraction of multi-source numerical features further includes extracting a hybrid feature set combining frequency domain features and time domain features, including frequency domain features such as the resonant frequency offset, resonant point amplitude change, and waveform envelope area difference of the frequency response curve, as well as time domain features such as the rising edge slope and falling edge time of the curve.
[0042] In another optional implementation, in step S100, the extraction of multi-source numerical features may further include using a deep learning automatic feature extraction method, where a one-dimensional convolutional neural network performs feature learning on the original frequency response curve and automatically extracts the most discriminative feature representation without the need for manual design of statistical indicators.
[0043] In this embodiment of the application, step S200, the adaptive global optimization includes performing parameter search using the mongoose optimization algorithm, specifically including steps S201~S205: S201: Set population size M=50, maximum number of iterations 200, leader ratio 0.2, sentinel ratio 0.1, and migration probability threshold 0.05; S202: Define the position of each individual as a two-dimensional vector, and the search range of the penalty factor is [2]. -5 ,2 15 The search range for kernel function parameters is [2]. -15 ,2 3 ].
[0044] S203: The fitness function is calculated using the classification accuracy of the support vector machine on the validation set. The formula for calculating the accuracy is as follows: in, For accuracy, For a real example, As a false positive example, This is a false counterexample. This is a true counterexample.
[0045] S204: In each iteration, the individual with the highest fitness is selected as the leader. The leader's position is updated based on the globally optimal individual and the group center position. Followers perform local search based on the leader's position and their own experience. The formula for updating the leader's position is expressed as follows: in, For the leader in the next generation t+1 position, The position of the current globally optimal individual. This is the current central location of the entire population. This represents the position of individual l at iteration number t. and is a uniformly random number generated in the interval (0, 1], used to introduce random perturbation, where t is the iteration number index; The follower position update formula is expressed as: in, For follower individual u, the position in the next generation. For the position of follower individual u in the current generation, For the current leader's position, This represents the position of the worst individual in the current global situation. and This is a random perturbation factor used to control the direction of movement and step size.
[0046] S205: Set the scout's position to be randomly reset to avoid premature convergence. If the optimal fitness improvement rate is below a threshold of 10 for several consecutive generations... -6 When the time comes, the offspring migration mechanism is triggered, the individual position boundary is checked, and the truncation process is performed to output the optimal parameter combination; The formula for resetting the scout's position is expressed as: in, For the position of the scout individual v in the next generation, This serves as the lower bound of the search space. This is the upper bound of the search space. To generate uniform random numbers in the range [0, 1]; The fitness and optimal solution update formulas are expressed as follows: in, Let be the fitness value of individual 'o'. This represents the optimal fitness value in the current population. This represents the worst fitness value in the current population.
[0047] In an optional implementation, in step S200, the adaptive global optimization further includes using an improved gray wolf optimization algorithm to optimize parameters. By simulating the social hierarchy and hunting behavior of gray wolf packs, including mechanisms such as alpha wolf-led search, β and δ wolves cooperating in encirclement, and ω wolves randomly exploring, the global exploration and local exploitation capabilities are balanced.
[0048] In another optional implementation, in step S200, the adaptive global optimization may further include using a hybrid frog-jumping algorithm for parameter optimization, combining meme calculus and population splitting and merging strategies, dividing the population into multiple memeplex subgroups for parallel search, exchanging information periodically, and achieving parameter optimization through local deep search and global information sharing.
[0049] In step S300, the classification model includes steps S301 to S303: S301: The support vector machine model is adopted, and the radial basis function is used as the kernel function. The model is trained based on the optimized penalty factor and kernel parameters. The formula for the support vector machine model is expressed as: in, , and For Lagrange multipliers, Let be the objective function of the dual problem. The total number of training samples. For the sample Category tags, For the sample Category tags, and The feature vector of the training samples is the numerical feature extracted from the frequency response curve. For transpose operation, As a limiting condition, For sample index; The kernel function is represented as: in, For the sample and kernel function values between, For kernel parameters, For Euclidean distance.
[0050] S302: Solve the dual form of the convex quadratic programming problem on the training samples to obtain the Lagrange multipliers.
[0051] S303: When the Lagrange multiplier is greater than zero, the support vectors calculate the bias term and construct a decision function for classification prediction; The decision function is expressed as: in, Let be the decision function. Here is the bias term, representing the difference between the test sample x and the training sample x. The kernel function value, This is a sign function that outputs the classification result (+1 or -1).
[0052] Furthermore, in step S300, the classification and discrimination includes steps S311 to S313. S311: Input the extracted numerical features into the optimized support vector machine model and use the 10-fold cross-validation method, training with 9 subsets and testing with 1 subset each time.
[0053] S312: Repeat 10 times to ensure the comprehensiveness of the evaluation, using accuracy, precision, recall and F1 score as performance evaluation metrics. Accuracy is expressed as: Accuracy is expressed as: Precision is expressed as: Recall rate is expressed as: The F1 value is expressed as: in, The true cases are the number of samples that actually belong to category A and are correctly predicted as category A by the model. False positives are the number of samples that do not actually belong to category A but are incorrectly predicted as category A by the model. False negatives are the number of samples that actually belong to category A but are incorrectly predicted by the model to belong to one of the remaining categories. A true counterexample is the number of samples that do not actually belong to category A but are correctly predicted by the model as not belonging to category A. For accuracy, For recall rate, This is the F1 value.
[0054] S313: Train the final classification model based on the evaluation results to classify fault types such as axial displacement, radial concavity, radial convexity, and radial misalignment; Among them, axial displacement includes an overall lateral shift of the frequency response curve in the mid-to-high frequency range (100kHz–1MHz), with a significant decrease in the correlation coefficient CC and an increase in the Euclidean distance ED and root mean square error RMSE; radial concavity includes a dip in the amplitude of the curve at a specific resonant frequency point, with local maxima of SSE (sum of squared errors) and ALSE (altotal sum of logarithmic errors) at the corresponding frequency point; radial convexity includes a bulge in the amplitude of the curve at a specific resonant frequency point, which is the opposite of the frequency characteristics of radial concavity, with an abnormal SSRE (square ratio error) at the corresponding frequency point; and radial misalignment includes a composite feature of local distortion and overall shift of the curve, manifested as significant changes in multiple statistical indicators (such as CC, ED, RMSE, CSD) simultaneously.
[0055] Furthermore, in this embodiment of the application, in step S300, the comparison verification includes steps S321 to S324: S321: Compare the mongoose optimization algorithm with the particle swarm optimization algorithm and the grid search algorithm.
[0056] S322: Test using the same three sets of sample data.
[0057] S323: Statistically evaluate the performance of each algorithm in terms of accuracy, computation time, and number of convergence iterations.
[0058] S324: Visualize the comparison results to verify the advantages of the dwarf mongoose optimization algorithm in terms of classification accuracy and convergence speed.
[0059] In an optional implementation, in step S300, the comparative verification further includes using statistical hypothesis testing to perform performance verification, using t-tests to compare the performance differences of different optimization algorithms in multiple runs, and proving the effectiveness of the algorithm improvement from a statistical significance perspective.
[0060] In another optional implementation, in step S300, the comparison verification may further include a comprehensive evaluation system based on multiple evaluation indicators, adding comprehensive evaluation indicators such as model complexity, generalization ability, and stability, using the analytic hierarchy process to determine the weight of each indicator, and conducting a comprehensive algorithm comparison.
[0061] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
[0062] Example 3, referring to Figure 4 This is the third embodiment of the present invention. This embodiment provides a winding deformation classification system based on the dwarf mongoose algorithm, including a data acquisition and interface module, a data processing and feature extraction module, an intelligent optimization module, a model training and diagnosis module, and a human-computer interaction and visualization module.
[0063] The data acquisition and interface module is used to establish a communication connection with the frequency response analyzer to receive and acquire the raw frequency response data of the transformer winding.
[0064] The data processing and feature extraction module is used to receive the original frequency response curve and automatically extract key numerical features characterizing the winding state through signal processing and feature engineering methods.
[0065] The intelligent optimization module is used to intelligently optimize the penalty factor and kernel function parameters of the support vector machine by simulating the division of labor and cooperation mechanism of the dwarf mongoose population, and output the optimal parameter combination with the classification accuracy as the fitness function.
[0066] The model training and diagnosis module is used to configure the support vector machine model with optimized hyperparameters, train the model using radial basis kernel functions, evaluate the model performance through 10-fold cross-validation, and automatically classify and diagnose new transformer winding states based on the trained model, identifying fault types and severity such as axial displacement, radial concavity, radial convexity, and radial misalignment.
[0067] The human-computer interaction and visualization module is used to provide users with a user-friendly graphical interface and a results display platform.
[0068] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
[0069] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0070] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0071] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0072] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
Claims
1. A winding deformation classification method based on the dwarf mongoose algorithm, characterized in that: include, A multi-source fault sample set is constructed, frequency response data of transformer windings are collected through multi-source data acquisition methods, and multi-source numerical features are extracted from the frequency response data. A swarm intelligence optimization algorithm is used to adaptively optimize the hyperparameters of the classification model globally, and parameter search is performed based on a population division of labor and cooperation mechanism; An optimized classification model is used to classify and identify the deformation state of transformer windings, and the performance advantages are evaluated through a comparative verification process.
2. The winding deformation classification method based on the dwarf mongoose algorithm as described in claim 1, characterized in that: The construction of the multi-source fault sample set includes synchronously collecting data through multi-source data acquisition methods and setting fault severity levels for different fault types. The collected raw data were standardized and preprocessed to eliminate dimensional differences, a sample database was established, and the processed data was divided into training and test sets.
3. The winding deformation classification method based on the dwarf mongoose algorithm as described in claim 2, characterized in that: The extraction of multi-source numerical features includes calculating a statistical deviation index between the frequency response curve and the reference curve, and extracting feature parameters characterizing the changes in winding state from multiple dimensions in the time and frequency domains. The extracted initial features are normalized to eliminate the magnitude differences between features. The most representative feature subset is selected through feature correlation analysis, and a standardized feature vector sequence is obtained.
4. The winding deformation classification method based on the dwarf mongoose algorithm as described in claim 3, characterized in that: The adaptive global optimization includes initializing algorithm parameters and setting the search space of hyperparameters, and establishing a collaborative search strategy among individuals based on the population division of labor and cooperation mechanism. By simulating the social behavior of biological groups to balance global exploration and local development, and dynamically adjusting the search direction based on fitness assessment results, the optimal parameter combination is output and the classification model configuration is updated.
5. The winding deformation classification method based on the dwarf mongoose algorithm as described in claim 4, characterized in that: The population division of labor and cooperation mechanism includes adopting the dwarf mongoose optimization algorithm and setting the population size, maximum number of iterations, leader ratio, sentinel ratio, and migration probability threshold; The position of each individual is defined as a two-dimensional vector, and the classification accuracy of the support vector machine on the validation set is used as the fitness function. In each iteration, the individual with the highest fitness is selected as the leader, and the position is updated based on the globally optimal individual and the group center position. Followers perform local searches based on the leader's position and their own experience, set scouts to randomly reset their positions, and trigger the offspring migration mechanism when the optimal fitness improvement rate is lower than the threshold for several consecutive generations. They check the individual position boundaries and truncate them, and output the optimal parameter combination. The classification model includes: using a support vector machine model, using radial basis functions as kernel functions, training the model based on optimized penalty factors and kernel parameters, solving the dual form of a convex quadratic programming problem on training samples to obtain Lagrange multipliers, and when the Lagrange multipliers are greater than zero, calculating bias terms for support vectors and constructing a decision function for classification prediction. The formula for the support vector machine model is expressed as: in, , and For Lagrange multipliers, Let the objective function of the dual problem be... The total number of training samples. For the sample Category tags, For the sample Category tags, and The feature vector of the training samples is the numerical feature extracted from the frequency response curve. For transpose operation, As a limiting condition, For sample index; The kernel function is represented as: in, For the sample and kernel function values between, For kernel parameters, For Euclidean distance.
6. The winding deformation classification method based on the dwarf mongoose algorithm as described in claim 5, characterized in that: The classification process includes inputting the extracted numerical features into the optimized support vector machine model, using a 10-fold cross-validation method repeated 10 times, using accuracy, precision, recall and F1 score as performance evaluation indicators, training the classification model based on the evaluation results, and classifying fault types such as axial displacement, radial concavity, radial convexity and radial misalignment. Accuracy is expressed as: Precision is expressed as: Recall rate is expressed as: The F1 value is expressed as: in, For accuracy, The true cases are the number of samples that actually belong to category A and are correctly predicted as category A by the model. False positives are the number of samples that do not actually belong to category A but are incorrectly predicted as category A by the model. False negatives are the number of samples that actually belong to category A but are incorrectly predicted by the model to belong to one of the remaining categories. A true counterexample is the number of samples that do not actually belong to category A but are correctly predicted by the model as not belonging to category A. For accuracy, For recall rate, This is the F1 value.
7. The winding deformation classification method based on the dwarf mongoose algorithm as described in claim 6, characterized in that: The comparative verification includes comparing the mongoose optimization algorithm with the particle swarm optimization algorithm and the grid search algorithm. The same three sets of sample data are used for testing. The performance of each algorithm in terms of accuracy, computation time and convergence iterations is statistically analyzed. The comparison results are displayed through visualization charts to verify the advantages of the mongoose optimization algorithm in terms of classification accuracy and convergence speed.
8. A winding deformation classification system based on the mongoose algorithm, employing the winding deformation classification method based on the mongoose algorithm as described in any one of claims 1 to 7, characterized in that, It includes a data acquisition and interface module, a data processing and feature extraction module, an intelligent optimization module, a model training and diagnosis module, and a human-computer interaction and visualization module; The data acquisition and interface module is used to establish a communication connection with the frequency response analyzer and receive and acquire the original frequency response data of the transformer winding. The data processing and feature extraction module is used to receive the original frequency response curve and automatically extract key numerical features characterizing the winding state through signal processing and feature engineering methods. The intelligent optimization module is used to intelligently optimize the penalty factor and kernel function parameters of the support vector machine by simulating the division of labor and cooperation mechanism of the pygmy mongoose population, and output the optimal parameter combination with the classification accuracy as the fitness function. The model training and diagnosis module is used to configure the support vector machine model with optimized hyperparameters, train the model using radial basis kernel functions, evaluate the model performance through 10-fold cross-validation, and automatically classify and diagnose new transformer winding states based on the trained model, identifying the fault types and severity of axial displacement, radial concavity, radial convexity, and radial misalignment. The human-computer interaction and visualization module is used to provide users with a user-friendly graphical interface and a results display platform.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the winding deformation classification method based on the dwarf mongoose algorithm as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the winding deformation classification method based on the dwarf mongoose algorithm as described in any one of claims 1 to 7.