Power transmission line fault identification method and system based on improved PSO-SVM incremental learning strategy

Abstract

The invention discloses a power transmission line fault identification method and system based on an improved PSO-SVM incremental learning strategy. The method comprises the following steps: training a PSO-SVM model by using historical fault data of a power transmission line to determine a support vector; calculating the importance degree of the support vector and the sparsity of the K-neighbor center to screen out playback data; merging the newly added fault data and the playback data into an incremental data set; new samples are generated in a sparse sample distribution area in the incremental data set through a zoning oversampling technology to achieve intra-class data balance of the incremental data set, then the incremental data set is used for training the PSO-SVM model again through multi-task cross integration learning to achieve inter-class data balance of the incremental data set, and a final fault recognition model is obtained. According to the method, the self-adaption and high-precision identification of the fault is realized under limited computing resources.

Description

Technical Field

[0001] This invention relates to transmission line fault diagnosis technology, specifically to a method and system for identifying transmission line faults based on an improved PSO-SVM incremental learning strategy. Background Technology

[0002] As the backbone of the power grid, transmission lines are characterized by high voltage levels, wide coverage, and complex operating environments. They are constantly subjected to severe challenges from extreme weather (such as lightning strikes, ice damage, and wind damage) and external interference (such as external force damage and bird damage), making them highly susceptible to various types of faults. Due to the diverse manifestations and intertwined electrical characteristics of different faults, accurate fault identification is difficult, significantly increasing the difficulty of fault location and elimination, and posing a serious threat to the safe and stable operation of the power grid. With the continuous advancement of smart grid construction, the power grid scale is constantly expanding and the network structure is becoming increasingly complex. The industry has placed higher demands on the real-time performance, accuracy, and adaptability of transmission line fault identification. Traditional fault diagnosis methods based on manual analysis or simple algorithms have shown significant limitations when dealing with multi-source heterogeneous fault data and complex fault scenarios, making it difficult to meet practical application needs. To overcome the bottlenecks of traditional methods, transmission line fault identification schemes based on artificial intelligence technology have become a research hotspot. Existing technologies are mainly classified into two categories: traditional machine learning methods and deep neural network methods. However, both types of methods have significant drawbacks and are difficult to adapt to the fault identification needs of real-world scenarios.

[0003] On the one hand, traditional machine learning methods typically rely on manually designed feature descriptions. After selecting key features from waveform data such as fault voltage and current, they combine classic classification algorithms such as Support Vector Machines (SVM) and Random Forests to achieve fault identification. The advantage of this type of method is that the feature extraction process is controllable, the physical meaning of the extracted features is clear, and the model has good interpretability. However, the model parameters usually need to be manually tuned, and the parameter settings are greatly affected by subjective factors, resulting in poor model adaptability. Furthermore, repeated training is required when facing dynamically added fault data, limiting the scope of application and making it difficult to promote and apply.

[0004] On the other hand, deep neural network-based methods can automatically learn features and classify faults directly from raw waveform data, achieving end-to-end fault identification. These methods are highly adaptive and require minimal human intervention, but they are essentially data-driven strategies, heavily reliant on large amounts of high-quality historical fault data. However, in real-world scenarios, obtaining transmission line fault data is difficult, the total amount is often very limited, fault category labeling is challenging, and data imbalance is a significant problem. This makes deep neural network models prone to overfitting, resulting in insufficient generalization ability when faced with new fault data, and hindering effective stable fault identification in real-world scenarios.

[0005] In summary, existing intelligent algorithm-based transmission line fault identification methods are either limited by cumbersome parameter tuning, poor adaptability and difficulty in handling dynamic data, or limited by dependence on high-quality labeled data and insufficient generalization ability in small sample scenarios. They cannot effectively solve the core problems of diverse fault types, limited available data and random addition of dynamic data in real-world applications. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a method and system for identifying transmission line faults based on an improved PSO-SVM incremental learning strategy, which achieves adaptive and high-precision fault identification under limited computing resources.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for transmission line fault identification based on an improved PSO-SVM incremental learning strategy includes the following steps: Acquire historical fault data of transmission lines and extract fault characteristics of transmission lines; Historical fault data is divided into positive and negative samples. The PSO-SVM model is trained using the transmission line fault features of historical fault data to obtain the initial fault identification model and determine the support vectors in the positive and negative samples. The importance of the support vectors is calculated to obtain the corresponding importance evaluation results. The K-nearest neighbor sparsity of the support vectors is calculated. The replay data evaluation function of the support vectors is calculated based on the importance evaluation results and K-nearest neighbor sparsity. The support vectors of all categories are cross-sorted based on the value of the replay data evaluation function, and the replay data is selected. Collect new fault data of transmission lines and merge the new fault data with the playback data into an incremental dataset; By using the oversampling technique, new samples are generated in sparse regions of the incremental dataset to achieve intra-class data balance. Then, the PSO-SVM model is retrained using the incremental dataset through multi-task cross-ensemble learning to achieve inter-class data balance and obtain the final fault identification model. The fault characteristics of the transmission line from the fault test data are input into the final fault identification model to obtain the fault identification results of the transmission line.

[0008] Furthermore, the fault characteristics of the transmission line include the peak value of the zero-sequence current, the standard deviation of the zero-sequence current, the skewness of the zero-sequence current, the peak value of the rate of change of current, the dominant frequency of the zero-sequence current, and the centroid of the spectrum.

[0009] Furthermore, the calculation of support vector importance includes the following steps: Calculate the cluster centers of positive and negative samples respectively; Calculate the support vectors of positive or negative samples. Distance to the corresponding cluster center And calculate support vectors Distance from the classification hyperplane of the initial fault identification model ; Distance With distance Substituting the values ​​into the importance evaluation function yields the support vectors. The importance evaluation results are as follows, and the mathematical expression of the importance evaluation function is as follows:

[0010] in, It is the minimum distance among all support vectors to their corresponding cluster centers. It is the maximum value among all the distances from the support vectors to their corresponding cluster centers. It is the minimum distance among all support vectors to the classification hyperplane. It is the maximum value among all the distances from the support vectors to the classification hyperplane.

[0011] Furthermore, the mathematical expression for calculating the K-nearest neighbor sparsity of support vectors is as follows:

[0012] in, Indicates the first The class of Support vectors, For support vectors The K-nearest neighbor center density is mathematically expressed as follows:

[0013] in, Represents support vectors The Let K be the nearest neighbor samples, and K represent the support vectors. The number of nearest neighbor samples, The average distance between the K-nearest neighbors of all support vectors is expressed mathematically as follows:

[0014] Where h represents the total number of support vectors of the same class. Represents support vectors With support vectors The squared distance.

[0015] Furthermore, when calculating the replay data evaluation function for support vectors based on the importance evaluation results and the sparsity of K-nearest neighbor centers, the following steps are included: Construct a decision matrix A for the evaluation indicators, where the elements are... Indicates the first The class of The values ​​of each evaluation index are calculated based on the decision matrix A, and then the weights are normalized. Based on the normalized weights, the importance evaluation results of the support vectors are weighted and summed with the K-nearest neighbor sparsity to obtain the replay data evaluation function for the support vectors. The mathematical expression is as follows:

[0016] in, This represents the i-th support vector. Represents support vectors Importance evaluation results Represents support vectors K-nearest neighbor center sparsity, and These are the weights of the first and second evaluation indicators after normalization.

[0017] Furthermore, the mathematical expression for the weights of the evaluation indicators is as follows:

[0018] in, It is the first The final weight of the evaluation index. It is an adjustment factor. It is the coefficient of variation. It is the first The mathematical expression for the sum of the Pearson correlation coefficients of the class evaluation index and all other indicators, and the coefficient of variation, is as follows:

[0019] in, This indicates that the decision matrix A has been standardized.

[0020] Furthermore, before generating new samples in sparsely distributed regions of the incremental dataset using the band oversampling technique, the following steps are also included: Calculate the distance between newly added fault data and the corresponding cluster center. Simultaneously, calculate the maximum distance among all replay data to their corresponding cluster centers. If the distance corresponding to the newly added fault data Less than or equal to the maximum distance The newly added fault data will be used as the new replay data. If a corresponding distance exists... Greater than the maximum distance The newly added fault data is used to generate new samples in sparsely distributed regions of the incremental dataset through the segmented oversampling technique.

[0021] Furthermore, when generating new samples in sparsely distributed regions of the incremental dataset using the band oversampling technique, the following steps are included: Sort the samples in the incremental dataset in ascending order according to their intra-class distance in the sample space, and then select samples from the sequence in turn. Each sample is placed into the corresponding set for each band. ,in This represents the number of zones in the sample space. To determine the number of replay samples, For all types of people in the first The sample set in each zone; The formula for calculating the total number of new samples to be generated across all zones is as follows:

[0022] in, The number of newly added samples that fall within a reasonable distribution range. This represents the number of newly added samples; The formula for calculating the number of new samples needed to be generated within each band is as follows:

[0023] Starting from the cluster center and ending at each sample, we treat the samples as vectors; for a sample set... For a single sample, calculate the minimum angle between the corresponding vectors. Finally, the sample pair used to generate new samples is determined by selecting the largest included angle from the calculation results. This process generates new samples and iterates through the steps until the required number of new samples are generated within the corresponding band. The formula for calculating the number of new samples is as follows:

[0024] in, Indicates the cluster center.

[0025] Furthermore, when retraining the PSO-SVM model using an incremental dataset through multi-task cross-ensemble learning, the following steps are included: The incremental dataset is divided into category subsets that correspond one-to-one with each fault category; Bootstrap sampling is used for each category subset. Next, the corresponding sampling subset is obtained, where It is the result of dividing the total number of samples in the incremental dataset by the number of categories; Combine all sampled subsets into the current training subset, train the current PSO-SVM model using the transmission line fault features of the current training subset, and then perform Bootstrap sampling again for each category subset. This process is repeated until a specified number M is reached, resulting in a final fault identification model containing M base classifiers. The final fault identification model uses all base classifiers to vote on the samples, counts the class frequency of each sample in the M votes, and takes the class with the highest number of votes as the final result. If multiple classes have the same number of votes, one of the classes with the same number of votes is randomly selected.

[0026] The present invention also proposes a transmission line fault identification system based on an improved PSO-SVM incremental learning strategy, comprising a microprocessor and a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, which is executed by the processor to implement the transmission line fault identification method based on the improved PSO-SVM incremental learning strategy.

[0027] Compared with the prior art, the advantages of the present invention are as follows: This invention effectively reduces the model's dependence on complete historical data by constructing an adaptively optimized PSO-SVM base classifier and combining it with an incremental learning strategy that integrates support vector importance and K-nearest neighbor sparsity. This reduces the subjectivity and computational overhead caused by manual parameter tuning and repeated training. By introducing a dual data balancing mechanism combining intra-class and inter-class data, the model's adaptability and robustness in imbalanced data scenarios are significantly improved. This enables more comprehensive and accurate processing of dynamically added small-sample fault data, enhancing the scientific rigor and accuracy of the fault identification model and providing reliable technical support for fault diagnosis of smart grid transmission lines. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0029] Figure 2 This is a spatial distribution map of the sample data for transmission line faults.

[0030] Figure 3 This is a diagram illustrating the cross-sorting of support vectors.

[0031] Figure 4 This is a schematic diagram of the sample space zoning.

[0032] Figure 5 This is a schematic diagram of the process of generating new samples within the band.

[0033] Figure 6 Flowchart for a multi-task cross-integrated learning method.

[0034] Figure 7 The following are zero-sequence current waveforms for transmission line fault data: (a) is the zero-sequence current waveform for ice damage fault, (b) is the zero-sequence current waveform for external force fault, (c) is the zero-sequence current waveform for lightning strike fault, (d) is the zero-sequence current waveform for wind damage fault, and (e) is the zero-sequence current waveform for bird damage fault. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0036] To address the challenges of diverse fault types, limited available data, random dynamic data additions, cumbersome parameter tuning of existing models, easy forgetting of historical knowledge, and insufficient recognition performance due to data imbalance in transmission line fault identification, this embodiment proposes a transmission line fault identification method based on an improved PSO-SVM incremental learning strategy. This method constructs a comprehensive evaluation function by fusing support vector importance and K-nearest neighbor sparsity, and introduces a dual data balancing mechanism of intra-class segmented oversampling and inter-class multi-task integration. This enables adaptive filtering of historical playback data and effective optimization of imbalanced data, improving the adaptability, stability, and recognition accuracy of the fault identification model in dynamic data environments.

[0037] like Figure 1 As shown, the method in this embodiment includes the following steps: S1) Obtain historical fault data of transmission lines and extract fault features of transmission lines. Divide the historical fault data into positive and negative samples. Use the fault features of transmission lines in the historical fault data to train the PSO-SVM model, obtain the initial fault identification model, and determine the support vectors in the positive and negative samples. S2) Calculate the importance of the support vectors to obtain the corresponding importance evaluation results, calculate the K-nearest neighbor sparsity of the support vectors, calculate the replay data evaluation function of the support vectors based on the importance evaluation results and K-nearest neighbor sparsity, and cross-sort the support vectors of all categories based on the value of the replay data evaluation function to select the replay data. S3) Collect new fault data of transmission lines and merge the new fault data with the playback data into an incremental dataset; S4) By using the oversampling technique, new samples are generated in sparsely distributed regions of the incremental dataset to achieve intra-class data balance in the incremental dataset. Then, the PSO-SVM model is retrained using the incremental dataset through multi-task cross-ensemble learning to achieve inter-class data balance in the incremental dataset and obtain the final fault identification model. S5) Input the fault characteristics of the transmission line from the fault test data into the final fault identification model to obtain the fault identification result of the transmission line.

[0038] The following is a detailed explanation of each step.

[0039] In step S1 of this embodiment, the PSO-SVM model adaptively optimizes the penalty parameter C and Gaussian kernel parameter of the support vector machine (SVM) using the particle swarm optimization (PSO) algorithm. This avoids repetitive training caused by manual tuning. When dividing historical fault data into positive and negative classes, the specific process involves determining whether the fault type in the historical fault data matches the type of fault to be identified, thus labeling the historical fault data as positive or negative. The required transmission line fault features for training the PSO-SVM model include the peak value of the zero-sequence current, the standard deviation of the zero-sequence current, the skewness of the zero-sequence current, the peak value of the current change rate, the dominant frequency of the zero-sequence current, and the spectral centroid. The relevant calculations are as follows: The zero-sequence current peak value reflects the amplitude intensity at the instant the fault occurs, and its calculation formula is as follows: (1) in, Let t be the instantaneous value of the zero-sequence current at any time t.

[0040] The zero-sequence current standard deviation measures the dispersion of a signal over time, and its calculation formula is as follows:

[0041] (2) Where N is the total number of sampling points, This represents the zero-sequence current value at the k-th sampling point. This is the average value of the zero-sequence current at all sampling points.

[0042] The zero-sequence current skewness reveals the asymmetry of the signal distribution, and its calculation formula is as follows: (3) The peak current change rate focuses on the abrupt change behavior in the early stages of a fault, and its calculation formula is as follows: (4) The dominant frequency of the zero-sequence current reveals the concentrated region of fault signal energy, and its calculation formula is as follows: (5) in, The zero-sequence current amplitude in the frequency domain can be calculated using Fourier transform.

[0043] The spectral centroid measures the "center of gravity" of the frequency distribution, and its calculation formula is as follows: (6) in, For frequency variables.

[0044] After training the PSO-SVM model in step S1, the sample space of historical fault data for transmission lines is as follows: Figure 2 As shown in the figure, H1 represents the positive class supporting hyperplane, and H2 represents the negative class supporting hyperplane. The sample points that fall exactly on H1 or H2 are the support vectors.

[0045] Step S2 of this embodiment uses a comprehensive evaluation function constructed from support vector importance and K-nearest neighbor sparsity to cross-sort the support vectors of historical data and select the replay dataset, including the following steps: S21) Calculate the support vector importance of the support vectors, including: Calculate the cluster centers of the positive and negative classes separately, as follows: Figure 2 As shown, the K-means clustering algorithm is used to calculate the cluster centers of all positive class samples. Cluster centers of all negative samples ; Calculate the support vectors of positive or negative samples. Distance to the corresponding cluster center And calculate support vectors Distance from the classification hyperplane of the initial fault identification model ; The first of the positive classes support vectors For example, calculate the first positive class. The support vectors are Its arrival at the center of the positive class sample The distance between the sample and the classification hyperplane, and the distance to the center of the positive class sample. The distance calculation formula is: (7) The formula for calculating the distance to the classification hyperplane is: (8) in, The parameters of the classification hyperplane.

[0046] Distance With distance Substituting the values ​​into the importance evaluation function yields the support vectors. The importance evaluation results are as follows, and the mathematical expression of the importance evaluation function is as follows: (9) in, It is the minimum distance among all support vectors to their corresponding cluster centers. It is the maximum value among all the distances from the support vectors to their corresponding cluster centers. It is the minimum distance among all support vectors to the classification hyperplane. It is the maximum value among all the distances from the support vectors to the classification hyperplane.

[0047] S22) Calculate the K-nearest neighbor sparsity of the support vectors. Assume the support vectors are ,in . Indicates the first The class of If there are 1 support vector, then the support vectors The formula for calculating the sparsity of K-nearest neighbor centers is: (10) in, Indicates the first The class of Support vectors, For support vectors The K-nearest neighbor center density is mathematically expressed as follows: (11) in, Represents support vectors The Let K be the nearest neighbor samples, and K represent the support vectors. The number of nearest neighbor samples, The average distance between the K-nearest neighbors of all support vectors is expressed mathematically as follows: (12) Where h represents the total number of support vectors of the same class. Represents support vectors With support vectors The squared distance for two Dimensional Samples , The formula for calculating the squared distance is: (13) S23) Calculate the replay data evaluation function for support vectors based on the importance evaluation results and the sparsity of K-nearest neighbor centers, including the following steps: Construct a decision matrix A for the evaluation indicators, assuming a total of [number] indicators are selected. Category evaluation indicators, each category of evaluation indicators includes The specific, quantifiable evaluation indicators, and the formula for calculating decision matrix A are as follows: (14) Among them, elements Indicates the first The class of The values ​​of each evaluation indicator.

[0048] After calculating the weights of the evaluation indicators based on the decision matrix A, the weights are normalized. First, the matrix is ​​standardized, and the calculation formula is as follows: (15) The coefficient of variation is introduced to measure the volatility of the evaluation index, and the calculation formula is as follows: (16) in, It is the first Standard deviation of the evaluation index It is the first The mean of the evaluation indicators.

[0049] The Pearson correlation coefficient is introduced to eliminate redundancy between indicators. The calculation formula is as follows: (17) in, For the first Category evaluation indicators and the first Correlation coefficients of various evaluation indicators.

[0050] Then, calculate the final weight of each indicator using the following formula: (18) in, It is the first The final weight of the evaluation index. It is an adjustment factor, with a value of . It is the first The sum of the correlations between this evaluation index and all other indicators.

[0051] The weights are normalized, and the calculation formula is as follows: (19) Finally, based on the normalized weights, the importance evaluation results of the support vectors are weighted and summed with the sparsity of the K-nearest neighbor centers to obtain the replay data evaluation function for the support vectors, the mathematical expression of which is as follows: (20) in, This represents the i-th support vector. Represents support vectors Importance evaluation results Represents support vectors K-nearest neighbor center sparsity, and These are the weights of the first and second evaluation indicators after normalization.

[0052] S24) Cross-sorting of support vectors like Figure 3 As shown, after calculating the replay data evaluation function of the support vectors in the positive and negative samples through the above steps, the support vectors of all categories are sorted alternately in descending order of the value of the replay data evaluation function. Then, the support vectors whose replay data evaluation function value is greater than the preset threshold or the specified number of support vectors at the head of the queue are selected as replay data.

[0053] Step S4 in this embodiment balances the intra-class data of the newly added dataset using segmented oversampling technology and balances the inter-class data of the newly added dataset using multi-task cross-ensemble learning to obtain the final PSO-SVM model, including the following steps: S41) Generation of new samples within the band Before generating new samples in sparsely distributed regions of the incremental dataset using a segmented oversampling technique, the distance between the newly added faulty data and its corresponding cluster center is calculated. Simultaneously, calculate the maximum distance among all replay data to their corresponding cluster centers. ,like Figure 4 As shown, if the distance corresponding to the newly added fault data... Less than or equal to the maximum distance This indicates that there are no sparsely distributed regions in the incremental dataset. Therefore, the newly added fault data is directly used as the new replay data. If a corresponding distance exists... Greater than the maximum distance The addition of fault data indicates that there are sparsely distributed regions in the incremental dataset, and new samples need to be generated in these sparsely distributed regions using the segmented oversampling technique.

[0054] In this embodiment, when generating new samples in sparsely distributed regions of the incremental dataset using the band oversampling technique, the following steps are included: First, determine the samples in each band and calculate the band interval. The formula for calculating the band interval is: (twenty one) in, For all types of people in the first The sample set in the band It is in the first Samples in each band.

[0055] Assuming the sample space is divided into Each band, the number of replay samples is The samples in the incremental dataset are sorted in ascending order according to their intra-class distance in the sample space. Then, samples are selected sequentially from the sequence. Each sample is placed into the corresponding set for each band. Number of samples in the outermost band ,in This represents the number of zones in the sample space. To determine the number of replay samples, For all types of people in the first The sample set in each zone; Then calculate the required number of new samples, and calculate the total number of new samples to be generated across all zones. The calculation formula is as follows: (twenty two) in, The number of newly added samples that fall within a reasonable distribution range. This represents the number of newly added samples; The formula for calculating the number of new samples needed to be generated within each band is as follows: (twenty three) Finally, based on the required number of new samples, new samples are generated using the existing samples in each zone. Specifically, the samples are treated as vectors, starting from the cluster centers and ending with each sample. For the sample set... For a single sample, calculate the minimum angle between the corresponding vectors. Finally, the maximum included angle from the calculated results is selected to satisfy the condition. The included angle is used to determine the sample pairs used to generate new samples. This process generates new samples and iterates through the steps until the required number of new samples are generated within the corresponding band. The formula for calculating the number of new samples is as follows: (twenty four) in, Indicates the cluster center.

[0056] like Figure 5 As shown, there are samples S1, S2 and S3 in the same segment. Calculate the minimum angle between the vectors of each pair to get three angles. Obviously, the minimum angle between the vectors of samples S1 and S2 is the largest angle among the three angles. Therefore, substitute samples S1 and S2 into formula (24) to calculate the new sample sn1.

[0057] S42) Multi-task cross-integrated learning like Figure 6As shown, when retraining the PSO-SVM model using an incremental dataset through multi-task cross-ensemble learning, the following steps are included: S421) Divide the incremental dataset into category subsets that correspond one-to-one with each fault category; S422) Bootstrap sampling is used for each category subset. Next, the corresponding sampling subset is obtained, where It is the result of dividing the total number of samples in the incremental dataset by the number of categories; S423) Combine all sampled subsets into the current training subset, train the current PSO-SVM model using the transmission line fault features of the current training subset, and then execute step S422 again until the specified number of times M is reached, thereby obtaining the final fault identification model containing M base classifiers.

[0058] Let the total number of samples in the incremental dataset used as the initial training set be... Among them, categories The sample sizes are respectively Since the training subset is required to have a perfect balance of the number of samples in each class, the number of samples in each class can be obtained as follows: For each category subset Using Bootstrap sampling Next, obtain an equal number of sampling subsets. Then, the sampled subsets of all categories are merged and assembled into a training subset with a perfectly balanced category distribution and a size exactly the same as the original training set. .

[0059] Use training subset After training the first PSO-SVM base classifier, repeat the above "class balanced sampling - merging - training" process. The wheel can produce A base classifier that exhibits significant differences, yet treats all subclasses equally.

[0060] In this embodiment, the final fault identification model uses all base classifiers to vote on the samples, counts the class frequency of each sample in the M votes, and takes the class with the highest number of votes as the final result according to the principle of majority rule. If multiple classes have the same number of votes, one of the classes with the same number of votes is randomly selected to ensure that the decision chain can still close quickly when there is a complete tie, thus maintaining the robustness and simplicity of the integrated system.

[0061] In step S5 of this embodiment, the fault test data includes the use of... Figure 7The newly generated fault test data and historical fault test data are shown in the zero-sequence current waveform diagram of the transmission line fault data. The test results show that the method of this embodiment achieves an accuracy of 90.48% on the real transmission line fault dataset.

[0062] Furthermore, this embodiment also proposes a transmission line fault identification system based on an improved PSO-SVM incremental learning strategy, including a microprocessor and a computer-readable storage medium. The computer-readable storage medium stores a computer program, which is executed by the processor to implement the transmission line fault identification method based on the improved PSO-SVM incremental learning strategy described in this embodiment.

[0063] In summary, this invention proposes a transmission line fault identification method and system based on an improved PSO-SVM incremental learning strategy. By constructing a PSO-SVM base classifier, the penalty coefficient and Gaussian kernel parameters of the Support Vector Machine (SVM) are adaptively optimized using the Particle Swarm Optimization (PSO) algorithm, effectively avoiding the repetitive training problem caused by manual parameter tuning in traditional methods, and improving the training efficiency and generalization ability of the model. Addressing the characteristic of continuously and dynamically added transmission line fault data, an incremental learning strategy integrating support vector importance and K-nearest neighbor sparsity is proposed. A cross-ranking replay data filtering mechanism is constructed to achieve intelligent selection and retention of historical key samples, thereby mitigating the catastrophic forgetting problem in the incremental learning process. To further improve the model's recognition performance in imbalanced data scenarios, a dual data balancing method combining intra-class and inter-class approaches is proposed. At the intra-class level, a sample space-based oversampling technique is employed. Based on the distance distribution characteristics between samples and cluster centers, new samples are generated in sparse regions to optimize the distribution uniformity of data for the same type of fault. At the inter-class level, a multi-task cross-integration learning strategy is introduced. A class-balanced training subset is constructed through hierarchical bootstrap sampling to train multiple PSO-SVM base classifiers. A majority voting mechanism is then used to integrate the prediction results, significantly improving the ability to identify minority class faults. This invention achieves adaptive and high-precision identification of multiple fault types under limited computational resources, effectively improving the model's robustness and practicality in dynamic data environments, and providing a feasible technical path for fault diagnosis of smart grid transmission lines.

[0064] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0065] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for transmission line fault identification based on an improved PSO-SVM incremental learning strategy, characterized in that, Includes the following steps: Acquire historical fault data of transmission lines and extract fault characteristics of transmission lines; Historical fault data is divided into positive and negative samples. The PSO-SVM model is trained using the transmission line fault features of historical fault data to obtain the initial fault identification model and determine the support vectors in the positive and negative samples. The importance of the support vectors is calculated to obtain the corresponding importance evaluation results. The K-nearest neighbor sparsity of the support vectors is calculated. The replay data evaluation function of the support vectors is calculated based on the importance evaluation results and K-nearest neighbor sparsity. The support vectors of all categories are cross-sorted based on the value of the replay data evaluation function, and the replay data is selected. Collect new fault data of transmission lines and merge the new fault data with the playback data into an incremental dataset; By using the oversampling technique, new samples are generated in sparse regions of the incremental dataset to achieve intra-class data balance. Then, the PSO-SVM model is retrained using the incremental dataset through multi-task cross-ensemble learning to achieve inter-class data balance and obtain the final fault identification model. The fault characteristics of the transmission line from the fault test data are input into the final fault identification model to obtain the fault identification results of the transmission line.

2. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, The fault characteristics of the transmission line include the peak value of the zero-sequence current, the standard deviation of the zero-sequence current, the skewness of the zero-sequence current, the peak value of the rate of change of current, the dominant frequency of the zero-sequence current, and the centroid of the spectrum.

3. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, The steps involved in calculating the importance of support vectors are as follows: Calculate the cluster centers of positive and negative samples respectively; Calculate the support vectors of positive or negative samples. Distance to the corresponding cluster center And calculate support vectors Distance from the classification hyperplane of the initial fault identification model ; Distance With distance Substituting the values ​​into the importance evaluation function yields the support vectors. The importance evaluation results are as follows, and the mathematical expression of the importance evaluation function is as follows: in, It is the minimum distance among all support vectors to their corresponding cluster centers. It is the maximum value among all the distances from the support vectors to their corresponding cluster centers. It is the minimum distance among all support vectors to the classification hyperplane. It is the maximum value among all the distances from the support vectors to the classification hyperplane.

4. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, The mathematical expression for calculating the K-nearest neighbor sparsity of support vectors is as follows: in, Indicates the first The class of Support vectors, For support vectors The K-nearest neighbor center density is mathematically expressed as follows: in, Represents support vectors The Let K be the nearest neighbor samples, and K represent the support vectors. The number of nearest neighbor samples, The average distance between the K-nearest neighbors of all support vectors is expressed mathematically as follows: Where h represents the total number of support vectors of the same class. Represents support vectors With support vectors The squared distance.

5. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, When calculating the replay data evaluation function for support vectors based on their importance evaluation results and K-nearest neighbor sparsity, the following steps are included: Construct a decision matrix A for the evaluation indicators, where the elements are... Indicates the first The class of The values ​​of each evaluation index are calculated based on the decision matrix A, and then the weights are normalized. Based on the normalized weights, the importance evaluation results of the support vectors are weighted and summed with the K-nearest neighbor sparsity to obtain the replay data evaluation function for the support vectors. The mathematical expression is as follows: in, This represents the i-th support vector. Represents support vectors Importance evaluation results Represents support vectors K-nearest neighbor center sparsity, and These are the weights of the first and second evaluation indicators after normalization.

6. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 5, characterized in that, The mathematical expressions for the weights of the evaluation indicators are as follows: in, It is the first The final weight of the evaluation index. It is an adjustment factor. It is the coefficient of variation. It is the first The mathematical expression for the sum of the Pearson correlation coefficients of the class evaluation index and all other indicators, and the coefficient of variation, is as follows: in, This indicates that the decision matrix A has been standardized.

7. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, Before generating new samples in sparsely distributed regions of the incremental dataset using the band oversampling technique, the following steps are also included: Calculate the distance between newly added fault data and the corresponding cluster center. Simultaneously, calculate the maximum distance among all replay data to their corresponding cluster centers. If the distance corresponding to the newly added fault data Less than or equal to the maximum distance The newly added fault data will be used as the new replay data. If a corresponding distance exists... Greater than the maximum distance The newly added fault data is used to generate new samples in sparsely distributed regions of the incremental dataset through the segmented oversampling technique.

8. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, When generating new samples in sparsely distributed regions of an incremental dataset using the band oversampling technique, the following steps are included: Sort the samples in the incremental dataset in ascending order according to their intra-class distance in the sample space, and then select samples from the sequence in turn. Each sample is placed into the corresponding set for each band. ,in This represents the number of zones in the sample space. To determine the number of replay samples, For all types of people in the first The sample set in each zone; The formula for calculating the total number of new samples to be generated across all zones is as follows: in, The number of newly added samples that fall within a reasonable distribution range. This represents the number of newly added samples; The formula for calculating the number of new samples needed to be generated within each band is as follows: Starting from the cluster center and ending at each sample, we treat the samples as vectors; for a sample set... For a single sample, calculate the minimum angle between the corresponding vectors. Finally, the sample pair used to generate new samples is determined by selecting the largest included angle from the calculation results. This step generates new samples and is repeated iteratively until the required number of new samples within the corresponding band are generated. The formula for calculating the number of new samples is: in, Indicates the cluster center.

9. The transmission line fault identification method based on the improved PSO-SVM incremental learning strategy according to claim 1, characterized in that, When retraining a PSO-SVM model using an incremental dataset through multi-task cross-ensemble learning, the following steps are included: The incremental dataset is divided into category subsets that correspond one-to-one with each fault category; Bootstrap sampling is used for each category subset. Next, the corresponding sampling subset is obtained, where It is the result of dividing the total number of samples in the incremental dataset by the number of categories; Combine all sampled subsets into the current training subset, train the current PSO-SVM model using the transmission line fault features of the current training subset, and then perform Bootstrap sampling again for each category subset. This process is repeated until a specified number M is reached, resulting in a final fault identification model containing M base classifiers. The final fault identification model uses all base classifiers to vote on the samples, counts the class frequency of each sample in the M votes, and takes the class with the highest number of votes as the final result. If multiple classes have the same number of votes, one of the classes with the same number of votes is randomly selected.

10. A transmission line fault identification system based on an improved PSO-SVM incremental learning strategy, characterized in that, The method includes a microprocessor and a computer-readable storage medium storing a computer program, which is executed by the processor to implement the transmission line fault identification method based on the improved PSO-SVM incremental learning strategy as described in any one of claims 1 to 9.

Images