A distribution network grounding fault cause classification method, system and medium based on HOA-CNN

Abstract

The present application relates to a method, system and medium for classifying the causes of grounding faults in distribution networks based on HOA-CNN. The method includes obtaining zero-sequence current waveform data under various working conditions of grounding faults in distribution networks caused by different reasons; using a sliding time window method to perform slicing operations and process the zero-sequence current waveform data; one-hot encoding the data labels of each working condition; training sets and test sets from the shuffled data; establishing a CNN model; reasonably setting the parameters of the CNN model through the HOA optimization algorithm; inputting the training set data into the HOA-CNN model, and continuing the training until the preset number of iterations is reached to complete the HOA-CNN model training; inputting the test set into the trained HOA-CNN model for testing to detect the classification ability of the HOA-CNN model. The present application optimizes the parameters of the convolutional neural network CNN by introducing HOA, and classifies the causes of grounding faults in distribution networks more efficiently and accurately, contributing to the intelligent development of power systems.

Description

Technical Field

[0001] The present application relates to the field of distribution network grounding fault classification, and more specifically, to a distribution network grounding fault cause classification method, system and medium based on HOA-CNN. Background Art

[0002] Each ground fault has its own unique physical mechanism and manifestation, but in actual situations, these fault characteristics are often masked or confused by the complex field environment. The fault signal may contain a lot of noise and interference, making it difficult to extract and identify useful fault features.

[0003] Ground faults include many factors, from natural environmental factors (such as bird nests on wires, birds hitting wires), to human factors (such as cars hitting wires, foreign objects hitting wires), to equipment problems (such as poor transformer insulation, lightning arrester failure), etc. This diversity leads to blurred boundaries between faults, increasing the difficulty of classification.

[0004] In actual operation, some types of ground faults may be more common, while other types of faults are relatively rare. This problem of unbalanced data volume may cause the machine learning algorithm to be biased towards the majority class during training, thereby affecting the ability to identify minority class faults.

[0005] Ground faults need to be identified and handled promptly and accurately to avoid greater impact on power systems and users. However, in practical applications, the real-time and accuracy of algorithms are often limited by multiple factors such as computing resources and data quality.

[0006] Due to the diversity and complexity of ground faults, it is very difficult to design a general algorithm that can be applied to all situations. The algorithm needs to maintain stable performance in different environments, different equipment, and different fault conditions.

[0007] In the field of classification and diagnosis of grounding faults in distribution networks, existing technologies mainly rely on traditional signal processing methods and machine learning algorithms, such as support vector machines (SVM), K-nearest neighbors (KNN) and decision trees (DT). Although these methods can achieve fault classification to a certain extent, they still have the following significant disadvantages: Limited feature extraction capability: Traditional methods often rely on manually designed features, which may not fully reflect the fault characteristics in the complex and changeable distribution network grounding fault environment, resulting in limited classification accuracy. Insufficient ability to process high-dimensional features: With the expansion of the scale of distribution networks and the improvement of their intelligence, fault data is becoming increasingly complex. Traditional methods are unable to cope with high-dimensional features and are difficult to capture subtle differences in fault characteristics. Poor real-time performance: When processing large-scale distribution network data, the computational complexity of traditional methods is high and it is difficult to meet real-time requirements, which is particularly unfavorable for grounding fault classification scenarios that require rapid response. Poor adaptability: The operating environment of the distribution network is complex and changeable, and there are many types of faults. When facing new faults or complex fault scenarios, traditional methods often need to redesign features or adjust parameters, and lack good adaptive capabilities. Difficulty in identifying high-resistance grounding faults: High-resistance grounding faults are difficult to identify effectively using traditional methods due to their large transition resistance and weak fault characteristics, making the handling of such faults a technical challenge. Summary of the invention

[0008] The purpose of the embodiments of the present application is to provide a method, system and medium for classifying the causes of grounding faults in distribution networks based on HOA-CNN, combining intelligent optimization algorithms with deep learning technology, providing new ideas and technical means for the intelligent operation and maintenance and fault management of distribution networks, and helping to improve the overall operating efficiency and reliability of the power system.

[0009] To achieve the above objectives, this application provides the following technical solutions:

[0010] In a first aspect, an embodiment of the present application provides a method for classifying causes of grounding faults in a distribution network based on HOA-CNN, comprising the following steps:

[0011] Through artificial grounding test, different types of grounding faults are simulated to obtain zero-sequence current waveform data under various working conditions of distribution network grounding faults caused by different reasons;

[0012] The sliding time window method is used to perform slicing operations and process the zero-sequence current waveform data;

[0013] One-hot encoding is performed on the data labels of each working condition;

[0014] After the data is processed, it is randomly shuffled, 80% of the shuffled data is extracted as the training set, and the remaining 20% ​​is used as the test set;

[0015] Establish a CNN model and determine the basic structure, loss function, optimizer and activation function of the model;

[0016] The HOA optimization algorithm is used to reasonably set the number of convolution kernels, convolution kernel size, number of pooling kernels, and pooling kernel size of the CNN model.

[0017] Input the training set data into the HOA-CNN model and continue training until the preset number of iterations is reached to complete the HOA-CNN model training;

[0018] The test set is input into the trained HOA-CNN model to test the classification ability of the HOA-CNN model.

[0019] The obtaining of zero-sequence current waveform data under various operating conditions of distribution network grounding faults caused by different reasons also includes analyzing actual grounding faults using real fault recording data, thereby obtaining more real and rich zero-sequence current waveform data.

[0020] The sliding time window method is used to perform slicing operation to process the zero-sequence current waveform data. Specifically, each operating condition contains N data points, and the slicing step is set to , which means that each slide will skip The next slicing is performed with n data points, and each sample is set to contain n data points to ensure the adequacy and representativeness of the sample. After such slicing, each working condition finally contains A data samples, and the relationship is shown in the formula:

[0021] (1).

[0022] The number of training set samples Q is shown in Formula 2, and the number of test set samples W is shown in Formula 3.

[0023] (2)

[0024] (3).

[0025] The specific contents of establishing a CNN model, determining the basic structure of the model, loss function, optimizer and activation function are as follows:

[0026] The cross entropy loss function is used, as shown in Formula 4;

[0027] (4)

[0028] is the predicted probability; is the actual label; L is the predicted result,

[0029] The optimizer updates the model parameters according to the gradient information of the loss function. During the training process, the model calculates the gradient of the loss function with respect to the model parameters through the back-propagation algorithm, and then the optimizer uses these gradients to update the model weights and biases, so that the value of the loss function gradually decreases. The stochastic gradient descent optimizer is used. The main function of the activation function is to introduce nonlinear characteristics to the neural network, so that the network can learn and represent more complex functional relationships. The tanh function is used, as shown in Formula 5. For any real number input x, the result of tanh(x) is another real number between -1 and 1, excluding -1 and 1. This is determined by the definition of the hyperbolic tangent function.

[0030] (5),

[0031] Here, e is the base of natural logarithms.

[0032] The HOA optimization algorithm is used to reasonably set the parameters of the CNN model, such as the number of convolution kernels, the size of convolution kernels, the number of pooling kernels, and the size of pooling kernels. Specifically,

[0033] First, the recognition accuracy of the CNN model on the validation set is used as the optimization target. The number of convolution kernels, the size of convolution kernels, the number of pooling kernels, and the size of pooling kernels are used as decision variables for the optimization problem and are set.

[0034] Among them, the optimization range of the number of convolution kernels and pooling kernels is [1, 100], and the optimization range of the size of convolution kernels and pooling kernels is [1, 20]. Then, the HOA algorithm is initialized, and each hiker represents a potential solution, that is, a set of specific number and size of convolution kernels, number and size of pooling kernels. During initialization, multiple positions of hikers are randomly generated, and these positions correspond to different combinations of number and size of convolution kernels and number and size of pooling kernels.

[0035] For each hiker, that is, each combination of the number and size of convolution kernels and the number and size of pooling kernels, the corresponding CNN model is trained and its performance is evaluated on the validation set. Here, the accuracy is used as the optimization target, and the evaluation result is used as the fitness value of the hiker. The higher the fitness value, the better the performance of the hiker, that is, the parameter combination.

[0036] According to Tobler's hiking function and current speed, the speed of each hiker is updated. The hiking function takes into account the terrain, i.e. the performance of the parameter combination, and the distance, i.e. the amplitude of the parameter change. According to the updated speed, the position of the hiker is adjusted, i.e. the number and size of each convolution kernel group, and the number and size of the pooling kernel group are adjusted.

[0037] In each generation, the hiker with the highest fitness is selected as the leader hiker, whose position may be the optimal solution found so far. The above steps are repeated until the maximum number of iterations is reached. After the optimization process is completed, the position of the leader hiker, that is, the optimal number and size combination of convolution kernels, is output. The CNN model is retrained using this set of optimal parameters, and its final performance is evaluated on the test set.

[0038] In a second aspect, an embodiment of the present application provides a distribution network grounding fault cause classification system based on HOA-CNN, the system comprising: a memory and a processor, the memory comprising a program of a distribution network grounding fault cause classification method based on HOA-CNN, and the program of the distribution network grounding fault cause classification method based on HOA-CNN is executed by the processor to implement the following steps: simulate different types of grounding fault conditions through artificial grounding tests, and obtain zero-sequence current waveform data under various working conditions of distribution network grounding faults caused by different reasons; use a sliding time window method to perform slicing operations to process the zero-sequence current waveform data; one-hot label each working condition data. encoding; after the data is processed, randomly shuffle the data, extract 80% of the shuffled data as the training set, and the remaining 20% ​​as the test set; establish a CNN model, determine the basic structure of the model, loss function, optimizer and activation function content; use the HOA optimization algorithm to reasonably set the number of convolution kernels, convolution kernel size, number of pooling kernels, and pooling kernel size of the CNN model; input the training set data into the HOA-CNN model, and continue training until the preset number of iterations is reached to complete the HOA-CNN model training; input the test set into the trained HOA-CNN model for testing to detect the classification ability of the HOA-CNN model.

[0039] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a program code, and when the program code is executed by a processor, the steps of the distribution network grounding fault cause classification method based on HOA-CNN as described above are implemented.

[0040] Compared with the prior art, the beneficial effects of the present invention are: utilizing the global search capability and adaptive optimization mechanism of the walking optimization algorithm (HOA) to optimize the parameters in the CNN model, improve the model convergence speed and classification accuracy, and avoid falling into the local optimal solution.

[0041] Reduce computing resource consumption: Through the effective guidance of the HOA algorithm, reduce invalid iterations in the CNN model training process, reduce computing resource consumption, and improve the practicality and promotion value of the method.

[0042] Improve the robustness of initial parameter settings: The HOA algorithm can adaptively adjust the search direction, reduce the impact of initial parameter settings on model performance, and improve the stability of the model under different initial conditions.

[0043] Enhance the adaptability of the model in specific scenarios: According to the specific needs of distribution network grounding fault classification, the CNN model is customized and optimized through the HOA algorithm to improve the classification performance of the model in specific fault scenarios.

[0044] Promoting the intelligent development of distribution networks: The distribution network grounding fault cause classification method based on HOA-CNN proposed in this study combines intelligent optimization algorithms with deep learning technology, providing new ideas and technical means for the intelligent operation and maintenance and fault management of distribution networks, which helps to improve the overall operation efficiency and reliability of the power system.

[0045] This application optimizes the convolutional neural network (CNN) parameters by introducing the walking optimization algorithm (HOA), and proposes a more efficient and accurate method for classifying the causes of grounding faults in distribution networks, contributing to the intelligent development of power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 It is a flow chart of the method of this application.

[0048] Figure 2 This is the CNN model structure diagram of this application. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0050] The terms "comprises," "comprising," or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0051] The terms "first", "second", etc. are only used to distinguish one entity or operation from another entity or operation, and should not be understood as indicating or implying relative importance, nor should they be understood as requiring or implying any such actual relationship or order between these entities or operations.

[0052] In this embodiment, simulation verification is used to further explain and illustrate the method proposed in this patent. The simulation model is built using Pscad, and its specific parameters are consistent with a domestic ±10k medium-voltage distribution system. The training and test data sets are generated using model simulation, and data feature extraction and HOA-CNN neural network algorithm are implemented in Python using the keras library.

[0053] like Figure 1 As shown, the distribution network grounding fault cause classification method using the HOA-CNN neural network proposed in this application includes the following steps.

[0054] (1) Build a simulation model using pscad software.

[0055] (2) Simulate various working conditions to collect experimental data, and each working condition contains N data points.

[0056] (3) Use the sliding time window method to perform slicing. Set the slicing step size to , which means that each slide will skip data points for the next slicing.

[0057] (4) Each sample is set to contain n data points to ensure the adequacy and representativeness of the sample.

[0058] (5) Each working condition ultimately contains A data samples.

[0059] (6) In order to match the output format of the model, the input label also needs to appear in the form of probability distribution. Therefore, one-hot encoding is performed on the data label of each working condition. 16 labels are set to correspond to 16 different grounding faults.

[0060] (7) In order to improve the generalization ability of the model, the data is randomly shuffled.

[0061] (8) 80% of the shuffled data is extracted as the training set, and the remaining 20% ​​is used as the test set. It should be noted that since the data is shuffled, the distribution of data for each working condition may be uneven.

[0062] (9) Establish a CNN model and determine the basic structure, loss function, optimizer, activation function, etc.

[0063] (10) Set the corresponding parameters of the HOA algorithm.

[0064] (11) Use HOA to optimize the hyperparameters in the CNN network and obtain the optimized HOA-CNN model;

[0065] (12) Use the training set to train the optimized HOA-CNN model to generate a prediction model;

[0066] (13) The test set feature matrix is ​​input into the trained HOA-CNN neural network to obtain the predicted class label, and then the fault cause classification result is obtained.

[0067] Furthermore, in step (9), the convolution layer is the core component of CNN. The convolution layer scans the input data through a series of learnable filters (or convolution kernels) to detect specific local features. These filters can automatically learn and capture low-level features such as edges, corners, textures, etc., providing richer and more meaningful information for subsequent layers. The formula is as follows:

[0068] ,

[0069] Where: f is the activation function; For the The output of the layer Features, For the Tier convolution kernel weight matrix; is the bias term; For the Layer output.

[0070] The pooling layer usually follows the convolutional layer. Its main function is to downsample and reduce the dimension of the data while retaining important features. This step not only helps to reduce the amount of calculation, but also improves the spatial invariance of the model, that is, it has a certain robustness to small position changes of the input data.

[0071] ,

[0072] Where: The first Tier Elements of the feature map; Indicates Pooling area; After pooling Layer elements in the feature map.

[0073] This model uses the cross entropy loss function, the formula is as follows:

[0074] ,

[0075] The model calculates the gradient of the loss function with respect to the model parameters through the back-propagation algorithm, and then the optimizer uses these gradients to update the weights and biases of the model so that the value of the loss function gradually decreases. This experiment uses the stochastic gradient descent optimizer. The main function of the activation function is to introduce nonlinear characteristics to the neural network so that the network can learn and represent more complex functional relationships. This experiment uses the tanh function.

[0076] Furthermore, in step (10), the mathematical basis of HOA is based on the well-known Tobler walking function, which is formulated as:

[0077] ,

[0078] in is the speed of the hiker at iteration or time t (i.e. km / h), is the slope of the trail or terrain, and the formula is:

[0079] ,

[0080] in, and Respectively represent the difference in altitude and distance covered by the hiker. In addition, is the inclination angle of the trajectory or terrain, in the range [0,50°].

[0081] The current speed of hiker i is:

[0082] ,

[0083] In the formula, is a number uniformly distributed in the range [0, 1]; and denote the current speed and initial speed of hiker i respectively. is the position of the leading hiker, is the scan factor (SF), which ranges from [1, 3].

[0084] The HOA optimization algorithm is used to reasonably set the parameters of the CNN model, such as the number of convolution kernels, the size of convolution kernels, the number of pooling kernels, and the size of pooling kernels. Specifically,

[0085] First, the recognition accuracy of the CNN model on the validation set is used as the optimization target. The number of convolution kernels, the size of convolution kernels, the number of pooling kernels, and the size of pooling kernels are used as decision variables for the optimization problem and are set accordingly.

[0086] The optimization range of the number of convolution kernels and pooling kernels is [1, 100], and the optimization range of the size of convolution kernels and pooling kernels is [1, 20]. Then, the HOA algorithm is initialized, and each walker represents a potential solution, that is, a specific number and size of convolution kernels, and number and size of pooling kernels. During initialization, multiple walker positions are randomly generated, which correspond to different combinations of number and size of convolution kernels and number and size of pooling kernels.

[0087] For each hiker (i.e., each combination of the number and size of convolution kernels and the number and size of pooling kernels), train the corresponding CNN model and evaluate its performance on the validation set, where accuracy is used as the optimization target. Use the evaluation result as the fitness value of the hiker. The higher the fitness value, the better the performance of the hiker (i.e., parameter combination).

[0088] Update the speed of each hiker based on Tobler's hiking function and current speed. The hiking function takes into account the terrain (i.e. the performance of the parameter combination) and the distance (i.e. the magnitude of the parameter change). Adjust the position of the hiker based on the updated speed (i.e. adjust the number and size of each convolution kernel, and the number and size of the pooling kernel).

[0089] In each generation, the hiker with the highest fitness is selected as the leader hiker, whose position may be the optimal solution found so far. Repeat the above steps until the maximum number of iterations is reached. After the optimization process is completed, the position of the leader hiker is output, that is, the optimal number and size combination of convolution kernels. The CNN model is retrained using this set of optimal parameters, and its final performance is evaluated on the test set.

[0090] Furthermore, in step (11), the hyperparameters to be optimized in the CNN network are the number of convolution kernels, with a search range of [1,100]; the convolution kernel size, with a search range of [1,20]; the number of pooling kernels, with a search range of [1,100]; the pooling kernel size, with a search range of [1,20]; and the number of iterations is 20.

[0091] The present application embodiment provides a method for classifying causes of grounding faults in a distribution network based on HOA-CNN, comprising the following steps:

[0092] (1) In order to further study the characteristics of distribution network grounding faults, different types of grounding faults are simulated through artificial grounding tests to obtain zero-sequence current waveform data of distribution network grounding faults caused by different reasons. At the same time, the actual grounding faults can also be analyzed using real fault recording data, so as to obtain more real and rich zero-sequence current waveform data, providing strong support for accurate diagnosis and handling of faults. The working conditions are shown in Table 1.

[0093] Table 1 Fault classification table

[0094]

[0095] (2) In this experiment, each working condition contains N data points, and the slicing step size is set to , which means that each slide will skip The next slicing is performed with n data points, and each sample is set to contain n data points to ensure the adequacy and representativeness of the sample. After such slicing, each working condition finally contains A data samples, and the relationship is shown in the formula:

[0096] (1)

[0097] (3) In deep learning, especially when using the softmax layer for multi-classification, the output of the model is a probability distribution. In order to match the output format of the model, the input label also needs to appear in the form of a probability distribution, so that the cross entropy loss can be directly calculated to evaluate the accuracy of the model prediction. Therefore, one-hot encoding is performed on the data labels of each working condition. The encoding is shown in Table 2.

[0098] Table 2 Data label table

[0099]

[0100] (4) After preprocessing the data, in order to improve the generalization ability of the model, the data is first randomly shuffled. Then, 80% of the shuffled data is extracted as the training set, and the remaining 20% ​​is used as the test set. It should be noted that since the data is shuffled, the distribution of data for each working condition may be uneven. The number of training set samples Q is shown in Formula 2. The number of test set samples W is shown in Formula 3.

[0101] (2)

[0102] (3)

[0103] (5) If Figure 2, establish a CNN model, determine the basic structure of the model, loss function, optimizer and activation function, etc. The main function of the loss function is to estimate the difference or error between the predicted value and the true value of the model. This experiment uses the cross entropy loss function, as shown in Formula 4; the optimizer updates the parameters of the model according to the gradient information of the loss function. During the training process, the model calculates the gradient of the loss function with respect to the model parameters through the back-propagation algorithm, and then the optimizer uses these gradients to update the weights and biases of the model, so that the value of the loss function gradually decreases. This experiment uses the stochastic gradient descent optimizer; the main function of the activation function is to introduce nonlinear characteristics to the neural network, so that the network can learn and represent more complex functional relationships. This experiment uses the tanh function, as shown in Formula 5. The basic structure of the CNN model is as follows Figure 2 shown.

[0104] (4)

[0105] (5)

[0106] (6) The hyperparameters of the HOA optimization algorithm were reasonably set. The parameters are shown in Table 2. Subsequently, the training set data was input into the HOA-CNN model, and the training was continued until the preset number of iterations was reached. At this point, the training process of the HOA-CNN model was successfully completed.

[0107] Table 3 HOA parameter setting table

[0108]

[0109] (7) Input the reserved test set into the trained HOA-CNN model for testing. Test the classification ability of the HOA-CNN model.

[0110] Table 4 shows the accuracy comparison of different optimization algorithms

[0111]

[0112] It can be seen from the above table that the solution of the present application improves the model convergence speed and classification accuracy, and avoids falling into the local optimal solution.

[0113] The embodiment of the present application provides a distribution network grounding fault cause classification system based on HOA-CNN, the system comprising: a memory and a processor, the memory comprising a distribution network grounding fault cause classification method based on HOA-CNN, the program of the distribution network grounding fault cause classification method based on HOA-CNN when executed by the processor implements the following steps: simulating different types of grounding fault conditions through artificial grounding tests, obtaining zero-sequence current waveform data under various working conditions of distribution network grounding faults caused by different reasons; using a sliding time window method to perform slicing operations to process the zero-sequence current waveform data; one-hot encoding the data labels of each working condition ; After the data is processed, randomly shuffle the data, extract 80% of the shuffled data as the training set, and the remaining 20% ​​as the test set; establish a CNN model, determine the basic structure of the model, loss function, optimizer and activation function content; use the HOA optimization algorithm to reasonably set the number of convolution kernels, convolution kernel size, number of pooling kernels, and pooling kernel size of the CNN model; input the training set data into the HOA-CNN model, and continue training until the preset number of iterations is reached to complete the HOA-CNN model training; input the test set into the trained HOA-CNN model for testing to detect the classification ability of the HOA-CNN model.

[0114] An embodiment of the present application provides a computer-readable storage medium, which stores program code. When the program code is executed by a processor, the steps of the distribution network grounding fault cause classification method based on HOA-CNN as described above are implemented.

[0115] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0116] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0117] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0119] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0120] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0121] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0122] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A distribution network grounding fault cause classification method based on HOA-CNN, characterized in that: The following steps are involved: Through artificial grounding test, different types of grounding faults are simulated to obtain zero-sequence current waveform data under various working conditions of distribution network grounding faults caused by different reasons; The sliding time window method is used to perform slicing operations and process the zero-sequence current waveform data; One-hot encoding is performed on the data labels of each working condition; After the data is processed, it is randomly shuffled, 80% of the shuffled data is extracted as the training set, and the remaining 20% ​​is used as the test set; Establish a CNN model and determine the basic structure, loss function, optimizer and activation function of the model; The HOA optimization algorithm is used to reasonably set the number of convolution kernels, convolution kernel size, number of pooling kernels, and pooling kernel size of the CNN model. Input the training set data into the HOA-CNN model and continue training until the preset number of iterations is reached to complete the HOA-CNN model training; The test set is input into the trained HOA-CNN model to test the classification ability of the HOA-CNN model. The HOA optimization algorithm is used to reasonably set the parameters of the CNN model, such as the number of convolution kernels, the size of convolution kernels, the number of pooling kernels, and the size of pooling kernels. Specifically, First, the recognition accuracy of the CNN model on the validation set is used as the optimization target. The number of convolution kernels, the size of convolution kernels, the number of pooling kernels, and the size of pooling kernels are used as decision variables for the optimization problem and are set. Among them, the optimization range of the number of convolution kernels and pooling kernels is [1, 100], and the optimization range of the size of convolution kernels and pooling kernels is [1, 20]. Then, the HOA algorithm is initialized, and each hiker represents a potential solution, that is, a set of specific number and size of convolution kernels, number and size of pooling kernels. During initialization, multiple positions of hikers are randomly generated, and these positions correspond to different combinations of number and size of convolution kernels and number and size of pooling kernels. For each hiker, that is, each combination of the number and size of convolution kernels and the number and size of pooling kernels, the corresponding CNN model is trained and its performance is evaluated on the validation set. Here, the accuracy is used as the optimization target, and the evaluation result is used as the fitness value of the hiker. The higher the fitness value, the better the performance of the hiker, that is, the parameter combination. According to Tobler's hiking function and current speed, the speed of each hiker is updated. The hiking function takes into account the terrain, i.e. the performance of the parameter combination, and the distance, i.e. the amplitude of the parameter change. According to the updated speed, the position of the hiker is adjusted, i.e. the number and size of each convolution kernel group, and the number and size of the pooling kernel group are adjusted. In each generation, the hiker with the highest fitness is selected as the leader hiker, whose position may be the optimal solution found so far. The above steps are repeated until the maximum number of iterations is reached. After the optimization process is completed, the position of the leader hiker, that is, the optimal number and size combination of convolution kernels, is output. The CNN model is retrained using this set of optimal parameters, and its final performance is evaluated on the test set.

2. The method for classifying grounding fault causes in distribution network based on HOA-CNN according to claim 1, characterized in that: The obtaining of zero-sequence current waveform data under various operating conditions of distribution network grounding faults caused by different reasons also includes analyzing actual grounding faults using real fault recording data, thereby obtaining more real and rich zero-sequence current waveform data.

3. The method for classifying grounding fault causes in distribution network based on HOA-CNN according to claim 1, characterized in that: The sliding time window method is used to perform slicing operation and process the zero-sequence current waveform data. Specifically, each operating condition contains N data points, and the slicing step is set to , which means that each slide will skip The next slicing is performed with n data points, and each sample is set to contain n data points to ensure the adequacy and representativeness of the sample. After such slicing, each working condition finally contains A data samples, and the relationship is shown in the formula: （1）。 4. The method for classifying grounding fault causes in distribution network based on HOA-CNN according to claim 1, characterized in that: The number of training set samples Q is shown in Formula 2, and the number of test set samples W is shown in Formula 3. （2） （3）。 5. The method for classifying grounding fault causes in distribution network based on HOA-CNN according to claim 1, characterized in that: The specific contents of establishing a CNN model, determining the basic structure of the model, loss function, optimizer and activation function are as follows: The cross entropy loss function is used, as shown in Formula 4; （4） is the predicted probability; is the actual label; L is the prediction result, The optimizer updates the model parameters according to the gradient information of the loss function. During the training process, the model calculates the gradient of the loss function with respect to the model parameters through the back-propagation algorithm, and then the optimizer uses these gradients to update the model weights and biases, so that the value of the loss function gradually decreases. The stochastic gradient descent optimizer is used. The main function of the activation function is to introduce nonlinear characteristics to the neural network, so that the network can learn and represent more complex functional relationships. The tanh function is used, as shown in Formula 5. For any real number input x, the result of tanh(x) is another real number between -1 and 1, excluding -1 and 1. This is determined by the definition of the hyperbolic tangent function. （5）， Here, e is the base of natural logarithms.

6. A distribution network grounding fault cause classification system based on HOA-CNN, used to implement any method of claims 1-5, characterized in that: The system includes: a memory and a processor, wherein the memory includes a program of a distribution network grounding fault cause classification method based on HOA-CNN, and when the program of the distribution network grounding fault cause classification method based on HOA-CNN is executed by the processor, the following steps are implemented: simulating different types of grounding fault conditions through artificial grounding tests, and obtaining zero-sequence current waveform data under various working conditions of distribution network grounding faults caused by different reasons; using a sliding time window method to perform slicing operations and process the zero-sequence current waveform data; one-hot encoding the data labels of each working condition; and after the data is processed, randomly shuffling the data. 80% of the shuffled data is extracted as the training set, and the remaining 20% ​​is used as the test set; a CNN model is established to determine the basic structure of the model, loss function, optimizer, and activation function content; the number of convolution kernels, convolution kernel size, number of pooling kernels, and pooling kernel size of the CNN model are reasonably set through the HOA optimization algorithm; the training set data is input into the HOA-CNN model, and training is continued until the preset number of iterations is reached to complete the HOA-CNN model training; the test set is input into the trained HOA-CNN model for testing to detect the classification ability of the HOA-CNN model.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, and when the program code is executed by a processor, the steps of the method for classifying causes of ground faults in a distribution network based on HOA-CNN are implemented as described in any one of claims 1 to 5.

Images