High-resistance fault detection method and device, computer equipment and storage medium

Abstract

The invention relates to a high-resistance fault detection method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: when a high-resistance fault of a line occurs, acquiring a zero-sequence current signal; performing denoising processing on the zero-sequence current signal, acquiring sample data of a normal current state and a fault current state with a time sequence, and performing multi-dimensional fault feature extraction according to the sample data to obtain a high-resistance fault feature data set containing multiple fault types; training a random forest neural network according to the high-resistance fault feature data set; and detecting the fault type of the high-resistance fault by using the trained random forest neural network. By adopting the method, multiple types of high-resistance faults can be accurately distinguished, so that the reliability and pertinence of fault processing of the power distribution network are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of resistance detection, in particular to a high-resistance fault detection method and device, computer equipment, computer readable storage medium and computer program product. BACKGROUND

[0002] With the expansion of the power distribution network and the increasing complexity of the operating environment, single-phase high-resistance grounding faults occur frequently, and it is difficult for traditional protection devices and algorithms to effectively detect them due to their weak fault currents and non-obvious characteristics. The structure of urban streets is complex, and the grounding medium is diverse (such as trees, cement, asphalt, etc.), resulting in a large number of fault types and overlapping characteristics, limited applicability of single detection methods, and high misjudgment rate.

[0003] Current high-resistance fault detection methods are mainly based on steady-state quantities, transient quantities, signal injection, or machine learning, but still have obvious limitations: steady-state and transient characteristics are easily lost or weak under high-resistance conditions; signal injection method relies on impedance characteristics and is limited in application scenarios; machine learning can extract multiple source characteristics, but its generalization ability is still insufficient when facing multiple types of high-resistance faults. In addition, current methods only determine whether a fault has occurred, and cannot identify the specific type of fault, which cannot provide effective guidance for operation and maintenance.

[0004] Therefore, there is an urgent need for a high-resistance fault detection method, device, computer equipment, computer readable storage medium and computer program product that can accurately distinguish between multiple types of high-resistance faults to improve the reliability and relevance of power distribution network fault handling. SUMMARY

[0005] Therefore, there is an urgent need for a high-resistance fault detection method, device, computer equipment, computer readable storage medium and computer program product that can accurately distinguish between multiple types of high-resistance faults to improve the reliability and relevance of power distribution network fault handling.

[0006] In a first aspect, the application provides a high-resistance fault detection method, comprising:

[0007] When a high-resistance fault occurs in a line, a zero-sequence current signal is obtained;

[0008] The zero-sequence current signal is denoised, sample data of normal current states and fault current states with time series are collected, and multi-dimensional fault feature extraction is performed according to the sample data to obtain a high-resistance fault feature data set containing multiple fault types;

[0009] According to the high-resistance fault feature data set, a random forest neural network is trained;

[0010] The trained random forest neural network is used to detect the fault type of the high-resistance fault.

[0011] In one embodiment, the detection of high-resistance fault types using a trained random forest neural network includes:

[0012] The multidimensional fault features detected in real time are input into multiple decision trees in the random forest neural network;

[0013] Each decision tree is used to perform fault analysis on the multidimensional fault features detected in real time, and the classification voting results of different fault types are output. The classification voting results reflect the judgment results of the decision tree on the local features of the fault.

[0014] Aggregate the classification voting results of all decision trees and determine the fault type with the highest number of votes as the high-resistance fault type.

[0015] In one embodiment, training the random forest neural network includes:

[0016] When constructing multiple decision trees in a random forest neural network, a sampling with replacement method is used to randomly select a subset of samples from the high-resistivity fault feature dataset, and the decision trees in the random forest neural network are trained using the subset of samples.

[0017] When splitting nodes in each decision tree, a subset of features is randomly selected from all multidimensional fault features, and the optimal splitting rule for the decision tree node is determined using the subset of features.

[0018] In one embodiment, determining the optimal splitting rule for the decision tree node using the feature subset includes:

[0019] For each feature in the randomly selected feature subset, calculate the information entropy and conditional entropy corresponding to different split points;

[0020] Based on the information entropy and the conditional entropy, calculate the information gain rate corresponding to each split point;

[0021] From all candidate split points, select the split point with the largest information gain ratio as the optimal split feature and split point for the current node.

[0022] In one embodiment, the step of extracting multidimensional fault features from the sample data to obtain a high-resistance fault feature dataset containing multiple fault types includes:

[0023] The sample data is subjected to time-frequency domain analysis to extract frequency domain features containing third and fifth harmonic components. These features are then combined with time-domain statistical features and energy features to construct multidimensional feature vectors for samples of different fault types. The time-domain statistical features include integral, zero-crossing duration, variance, skewness, and peak value. The energy features include energy entropy and Shannon entropy.

[0024] The multidimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct the high-resistance fault feature dataset.

[0025] In one embodiment, training a random forest neural network based on the high-resistance fault feature dataset includes:

[0026] The Hyperband hyperparameter optimization algorithm is used to optimize the hyperparameters of the random forest neural network using the training set and the validation set, so as to obtain a random forest neural network with the optimal combination of hyperparameters.

[0027] The training set is used to train the random forest neural network with the optimal hyperparameter combination to generate a high-resistance fault detection model.

[0028] The high-resistance fault detection model is tested using the test set to verify its ability to classify and identify various high-resistance fault types and normal current states.

[0029] Secondly, this application also provides a high-resistance fault detection device, comprising:

[0030] The signal acquisition module is used to acquire the zero-sequence current signal when a high-resistance fault occurs in the line.

[0031] The signal processing module is used to denoise the zero-sequence current signal, collect sample data of normal current state and fault current state with time series, and extract multi-dimensional fault features based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0032] The model training module is used to train a random forest neural network based on the high-resistivity fault feature dataset.

[0033] The fault detection module is used to detect the fault type of high-resistance faults using a trained random forest neural network.

[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0035] When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired;

[0036] The zero-sequence current signal is denoised, and sample data of normal current state and fault current state with time series are collected. Multidimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0037] A random forest neural network is trained based on the high-resistivity fault feature dataset.

[0038] The trained random forest neural network is used to detect the fault type of high-resistance faults.

[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0040] When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired;

[0041] The zero-sequence current signal is denoised, and sample data of normal current state and fault current state with time series are collected. Multidimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0042] A random forest neural network is trained based on the high-resistivity fault feature dataset.

[0043] The trained random forest neural network is used to detect the fault type of high-resistance faults.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0045] When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired;

[0046] The zero-sequence current signal is denoised, and sample data of normal current state and fault current state with time series are collected. Multidimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0047] A random forest neural network is trained based on the high-resistivity fault feature dataset.

[0048] The trained random forest neural network is used to detect the fault type of high-resistance faults.

[0049] The aforementioned high-resistivity fault detection methods, devices, computer equipment, computer-readable storage media, and computer program products significantly enhance the characterization and discrimination capabilities of faults caused by different grounding media (such as asphalt, sand, and tree branches) by integrating multi-dimensional fault feature extraction in the time, frequency, and energy domains. They overcome the shortcomings of traditional single-feature detection algorithms, which suffer from poor adaptability and high false positive rates under high-resistivity conditions. By utilizing a random forest neural network to integrate multiple decision trees for collaborative decision-making, overfitting is effectively suppressed, improving the classification accuracy and generalization ability for various high-resistivity fault types and normal states in complex distribution network environments. This represents a technological leap from binary judgment of "whether it is a fault" to multi-class identification of "what kind of fault," providing maintenance personnel with more instructive fault type information, thereby supporting precise inspection and fault handling and improving the intelligent level of distribution network operation and maintenance. Furthermore, the robustness of the fault detection system to noisy environments and sample imbalance is improved. Through denoising preprocessing of zero-sequence current signals, Bootstrap random sampling, and random feature selection, the stability and practicality of the method in real-world complex scenarios are ensured. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is an application environment diagram of a high-resistivity fault detection method in one embodiment;

[0052] Figure 2 This is a flowchart illustrating a high-resistance fault detection method in one embodiment;

[0053] Figure 3 This is a flowchart illustrating a high-resistance fault detection method in another embodiment;

[0054] Figure 4 This is a wiring diagram of the experimental platform in the most detailed embodiment;

[0055] Figure 5 The flowchart for constructing the high-resistivity fault identification and classification model in the most detailed embodiment is shown below.

[0056] Figure 6 This is a schematic diagram illustrating the principle of model hyperparameter optimization in the most detailed embodiment;

[0057] Figure 7 This is a schematic diagram of the random forest algorithm structure in the most detailed embodiment;

[0058] Figure 8 This is a structural block diagram of a high-resistance fault detection device in one embodiment;

[0059] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0060] With the continuous development of power distribution networks, their operating environment is becoming increasingly complex, leading to frequent grounding faults, with single-phase grounding faults accounting for the largest proportion. If not handled promptly, these faults can easily lead to complex grounding faults, posing a serious threat to personal safety and power distribution equipment. The complex structure of urban streets makes high-resistance faults highly complex. Typically, the current in these grounding faults is extremely weak, making it difficult for existing protection devices and traditional fault protection algorithms to effectively detect them. Furthermore, high-resistance faults occur in many different types, and the fault feature value extraction for a single high-resistance type suffers from a high false positive rate when different types of high-resistance faults occur, making it impossible to guarantee applicability to different types of high-resistance faults.

[0061] Currently, high-resistivity grounding fault detection technologies mainly revolve around steady-state characteristics, transient characteristics, active signal injection, and machine learning. The zero-sequence current detection method based on steady-state quantities collects the zero-sequence current of each outgoing line using a zero-sequence current transformer, and determines the faulty line by comparing amplitude, phase, or power direction. The traveling wave method based on transient quantities utilizes the reflection characteristics of the traveling wave generated by the fault between the bus and the fault point to calculate the fault distance; the transient energy method distinguishes between faulty and non-faulty lines by extracting the polarity, magnitude, or distribution characteristics of transient zero-sequence energy. The signal injection method addresses scenarios where the steady-state / transient characteristics disappear due to weak high-resistivity fault current. It injects a low-frequency signal at the neutral point of a voltage transformer, utilizing the low impedance of the faulty line to ground, and detects the attenuation characteristics of the injected signal at the end of the faulty line to locate the fault section. In recent years, machine learning and deep learning technologies have been introduced to improve fault identification capabilities in complex scenarios by mining the correlation features of multi-source data.

[0062] However, traditional technologies still have many shortcomings. When high-resistance grounding faults occur in cities, they are difficult to reliably identify due to the complexity of the lines and street structures, the typically large transition resistance, high concealment, and weak electrical quantity abrupt changes. Furthermore, high-resistance faults occur frequently, and extracting fault feature values ​​for a single high-resistance type results in a high false alarm rate when different types of high-resistance faults occur, and the applicability to different types of high-resistance faults cannot be guaranteed. Multiple types of high-resistance faults have overlapping characteristics, and the robustness of a single feature is insufficient. Urban high-resistance fault types are complex (grounding media include trees, cement, asphalt, sand, etc.), and the differences in zero-sequence current amplitude, phase, and harmonic characteristics between different types of faults are subtle. Traditional amplitude and phase comparison methods are prone to misjudgment. The above high-resistance fault identification methods only address whether a high-resistance fault has occurred, without specifying the type of high-resistance fault. Many high-resistance fault scenarios exist, such as unlined lines contacting the ground, branches, concrete, etc., or cable aging leading to insulation damage and direct contact with the ground. Simply identifying whether a fault has occurred provides no guidance for maintenance personnel. Therefore, there is an urgent need to develop intelligent detection technologies that can accurately distinguish between various types of high-resistance faults in order to improve the reliability and targetedness of power distribution network fault handling.

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0065] The high-resistance fault detection method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0066] When a high-resistance fault occurs on the line, server 104 controls terminal 102 to acquire zero-sequence current signal; server 104 performs noise reduction processing on zero-sequence current signal, collects sample data of normal current state and fault current state with time series, and performs multi-dimensional fault feature extraction based on sample data to obtain a high-resistance fault feature dataset containing multiple fault types; based on the high-resistance fault feature dataset, a random forest neural network is trained; and the trained random forest neural network is used to detect the fault type of high-resistance fault.

[0067] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0068] In one exemplary embodiment, such as Figure 2 As shown, a high-resistance fault detection method is provided, which is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S202 to S208. Wherein:

[0069] Step S202: When a high-resistance fault occurs in the line, acquire the zero-sequence current signal.

[0070] Specifically, a high-resistance fault refers to a non-metallic connection between the conductor of a distribution network line and the ground or other objects (such as trees, asphalt, or cement) through a high transition resistance (usually greater than 100 ohms). Its significant characteristics are weak fault current and inconspicuous electrical characteristics.

[0071] The zero-sequence current signal refers to the vector sum signal that reflects the degree of imbalance of the three-phase current under the above-mentioned fault conditions. Its mathematical expression is I0=(I a +I b +I c ), is the most direct and sensitive characteristic quantity for identifying grounding faults.

[0072] The means of acquisition include:

[0073] The current is directly acquired via a zero-sequence current transformer (CT), or by acquiring signals from a three-phase current transformer and performing digital calculations (I0=I...) via a summing circuit or a microprocessor. a +I b +I c )get.

[0074] Step S204: Denoise the zero-sequence current signal, collect sample data of normal current state and fault current state with time series, and extract multi-dimensional fault features based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0075] Specifically, the purpose of denoising zero-sequence current signals is to eliminate noise interference (such as electromagnetic interference and measurement errors) in the original signal and enhance the significance of effective fault characteristics. Signal processing techniques such as wavelet transform and empirical mode decomposition (EMD) are typically used to separate noise and fault components.

[0076] Sample data of normal and fault current states with time series were collected. Each sample should contain continuous time series data of several cycles before the fault (normal state) and dozens of cycles after the fault (fault state) to provide a basis for comparative analysis. From the denoised time-domain signal, multi-dimensional quantitative indicators that can characterize different fault types were extracted through mathematical transformation and statistical calculation.

[0077] Specific characteristics include: frequency domain characteristics, such as the third harmonic component and the fifth harmonic component (reflecting the nonlinear characteristics of the fault); time domain statistical characteristics, such as variance (degree of fluctuation), skewness (waveform asymmetry), peak value (abrupt change intensity), and integral (energy accumulation); and energy entropy characteristics, such as Shannon entropy (signal complexity) and energy entropy (uniformity of energy distribution).

[0078] All extracted feature vectors are associated with corresponding fault type labels (such as "tree branch grounding", "cement grounding", "normal") to form a structured dataset for subsequent training and validation of machine learning models.

[0079] Step S206: Train a random forest neural network based on the high-resistivity fault feature dataset.

[0080] Specifically, Random Forest is a machine learning algorithm that makes decisions by ensembled multiple decision trees. Its name comes from its "forest"-like ensemble structure and the randomness inherent within it. The training process aims to allow the Random Forest model to learn classification patterns for different fault types from data. This involves two core randomization processes: Bootstrap sampling (random samples): multiple subsets of samples are randomly drawn with replacement from the dataset to train different decision trees in the forest, increasing model diversity and generalization ability. Random feature selection: when splitting at each node of each tree, only a subset of features is randomly selected from all features to find the optimal split point, avoiding a few strong features dominating all trees, further improving model diversity and robustness.

[0081] After training, a strong classifier composed of multiple decision trees (i.e., the "high-resistance fault detection model") is obtained. This model can give a final high-resistance fault type classification decision by combining the voting results of all trees on the features of the new input sample.

[0082] Step S208: Use the trained random forest neural network to detect the fault type of high-resistance fault.

[0083] Specifically, a multidimensional feature vector is obtained from the real-time zero-sequence current signal after undergoing the same denoising and feature extraction process as in the training phase. This feature vector is then input into the trained random forest model. Each decision tree in the forest independently makes a judgment and outputs a classification vote result for the fault type. The model aggregates the vote results of all decision trees, follows the majority voting principle, and outputs the fault type with the highest number of votes as the final diagnostic result (e.g., "asphalt grounding", "tree grounding", "normal", etc.).

[0084] The aforementioned high-resistivity fault detection method significantly enhances the characterization and discrimination capabilities of faults caused by different grounding media (such as asphalt, sand, and tree branches) by integrating multi-dimensional fault feature extraction from the time, frequency, and energy domains. It overcomes the shortcomings of traditional single-feature detection algorithms, which suffer from poor adaptability and high misclassification rates under high-resistivity conditions. By utilizing a random forest neural network to integrate multiple decision trees for collaborative decision-making, it effectively suppresses overfitting and improves the classification accuracy and generalization ability for various high-resistivity fault types and normal states in complex distribution network environments. It achieves a technological leap from binary judgment of "whether it is a fault" to multi-class identification of "what kind of fault," providing maintenance personnel with more instructive fault type information, thereby supporting precise inspection and fault handling and improving the intelligent level of distribution network operation and maintenance. Furthermore, it improves the robustness of the fault detection system to noisy environments and sample imbalances. Through denoising preprocessing of zero-sequence current signals, Bootstrap random sampling, and random feature selection, it ensures the stability and practicality of the method in real-world complex scenarios.

[0085] In one exemplary embodiment, a trained random forest neural network is used to detect the fault type of a high-resistance fault, including:

[0086] The multidimensional fault features detected in real time are input into multiple decision trees in a random forest neural network;

[0087] Each decision tree is used to perform fault analysis on the multidimensional fault features detected in real time, and the classification voting results of different fault types are output. The classification voting results reflect the judgment results of the decision tree on the local features of the fault.

[0088] Aggregate the classification voting results of all decision trees and determine the fault type with the highest number of votes as the high-resistance fault type.

[0089] Specifically, the multi-dimensional fault features detected in real time are input into multiple decision trees to ensure that the dimensions, types, and order of the features extracted in real time are completely consistent with the high-resistance fault feature dataset used during model training. This is a prerequisite for the model to work correctly.

[0090] Each decision tree independently analyzes the input features based on unique rules learned during training and formed by random feature selection, generating a preliminary classification opinion (i.e., voting result). This opinion reflects the locality and diversity judgment made by the tree based on a subset of features.

[0091] The Random Forest algorithm uses the majority voting principle to count the votes of all decision trees and takes the fault type with the most votes as the final output. This effectively balances the judgment bias of individual trees, thereby significantly improving the accuracy and robustness of the final classification result.

[0092] In this embodiment, by integrating the local judgments of multiple decision trees and performing collective voting, the shortcomings of single decision tree models, such as easy overfitting and weak generalization ability, are effectively overcome, significantly improving the classification accuracy and reliability of high-resistance fault types in complex distribution network environments. It can synthesize weak information from various fault characteristics to accurately identify high-resistance faults caused by different grounding media (such as trees, asphalt, etc.), providing maintenance personnel with clear fault type diagnostic results.

[0093] like Figure 3 As shown, in an exemplary embodiment, training a random forest neural network includes:

[0094] Step S302: When constructing multiple decision trees in the random forest neural network, a sampling with replacement method is used to randomly select a subset of samples from the high-resistance fault feature dataset and use the subset of samples to train the decision trees in the random forest neural network.

[0095] Step S304: When splitting nodes in each decision tree, a subset of features is randomly selected from all multidimensional fault features, and the optimal splitting rule for the decision tree nodes is determined using the subset of features.

[0096] Specifically, a subset of samples is randomly selected using sampling with replacement to train the decision trees, generating slightly different training datasets for each tree. This aims to improve overall generalization ability and avoid overfitting by increasing model diversity. A subset of features is randomly selected from all multidimensional fault features to determine node splitting rules. When splitting nodes, the optimal splitting rule is forced to be found only from the randomly selected feature subset (rather than all features). This aims to further enhance the differences between trees and ensure that the model can fully learn the association between different feature combinations and fault types.

[0097] In this embodiment, by introducing a dual randomness mechanism of samples and features, the training method effectively improves the model's generalization ability and robustness. Bootstrap sampling enables each decision tree to be trained based on a differentiated subset of data, enhancing model diversity; random feature selection avoids a few strong features dominating the splitting process, prompting the model to explore the correlation between different feature combinations and fault types. The final constructed random forest neural network can more comprehensively learn the complex feature patterns of high-resistance faults, significantly reducing the risk of overfitting and achieving accurate and stable classification of multiple high-resistance fault types.

[0098] In an exemplary embodiment, determining the optimal splitting rule for a decision tree node using a subset of features includes:

[0099] For each feature in a randomly selected feature subset, calculate the information entropy and conditional entropy corresponding to different split points;

[0100] Based on information entropy and conditional entropy, calculate the information gain rate corresponding to each split point;

[0101] From all candidate split points, select the split point with the largest information gain ratio as the optimal split feature and split point for the current node.

[0102] Specifically, information entropy measures the uncertainty or disorder of the current node's data; a higher entropy value indicates a more mixed sample category (e.g., fault type). Conditional entropy measures the uncertainty of the data in each child node after splitting at a certain split point based on a certain feature; a lower value indicates higher purity of the subset after splitting. Information gain ratio is the ratio of information gain to split information. This metric considers not only the improvement in purity after splitting (information gain) but also the number and size of child nodes generated by the split itself (split information), thus effectively avoiding the bias caused by selecting features with more values, making the splitting criteria fairer and more robust. By comparing all candidate split points, the split point that minimizes uncertainty while avoiding the risk of overfitting is selected as the optimal splitting rule for the current node.

[0103] A feature dataset D is constructed by fusing time-series features from normal and high-resistance fault data. The Bootstrap resampling method is then used to extract data with replacement from D. Secondary sample, generate A separate training subset Each subset The probability distribution function is defined as Its information entropy expression is:

[0104] ;

[0105] For feature factors conditional entropy Its information gain rate The calculation is as follows:

[0106] ;

[0107] in, Features The inherent value reflects the distributional complexity of the feature values.

[0108] For each training subset Construct a decision tree by randomly selecting features from all features. Victozae:

[0109] ;

[0110] Calculate the information gain ratio of each feature Select the feature with the highest information gain ratio. Perform the splitting and recursive execution until the stopping condition is met: or number of node samples ,in, The preset minimum information gain threshold, This represents the minimum sample size for a leaf node.

[0111] Will A forest collection is composed of independently generated decision trees. For the sample to be predicted, the final classification result is determined by a majority voting mechanism.

[0112] In this embodiment, by introducing information gain ratio as the node splitting criterion, the shortcomings of traditional information gain preference for multi-valued features are effectively overcome, the objectivity and fairness of splitting selection are improved, and the splitting point with the strongest classification ability and the best anti-overfitting performance can be automatically identified from the random feature subset, ensuring the efficiency and accuracy of each decision tree growth.

[0113] In an exemplary embodiment, multidimensional fault feature extraction is performed based on sample data to obtain a high-resistance fault feature dataset containing multiple fault types, including:

[0114] Time-frequency domain analysis was performed on the sample data to extract frequency domain features containing third and fifth harmonic components. Combined with time-domain statistical features and energy features, multidimensional feature vectors for samples of different fault types were constructed. Among them, the time-domain statistical features include integral, zero-crossing duration, variance, skewness and peak value, and the energy features include energy entropy and Shannon entropy.

[0115] The multidimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct a high-resistance fault feature dataset.

[0116] Specifically, the essence of multidimensional fault feature extraction is to extract quantitative indicators that can characterize the nature of different faults from signals through time-frequency domain analysis and information entropy calculation.

[0117] Frequency domain characteristics: such as the third harmonic component and the fifth harmonic component (used to capture specific harmonic characteristics caused by faults); Time domain statistical characteristics: such as integral (signal energy), zero-crossing duration (waveform change rate), variance (fluctuation amplitude), skewness (waveform asymmetry), peak value (abrupt change intensity); Energy characteristics: such as energy entropy (uniformity of energy distribution) and Shannon entropy (signal complexity).

[0118] A high-resistance fault feature dataset is constructed, and the multi-dimensional feature vector of each sample is associated with its corresponding fault type label (such as "cement grounding" or "tree grounding") to form structured data. The data is randomly divided into a training set (for model training), a validation set (for parameter tuning), and a test set (for final evaluation) according to a preset ratio (such as 7:2:1) to ensure the objectivity and generalization ability of the model evaluation.

[0119] In this embodiment, the core technical challenge of weak and difficult-to-distinguish high-resistance fault features is solved by multi-dimensional fault feature extraction and standardized dataset construction; the original, noisy current signal is transformed into a structured feature vector that can accurately characterize different fault types, and the reliability of model training and evaluation is ensured by dataset partitioning.

[0120] In one exemplary embodiment, training a random forest neural network based on a high-resistance fault feature dataset includes:

[0121] The Hyperband hyperparameter optimization algorithm is used to optimize the hyperparameters of the random forest neural network using the training set and validation set, so as to obtain a random forest neural network with the optimal hyperparameter combination.

[0122] A high-resistance fault detection model is generated by training a random forest neural network with the optimal hyperparameter combination using the training set.

[0123] The high-resistance fault detection model was tested using a test set to verify its ability to classify and identify various high-resistance fault types and normal current states.

[0124] Specifically, the Hyperband hyperparameter optimization algorithm is used for optimization. The Hyperband algorithm is an adaptive resource allocation hyperparameter optimization method that efficiently searches the hyperparameter space by stopping poorly performing parameter combinations early, thereby quickly determining the optimal hyperparameter combination (such as the number of decision trees, maximum depth, etc.) for the random forest within limited computational resources. Using the hyperparameter-optimized random forest structure, the model is trained using the training set data to generate the final deployable high-resistance fault detection model.

[0125] Using test set data that was not used for training and optimization, the model's classification accuracy for various high-resistivity fault types (such as asphalt grounding, tree branch grounding, etc.) and normal states is objectively evaluated to ensure the model's generalization ability.

[0126] In this embodiment, the problems of parameter dependence on empirical tuning and insufficient generalization ability of random forest models are solved by using the Hyperband hyperparameter optimization algorithm and the three-stage dataset verification mechanism. The optimal hyperparameter combination is obtained automatically and quickly, which significantly improves the model convergence efficiency and classification accuracy. The high-resistance fault detection model constructed in the end can accurately distinguish between various high-resistance fault types and normal states, and has strong generalization ability, which meets the requirements of accuracy and reliability of fault identification in complex power distribution network sites.

[0127] The most detailed embodiment of this application is as follows:

[0128] 1. Data Acquisition and Preprocessing:

[0129] refer to Figure 4 (Experimental platform wiring diagram) The experimental platform of this application is built on a 10kV distribution network and includes a power supply, fault carriers (such as simulated devices for trees, asphalt, cement, etc.), experimental loads, CAAP waveform recording and detection devices, and a control terminal. When a high-resistance fault occurs on the line, the zero-sequence current signal is collected by the CAAP waveform recording and detection device and transmitted to the control terminal for storage and processing.

[0130] The acquired raw zero-sequence current signal is preprocessed, including:

[0131] Denoising: Wavelet transform (such as Db4 wavelet basis) is used to reduce the noise of the signal, eliminating electromagnetic interference and measurement noise;

[0132] Normalization: Normalizes the signal amplitude to the range of [0,1] to eliminate the influence of dimensions;

[0133] Sample construction: Data from the four cycles before the fault and the forty cycles after the fault are extracted to form time series sample data that simultaneously includes normal and fault states.

[0134] 2. Multidimensional Fault Feature Extraction and Dataset Construction:

[0135] refer to Figure 5 (Flowchart of high-resistivity fault identification and classification model construction) Multidimensional feature extraction is performed on the preprocessed sample data:

[0136] Frequency domain characteristics: The third and fifth harmonic components are extracted using Fast Fourier Transform (FFT);

[0137] Time-domain statistical characteristics: Calculate the integral, zero-crossing duration, variance, skewness, and peak value of the signal;

[0138] Energy characteristics: Calculate energy entropy and Shannon entropy to characterize signal complexity and energy distribution characteristics.

[0139] The extracted multidimensional feature vectors are associated with fault type labels (such as "tree-based grounding", "asphalt-based grounding", "normal", etc.) to construct a high-resistivity fault feature dataset. This dataset is then randomly divided into training, validation, and test sets in a 7:2:1 ratio.

[0140] 3. Random Forest Model Training and Hyperparameter Optimization:

[0141] refer to Figure 6 (Schematic diagram of model hyperparameter optimization principle) The Hyperband hyperparameter optimization algorithm is used to optimize the random forest model:

[0142] Hyperparameter range settings include n_estimators (number of decision trees), max_depth (maximum tree depth), min_samples_leaf (minimum number of samples per leaf node), and max_features (size of feature subsets).

[0143] Optimization process: Use training and validation sets to train and evaluate different hyperparameter combinations, eliminate poor-performing combinations through an early stopping mechanism, and finally select the hyperparameter combination with the smallest validation set error;

[0144] Model training: Based on the optimal combination of hyperparameters, a random forest model is trained using all training set data to generate a high-resistance fault detection model.

[0145] 4. Model Validation and Fault Classification:

[0146] refer to Figure 7 (Schematic diagram of the random forest algorithm structure), and the performance of the trained model is evaluated using the test set:

[0147] Voting mechanism: The test sample is input into multiple decision trees in the random forest, and each tree outputs a classification result;

[0148] Result aggregation: The majority voting principle is used to select the fault type with the most votes as the final output;

[0149] Performance metrics: Calculate the model's accuracy, recall, F1 score, and other metrics on the test set to verify the model's ability to classify various high-resistance fault types and normal states.

[0150] 5. Practical Applications and Fault Early Warning:

[0151] The trained high-resistivity fault detection model is deployed to the distribution network fault early warning system. Zero-sequence current signals are acquired in real time, preprocessed, and feature extracted before being input into the model to achieve real-time identification and early warning of high-resistivity fault types, providing maintenance personnel with fault type information and repair guidance.

[0152] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0153] Based on the same inventive concept, this application also provides a high-resistance fault detection device for implementing the high-resistance fault detection method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the high-resistance fault detection device provided below can be found in the limitations of the high-resistance fault detection method described above, and will not be repeated here.

[0154] In one exemplary embodiment, such as Figure 8 As shown, a high-resistance fault detection device is provided, comprising:

[0155] The signal acquisition module 802 is used to acquire the zero-sequence current signal when a high-resistance fault occurs in the line;

[0156] The signal processing module 804 is used to denoise the zero-sequence current signal, collect sample data of normal current state and fault current state with time series, and extract multi-dimensional fault features based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0157] Model training module 806 is used to train a random forest neural network based on a high-resistivity fault feature dataset;

[0158] The fault detection module 808 is used to detect the fault type of high-resistance faults using a trained random forest neural network.

[0159] In one embodiment, the fault detection module 808 is specifically used to input the multidimensional fault features detected in real time into multiple decision trees in a random forest neural network; to perform fault analysis on the multidimensional fault features detected in real time using each decision tree, and to output the classification voting results of different fault types, wherein the classification voting results reflect the judgment results of the decision trees on the local features of the fault; to aggregate the classification voting results of all decision trees, and to determine the fault type with the highest number of votes as the fault type of high resistance fault.

[0160] In one embodiment, the model training module 806 is specifically used to, when constructing multiple decision trees in the random forest neural network, adopt a sampling with replacement method to randomly extract a sample subset from the high-resistance fault feature dataset, and use the sample subset to train the decision trees in the random forest neural network; when splitting nodes in each decision tree, a feature subset is randomly selected from all multi-dimensional fault features, and the optimal splitting rule of the decision tree node is determined using the feature subset.

[0161] In one embodiment, the model training module 806 is specifically used to calculate the information entropy and conditional entropy corresponding to different split points for each feature in a randomly selected feature subset; calculate the information gain ratio corresponding to each split point based on the information entropy and conditional entropy; and select the split point with the largest information gain ratio from all candidate split points as the optimal split feature and split point of the current node.

[0162] In one embodiment, the signal processing module 804 is specifically used to perform time-frequency domain analysis on the sample data, extract frequency domain features containing third and fifth harmonic components, and combine time-domain statistical features and energy features to form multi-dimensional feature vectors for samples of different fault types; wherein, the time-domain statistical features include integral, zero-crossing duration, variance, skewness and peak value, and the energy features include energy entropy and Shannon entropy; the multi-dimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct a high-resistivity fault feature dataset.

[0163] In one embodiment, the model training module 806 is specifically used to employ the Hyperband hyperparameter optimization algorithm to optimize the hyperparameters of the random forest neural network using the training set and validation set, thereby obtaining a random forest neural network with the optimal hyperparameter combination; to train the random forest neural network with the optimal hyperparameter combination using the training set to generate a high-resistance fault detection model; and to test the high-resistance fault detection model using the test set to verify the high-resistance fault detection model's ability to classify and identify various high-resistance fault types and normal current states.

[0164] Each module in the aforementioned high-resistance fault detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0165] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores sample data of normal and fault current states with time series. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a high-resistance fault detection method.

[0166] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0167] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0168] When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired;

[0169] The zero-sequence current signal is denoised, sample data of normal current state and fault current state with time series are collected, and multi-dimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0170] A random forest neural network was trained based on a high-resistivity fault feature dataset.

[0171] The trained random forest neural network is used to detect the fault type of high-resistance faults.

[0172] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0173] The multidimensional fault features detected in real time are input into multiple decision trees in a random forest neural network;

[0174] Each decision tree is used to perform fault analysis on the multidimensional fault features detected in real time, and the classification voting results of different fault types are output. The classification voting results reflect the judgment results of the decision tree on the local features of the fault.

[0175] Aggregate the classification voting results of all decision trees and determine the fault type with the highest number of votes as the high-resistance fault type.

[0176] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0177] When constructing multiple decision trees in a random forest neural network, a sampling with replacement method is used to randomly select a subset of samples from the high-resistivity fault feature dataset and use the subset of samples to train the decision trees in the random forest neural network.

[0178] When splitting nodes in each decision tree, a subset of features is randomly selected from all multidimensional fault features, and the optimal splitting rule for the decision tree node is determined using the subset of features.

[0179] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0180] For each feature in a randomly selected feature subset, calculate the information entropy and conditional entropy corresponding to different split points;

[0181] Based on information entropy and conditional entropy, calculate the information gain rate corresponding to each split point;

[0182] From all candidate split points, select the split point with the largest information gain ratio as the optimal split feature and split point for the current node.

[0183] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0184] Time-frequency domain analysis was performed on the sample data to extract frequency domain features containing third and fifth harmonic components. Combined with time-domain statistical features and energy features, multidimensional feature vectors for samples of different fault types were constructed. Among them, the time-domain statistical features include integral, zero-crossing duration, variance, skewness and peak value, and the energy features include energy entropy and Shannon entropy.

[0185] The multidimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct a high-resistance fault feature dataset.

[0186] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0187] The Hyperband hyperparameter optimization algorithm is used to optimize the hyperparameters of the random forest neural network using the training set and validation set, so as to obtain a random forest neural network with the optimal hyperparameter combination.

[0188] A high-resistance fault detection model is generated by training a random forest neural network with the optimal hyperparameter combination using the training set.

[0189] The high-resistance fault detection model was tested using a test set to verify its ability to classify and identify various high-resistance fault types and normal current states.

[0190] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0191] When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired;

[0192] The zero-sequence current signal is denoised, sample data of normal current state and fault current state with time series are collected, and multi-dimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0193] A random forest neural network was trained based on a high-resistivity fault feature dataset.

[0194] The trained random forest neural network is used to detect the fault type of high-resistance faults.

[0195] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0196] The multidimensional fault features detected in real time are input into multiple decision trees in a random forest neural network;

[0197] Each decision tree is used to perform fault analysis on the multidimensional fault features detected in real time, and the classification voting results of different fault types are output. The classification voting results reflect the judgment results of the decision tree on the local features of the fault.

[0198] Aggregate the classification voting results of all decision trees and determine the fault type with the highest number of votes as the high-resistance fault type.

[0199] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0200] When constructing multiple decision trees in a random forest neural network, a sampling with replacement method is used to randomly select a subset of samples from the high-resistivity fault feature dataset and use the subset of samples to train the decision trees in the random forest neural network.

[0201] When splitting nodes in each decision tree, a subset of features is randomly selected from all multidimensional fault features, and the optimal splitting rule for the decision tree node is determined using the subset of features.

[0202] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0203] For each feature in a randomly selected feature subset, calculate the information entropy and conditional entropy corresponding to different split points;

[0204] Based on information entropy and conditional entropy, calculate the information gain rate corresponding to each split point;

[0205] From all candidate split points, select the split point with the largest information gain ratio as the optimal split feature and split point for the current node.

[0206] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0207] Time-frequency domain analysis was performed on the sample data to extract frequency domain features containing third and fifth harmonic components. Combined with time-domain statistical features and energy features, multidimensional feature vectors for samples of different fault types were constructed. Among them, the time-domain statistical features include integral, zero-crossing duration, variance, skewness and peak value, and the energy features include energy entropy and Shannon entropy.

[0208] The multidimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct a high-resistance fault feature dataset.

[0209] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0210] The Hyperband hyperparameter optimization algorithm is used to optimize the hyperparameters of the random forest neural network using the training set and validation set, so as to obtain a random forest neural network with the optimal hyperparameter combination.

[0211] A high-resistance fault detection model is generated by training a random forest neural network with the optimal hyperparameter combination using the training set.

[0212] The high-resistance fault detection model was tested using a test set to verify its ability to classify and identify various high-resistance fault types and normal current states.

[0213] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0214] When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired;

[0215] The zero-sequence current signal is denoised, sample data of normal current state and fault current state with time series are collected, and multi-dimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types.

[0216] A random forest neural network was trained based on a high-resistivity fault feature dataset.

[0217] The trained random forest neural network is used to detect the fault type of high-resistance faults.

[0218] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0219] The multidimensional fault features detected in real time are input into multiple decision trees in a random forest neural network;

[0220] Each decision tree is used to perform fault analysis on the multidimensional fault features detected in real time, and the classification voting results of different fault types are output. The classification voting results reflect the judgment results of the decision tree on the local features of the fault.

[0221] Aggregate the classification voting results of all decision trees and determine the fault type with the highest number of votes as the high-resistance fault type.

[0222] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0223] When constructing multiple decision trees in a random forest neural network, a sampling with replacement method is used to randomly select a subset of samples from the high-resistivity fault feature dataset and use the subset of samples to train the decision trees in the random forest neural network.

[0224] When splitting nodes in each decision tree, a subset of features is randomly selected from all multidimensional fault features, and the optimal splitting rule for the decision tree node is determined using the subset of features.

[0225] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0226] For each feature in a randomly selected feature subset, calculate the information entropy and conditional entropy corresponding to different split points;

[0227] Based on information entropy and conditional entropy, calculate the information gain rate corresponding to each split point;

[0228] From all candidate split points, select the split point with the largest information gain ratio as the optimal split feature and split point for the current node.

[0229] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0230] Time-frequency domain analysis was performed on the sample data to extract frequency domain features containing third and fifth harmonic components. Combined with time-domain statistical features and energy features, multidimensional feature vectors for samples of different fault types were constructed. Among them, the time-domain statistical features include integral, zero-crossing duration, variance, skewness and peak value, and the energy features include energy entropy and Shannon entropy.

[0231] The multidimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct a high-resistance fault feature dataset.

[0232] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0233] The Hyperband hyperparameter optimization algorithm is used to optimize the hyperparameters of the random forest neural network using the training set and validation set, so as to obtain a random forest neural network with the optimal hyperparameter combination.

[0234] A high-resistance fault detection model is generated by training a random forest neural network with the optimal hyperparameter combination using the training set.

[0235] The high-resistance fault detection model was tested using a test set to verify its ability to classify and identify various high-resistance fault types and normal current states.

[0236] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0237] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0238] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0239] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A high-resistance fault detection method, characterized in that, The method includes: When a high-resistance fault occurs in the line, the zero-sequence current signal is acquired; The zero-sequence current signal is denoised, and sample data of normal current state and fault current state with time series are collected. Multidimensional fault features are extracted based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types. A random forest neural network is trained based on the high-resistivity fault feature dataset. The trained random forest neural network is used to detect the fault type of high-resistance faults.

2. The method according to claim 1, characterized in that, The method of detecting high-resistance fault types using a trained random forest neural network includes: The multidimensional fault features detected in real time are input into multiple decision trees in the random forest neural network; Each decision tree is used to perform fault analysis on the multidimensional fault features detected in real time, and the classification voting results of different fault types are output. The classification voting results reflect the judgment results of the decision tree on the local features of the fault. Aggregate the classification voting results of all decision trees and determine the fault type with the highest number of votes as the high-resistance fault type.

3. The method according to claim 1, characterized in that, The training of the random forest neural network includes: When constructing multiple decision trees in a random forest neural network, a sampling with replacement method is used to randomly select a subset of samples from the high-resistivity fault feature dataset, and the decision trees in the random forest neural network are trained using the subset of samples. When splitting nodes in each decision tree, a subset of features is randomly selected from all multidimensional fault features, and the optimal splitting rule for the decision tree node is determined using the subset of features.

4. The method according to claim 3, characterized in that, The step of determining the optimal splitting rule for decision tree nodes using the feature subset includes: For each feature in the randomly selected feature subset, calculate the information entropy and conditional entropy corresponding to different split points; Based on the information entropy and the conditional entropy, calculate the information gain rate corresponding to each split point; From all candidate split points, select the split point with the largest information gain ratio as the optimal split feature and split point for the current node.

5. The method according to claim 1, characterized in that, The step involves extracting multidimensional fault features from the sample data to obtain a high-resistance fault feature dataset containing multiple fault types, including: The sample data is subjected to time-frequency domain analysis to extract frequency domain features containing third and fifth harmonic components. These features are then combined with time-domain statistical features and energy features to construct multidimensional feature vectors for samples of different fault types. The time-domain statistical features include integral, zero-crossing duration, variance, skewness, and peak value. The energy features include energy entropy and Shannon entropy. The multidimensional feature vectors are randomly divided into training set, validation set and test set according to a preset ratio to construct the high-resistance fault feature dataset.

6. The method according to claim 5, characterized in that, The step of training a random forest neural network based on the high-resistance fault feature dataset includes: The Hyperband hyperparameter optimization algorithm is used to optimize the hyperparameters of the random forest neural network using the training set and the validation set, so as to obtain a random forest neural network with the optimal combination of hyperparameters. The training set is used to train the random forest neural network with the optimal hyperparameter combination to generate a high-resistance fault detection model. The high-resistance fault detection model is tested using the test set to verify its ability to classify and identify various high-resistance fault types and normal current states.

7. A high-resistance fault detection device, characterized in that, The device includes: The signal acquisition module is used to acquire the zero-sequence current signal when a high-resistance fault occurs in the line. The signal processing module is used to denoise the zero-sequence current signal, collect sample data of normal current state and fault current state with time series, and extract multi-dimensional fault features based on the sample data to obtain a high-resistivity fault feature dataset containing multiple fault types. The model training module is used to train a random forest neural network based on the high-resistivity fault feature dataset. The fault detection module is used to detect the fault type of high-resistance faults using a trained random forest neural network.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images