A fault section locating and credibility checking method for distribution network local security
Patent Information
- Application Number
- CN202611300533.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-09-25
AI Technical Summary
其中，矩阵法依赖馈线终端单元上报的故障过流信息及配电网拓扑关系，其定位效果受监测终端覆盖范围影响；状态估计法则依赖较准确的网络拓扑、线路阻抗参数和量测信息，在线路参数实际运维中难以精确获知的场景下实用性受限
本发明通过神经网络进行初步故障区段筛选，再利用电压变化量空间分布规律对候选区段进行物理校验，既利用了深度学习模型的拟合能力，又通过物理规律对模型输出进行可信度验证，克服了纯数据驱动模型缺乏物理约束的黑箱问题，提升了定位结果的透明性和可信度。区别于依赖精确线路阻抗参数的状态估计法，本发明仅利用配电变压器低压侧电压监测数据即可完成物理校验，无需构建节点导纳矩阵，工程实施难度低，对实际配电网中参数缺失或不准场景具有良好的适应性。实验结果表明，本发明方法整体定位准确率达98.7%，显著优于纯数据驱动模型（78.8%）和状态估计法（93.8%），在不同接地故障电阻工况下均能保持较高的定位性能。
Smart Images
Figure CN122815092A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power distribution network safety and fault analysis technology, and more specifically to a method for fault segment location and reliability verification for local power distribution network safety. Background Technology
[0002] The security of the distribution network in real-time operation, i.e., the local security of the distribution network, directly affects the reliability of power supply. In recent years, under the influence of factors such as the massive decentralized resource access and extreme weather, the potential for distribution network faults has intensified, bringing about local security risks to the distribution network. Quickly and accurately locating the faulty section is not only a prerequisite for fault isolation and power restoration, but also a crucial step in identifying local security risks in the distribution network.
[0003] Existing methods for locating fault sections in distribution networks can be mainly divided into two categories. One category is based on physical models, primarily including matrix methods and state estimation methods. Matrix methods rely on fault overcurrent information reported by feeder terminal units and the distribution network topology; their location accuracy is affected by the coverage area of monitoring terminals. State estimation methods rely on relatively accurate network topology, line impedance parameters, and measurement information, but their practicality is limited in scenarios where line parameters are difficult to obtain precisely in actual operation and maintenance. The other category is based on data-driven methods, using machine learning or deep learning models to learn the mapping relationship between measurement information and fault location from historical fault data. However, this type of method lacks physical constraints, resulting in "black box" characteristics in the location results, insufficient reliability, and difficulty in timely and accurately identifying distribution network safety risks and weaknesses.
[0004] Therefore, there is an urgent need in this field for a fault location scheme that does not rely on precise line impedance parameters and can effectively verify the physical reliability of data-driven location results. Summary of the Invention
[0005] In view of the above problems, the present invention is proposed to provide a fault section location and reliability verification method for local security of distribution networks that overcomes or at least partially solves the above problems.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] S1. Acquire electrical measurement signals from multiple monitoring nodes in the distribution network, as well as voltage measurement signals from the low-voltage side of each distribution transformer; S2. Construct a training sample set based on the electrical measurement signals and their corresponding fault section labels, and train a classification model; input the electrical measurement signals of the fault samples to be tested into the trained classification model, output the fault probability of each section of the distribution network, and select the K sections with the highest fault probability as candidate fault sections, where K is a preset positive integer; S3. Based on the distribution network topology, determine the downstream monitoring nodes of each candidate section; calculate the voltage change of each monitoring node before and after the fault based on the voltage measurement signal on the low-voltage side of the distribution transformer; establish physical verification criteria, which include: The first criterion is that the coefficient of variation of voltage change at each monitoring node downstream of the candidate section is less than a preset threshold. The second criterion is to take each candidate segment as a hypothetical fault point and check whether the monitoring node with the largest voltage change amplitude in the entire network is located within the downstream topology range of the hypothetical fault point. S4. Perform the verification of the first criterion and the second criterion on the candidate fault segments respectively; among the candidate fault segments that simultaneously satisfy the first criterion and the second criterion, select the candidate fault segment with the highest output probability of the classification model and determine it as the final fault segment location result.
[0008] Preferably, the electrical measurement signals include three-phase voltage signals and three-phase current signals of each monitoring node; the electrical measurement signals are collected by feeder terminal units located at the feeder of the distribution network.
[0009] Preferably, in step S2, the training classification model specifically includes: Discrete wavelet transform is performed on the voltage and current signals of each phase in the electrical measurement signals to extract the detail coefficients and approximation coefficients of each layer; For each phase voltage or current signal, its discrete sampling sequence is recorded as the original signal. ,go through After layer wavelet decomposition, the first Layer approximation coefficient With detail coefficient Calculated using Mallat's recursive formula: (1) in, , These are the low-pass and high-pass filter coefficients corresponding to the selected wavelet basis; index Implicit downsampling operation; Original signal ; Decomposition layer number Based on sampling frequency Power supply base frequency Number of valid sampling points during the fault period and wavelet filter length To determine common constraints, the following conditions must be met simultaneously:
[0010] If there exists a positive integer that satisfies both constraints... Then the maximum value among them is selected as the final decomposition level. If there is no positive integer that simultaneously satisfies both constraints. Then select the one that satisfies The largest positive integer As the final decomposition level; Calculate the standard deviation, skewness, and kurtosis of the detail coefficients and approximation coefficients of each layer as time-frequency domain statistical features; for any layer coefficient sequence Its standard deviation skewness and kurtosis Calculate according to the following formulas respectively:
[0011] in, This is the length of the coefficient for this layer; The coefficient mean is used; the calculated statistics are combined to construct a feature vector; Using the feature vector as input and the fault segment label as output label, a fully connected neural network is trained.
[0012] Preferably, a positive integer is preset. K The value is determined by the recall rate on the validation set: Calculate the number of different candidates corresponding Top-K Recall rate :
[0013]
[0014] in, The total number of samples in the validation set; For the first The neural network outputs a probability vector for each sample; To take the probability before Large segment index set functions; Labels for actual faulty sections; It is an indicator function; This is the recall rate threshold; This is the upper limit for the search.
[0015] Preferably, in step S3, calculating the voltage changes of each monitoring node before and after the fault specifically includes: For the n In the nth sample j Each monitoring node extracts the absolute value of the change in the three-phase voltage amplitude before and after the fault. , , The voltage change at the monitoring node is determined by taking the largest change among the three phases.
[0016] For candidate segments Extract the voltage changes of all monitoring nodes downstream of this section and calculate the average downstream voltage change:
[0017] Calculate the coefficient of variation:
[0018] in Candidate segment The collection of all downstream monitoring nodes The number of elements in the set. A very small positive number set to prevent the denominator from being zero.
[0019] Preferably, the candidate fault segments of the fault sample to be tested are respectively verified by the first criterion and the second criterion, and the candidate fault segment with the highest output probability of the classification model is selected from the candidate fault segments that simultaneously satisfy both criteria, specifically including: For candidate segments Define the first verification result Second verification result :
[0020]
[0021] in:
[0022] It is the monitoring node with the largest voltage change amplitude in the entire network. It is a collection of monitoring nodes across the entire network. Candidate segment The end node, Candidate segment The set of all downstream monitoring nodes; Final Failure Section Select according to the following formula:
[0023] in For the candidate segment set, For neural networks to segment The output probability.
[0024] Preferably, in step S4, if there is no segment in the candidate fault segment of the current fault sample that simultaneously satisfies the first criterion and the second criterion, the location result is deemed to be unreliable and is transferred to manual review.
[0025] Preferably, the fully connected neural network contains multiple hidden layers, and the output layer uses the Softmax function to normalize the output values of each segment to obtain the fault probability vector of each segment.
[0026] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following: This invention uses a neural network for initial fault segment screening, and then utilizes the spatial distribution law of voltage changes to perform physical verification of candidate segments. It leverages the fitting ability of deep learning models and verifies the reliability of model outputs through physical laws, overcoming the black-box problem of purely data-driven models lacking physical constraints, thus improving the transparency and reliability of the location results. Unlike state estimation methods that rely on precise line impedance parameters, this invention only uses voltage monitoring data from the low-voltage side of the distribution transformer to complete physical verification, eliminating the need to construct node admittance matrices. This reduces engineering implementation difficulty and demonstrates good adaptability to scenarios with missing or inaccurate parameters in actual distribution networks. Experimental results show that the overall location accuracy of this method reaches 98.7%, significantly better than purely data-driven models (78.8%) and state estimation methods (93.8%), and maintains high location performance under different ground fault resistance conditions. Attached Figure Description
[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0028] Figure 1 This is a flowchart of a method for fault section location and reliability verification for local security of distribution networks provided in an embodiment of the present invention; Figure 2 This is a comparison chart of the fault section location accuracy under different grounding fault resistances provided in the embodiments of the present invention. Detailed Implementation
[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0030] This invention discloses a method for fault section location and reliability verification for local security in distribution networks, comprising the following steps: S1. Acquire electrical measurement signals from multiple monitoring nodes in the distribution network, as well as voltage measurement signals from the low-voltage side of each distribution transformer; S2. Construct a training sample set based on electrical measurement signals and their corresponding fault section labels, and train a classification model; input the electrical measurement signals of the fault samples to be tested into the trained classification model, output the fault probability of each section of the distribution network, and select the K sections with the highest fault probability as candidate fault sections, where K is a preset positive integer; S3. Based on the distribution network topology, determine the downstream monitoring nodes of each candidate section; calculate the voltage change of each monitoring node before and after the fault based on the voltage measurement signal on the low-voltage side of the distribution transformer; establish physical verification criteria, which include: The first criterion is that the coefficient of variation of voltage change at each monitoring node downstream of the candidate section is less than a preset threshold. The second criterion is to take each candidate segment as a hypothetical fault point and check whether the monitoring node with the largest voltage change amplitude in the entire network is located within the downstream topology range of the hypothetical fault point. S4. Perform the verification of the first criterion and the second criterion on the candidate fault segments respectively; among the candidate fault segments that simultaneously satisfy the first criterion and the second criterion, select the candidate fault segment with the highest output probability of the classification model and determine it as the final fault segment location result.
[0031] The following is in conjunction with the appendix Figure 1 Steps S1-S4 of the present invention will be described in detail respectively: S1: Data Acquisition and Sample Construction This embodiment constructs a simulation model based on the IEEE 33-node standard distribution network topology to simulate various fault scenarios in distribution network operation. Feeder terminal units (FTUs) are configured at nodes 1, 2, 3, 6, 18, 22, 25, and 33, with a sampling frequency set to 1kHz. Voltage monitoring devices on the low-voltage side of distribution transformers are configured at all load nodes, also with a sampling frequency set to 1kHz. The feeder terminal units are used to collect the three-phase voltage and three-phase current signals of each monitoring node, while the voltage monitoring devices on the low-voltage side of the distribution transformers are used to collect the three-phase voltage signals on the low-voltage side of each distribution transformer.
[0032] This embodiment uses a single-phase ground fault as an example. The fault locations are set at 30%, 50%, and 70% of the line length, with ground fault resistances set to 0.01Ω, 2Ω, 10Ω, 20Ω, 50Ω, 100Ω, 500Ω, and 1000Ω, respectively, generating a total of 768 fault samples. For each fault scenario, the three-phase voltage and current signals of each FTU monitoring node and the three-phase voltage signals of the low-voltage side of each distribution transformer are simultaneously collected to construct the original dataset.
[0033] The original dataset was divided into training, validation, and test sets in a 6:1:3 ratio. Each sample was labeled with its fault section number, fault location, and ground fault resistance value. The training set was used for neural network parameter learning, the validation set was used for physical verification hyperparameter optimization, and the test set was used for final localization performance evaluation.
[0034] Step S2: Model Training and Candidate Segment Selection Step S2.1: Multi-scale time-frequency decomposition of FTU signal based on DWT The three-phase voltage and three-phase current signals of each FTU monitoring node are decomposed by Discrete Wavelet Transform (DWT) to convert the original time-domain signals into multi-scale time-frequency feature representations.
[0035] This embodiment uses the db4 wavelet basis to decompose the signal. For a certain phase voltage or current signal, its discrete sampling sequence is recorded as the original signal. ,go through After layer wavelet decomposition, the first Layer approximation coefficient With detail coefficient The calculation is performed using Mallat's recursive formula as follows:
[0036] In the formula , These are the low-pass and high-pass filter coefficients corresponding to the selected wavelet basis; index Implicit downsampling operation; Original signal .
[0037] Decomposition layer number Based on sampling frequency Power supply base frequency Number of valid sampling points during the fault period and wavelet filter length To determine common constraints, the following conditions must be met simultaneously:
[0038] Within the range of feasible positive integers that satisfy the above constraints, the largest feasible value is selected as the final decomposition level. If there is no positive integer that simultaneously satisfies both constraints. At the same time, priority should be given to ensuring signal length constraints. That is, select the one that satisfies The largest positive integer The final number of decomposition layers is used to ensure the reliability of the decomposition coefficients. In this embodiment, the sampling frequency is 1kHz and the power supply base frequency is 50Hz. The number of decomposition layers is determined through calculation. =3. After decomposition, we get... Group detail coefficient and a set of coarsest-scale approximation coefficients .
[0039] Step S2.2: Constructing the fault feature vector For each layer of coefficient sequence obtained in S2.1 (including all) and ), calculate the standard deviation respectively skewness With kurtosis As a fault characteristic of this layer, This indicates the first layer. There are several coefficient values. The calculation formula is shown below:
[0040] in, This is the length of the coefficient for this layer; This represents the mean of the coefficients.
[0041] For each phase of each electrical variable, its eigenvector is composed of Layer detail factor It consists of three statistics: a statistical measure and an approximation coefficient. Dimension. The eigenvectors of each phase of all electrical variables are concatenated to form the final eigenvector. ,in, , This represents the number of variables to be monitored.
[0042] Step S2.3: Constructing and training a fully connected neural network model Feature vectors constructed using S2.2 Using the input, construct a fully connected neural network.
[0043] In one embodiment, the network comprises three hidden layers with 512, 256, and 128 neurons, respectively. Each layer is followed by a batch normalization layer, a ReLU activation layer, and a Dropout layer (with the dropout rate decreasing from 0.4 to 0.2). The output layer uses the Softmax function to normalize the output values of each segment, obtaining the fault probability vector for each segment. ,in This represents the total number of distribution network sections.
[0044] Assume the network includes Hidden layer, first The layer output is:
[0045] in , This is the weight matrix. For bias vectors, The activation function is used. The Adam optimizer is employed for training, with an initial learning rate of 5 × 10⁻⁶. -4 The error rate decays every 50 rounds. A cross-entropy loss function is used, with a maximum training epoch of 200 rounds, and early stopping is employed to prevent overfitting. After training, test set samples are input into the model, which outputs the fault probability of each segment for subsequent Top-K candidate segment extraction and localization performance evaluation.
[0046] Step S3: Establish physical verification criteria Step S3.1: Optimize the number of candidate segments K Calculate the number of different candidates on the validation set. Corresponding Top-K recall Select the smallest value that meets the recall threshold. As the optimal number of candidate segments The calculation formula is as follows:
[0047]
[0048] In the formula The total number of samples in the validation set; For the first The neural network outputs a probability vector for each sample; To take the probability before Large segment index set functions; Labels for actual faulty sections; It is an indicator function; This is the recall rate threshold; This is the upper limit for the search. In this embodiment, it is set to... After optimization of the validation set, the recall threshold requirement is met when K = 5, so the optimal number of candidate segments is determined to be K = 5.
[0049] Step S3.2: Calculate the coefficient of variation of downstream voltage change and determine the threshold. For the In the nth sample For each transformer substation's low-voltage side voltage monitoring node, the absolute value of the change in three-phase voltage amplitude before and after the fault is extracted and denoted as... , , The voltage change at the monitoring node is determined by the largest change among the three phases.
[0050] For each sample in the validation set and its corresponding candidate region Extract the voltage changes of all monitoring nodes downstream of this section and calculate the average downstream voltage change:
[0051] Calculate its coefficient of variation This characterizes the dispersion of downstream voltage fluctuations.
[0052] In the formula, A very small positive number is set to prevent the denominator from being zero. Candidate segment The set of all downstream monitoring nodes; The number of elements in the set. The smaller the value, the more consistent the voltage changes at the downstream monitoring nodes of the section, which is consistent with the characteristic that voltage changes are concentrated downstream of the fault section after a fault.
[0053] Based on the distribution of the coefficient of variation of downstream voltage changes in all samples of the real fault section on the validation set, the threshold is determined using the quantile method. :
[0054] In the formula For sets Quantile function; This is the confidence level parameter. In this embodiment, the coefficients of variation downstream of all real fault sections in the validation set are arranged in ascending order, and the threshold is determined using the 95th percentile method. =0.028.
[0055] Step S4: Joint Decision and Output Step S4.1: Extract Top-K candidate segments For each sample in the test set Output probability vector based on fully connected neural network Select the first ones in descending order of probability value. K Each segment constitutes a candidate set. In this embodiment K=5 .
[0056] Step S4.2: Verify the downstream voltage variation coefficient For candidate set Each segment The coefficient of variation of the measured downstream voltage change is calculated according to the aforementioned formula. And determine whether the physical consistency condition is met:
[0057] At that time, it was indicated that the downstream section of the segment exhibited a concentrated distribution of voltage changes, which was verified by the first criterion.
[0058] Step S4.3: Verify the location of peak nodes across the entire network Identify test samples The node with the largest voltage change amplitude across the entire network:
[0059] Determine whether it is located in the candidate segment The end or downstream:
[0060] in, It is a collection of monitoring nodes across the entire network; Candidate segment The terminal node; For section The set of all downstream monitoring nodes. When the peak node of the entire network is located in the segment. At the end or downstream, This indicates that the location of the most severe voltage drop is consistent with the fault path topology of that section, which is verified by the second criterion.
[0061] Step S4.4: Joint Determination of Faulty Sections Based on the above verification results, the candidate set... The various sections in the process undergo joint judgment, and the final faulty section is output. :
[0062] in, For neural networks to segment The output probability; and These are the physical verification results.
[0063] If at least one segment in the candidate set satisfies both physical checks simultaneously, the final faulty segment is selected according to the above formula, and the location result is output as the basis for accurate identification of local security risks and vulnerabilities in the distribution network, supporting subsequent security assurance decisions; otherwise, the location result is deemed insufficiently reliable and is transferred to manual review. To verify the effectiveness of the method of the present invention, this embodiment conducted a fault segment location experiment on a test set and compared it with existing methods.
[0064] Accuracy of fault section location under different grounding fault resistances, such as Figure 2 As shown in the figure. Based on the comprehensive simulation results, the method of the present invention has excellent location performance for low-resistance and medium-resistance faults. Even under the condition that the voltage change is greatly reduced due to high-resistance faults (greater than 100Ω), it can still maintain a high location accuracy, which verifies the robustness of the proposed method for local safety identification of distribution networks under different fault conditions.
[0065] The following table compares the accuracy of fault location using different methods:
[0066] The results show that, compared with pure data-driven models and state estimation methods that rely on precise parameters, the method proposed in this invention is more adaptable to scenarios where actual distribution network parameters are incomplete, effectively improving risk identification capabilities and providing a reliable basis for local security assurance.
[0067] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0068] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for fault section location and reliability verification for local security in distribution networks, characterized in that, Includes the following steps: S1. Acquire electrical measurement signals from multiple monitoring nodes in the distribution network, as well as voltage measurement signals from the low-voltage side of each distribution transformer; S2. Construct a training sample set based on the electrical measurement signals and their corresponding fault section labels, and train a classification model; input the electrical measurement signals of the fault samples to be tested into the trained classification model, output the fault probability of each section of the distribution network, and select the K sections with the highest fault probability as candidate fault sections, where K is a preset positive integer; S3. Based on the distribution network topology, determine the downstream monitoring nodes for each candidate section; Based on the voltage measurement signal on the low-voltage side of the distribution transformer, calculate the voltage change of each monitoring node before and after the fault. Establish physical verification criteria, which include: The first criterion is that the coefficient of variation of voltage change at each monitoring node downstream of the candidate section is less than a preset threshold. The second criterion is to take each candidate segment as a hypothetical fault point and check whether the monitoring node with the largest voltage change amplitude in the entire network is located within the downstream topology range of the hypothetical fault point. S4. Perform the verification of the first criterion and the second criterion on the candidate fault segments respectively; among the candidate fault segments that simultaneously satisfy the first criterion and the second criterion, select the candidate fault segment with the highest output probability of the classification model and determine it as the final fault segment location result.
2. The method according to claim 1, characterized in that, The electrical measurement signals include the three-phase voltage signals and three-phase current signals of each monitoring node; the electrical measurement signals are collected by the feeder terminal unit located at the feeder of the distribution network.
3. The method according to claim 1, characterized in that, In step S2, the training of the classification model specifically includes: Discrete wavelet transform is performed on the voltage and current signals of each phase in the electrical measurement signals to extract the detail coefficients and approximation coefficients of each layer; For each phase voltage or current signal, its discrete sampling sequence is recorded as the original signal. ,go through After layer wavelet decomposition, the first Layer approximation coefficient With detail coefficient Calculated using Mallat's recursive formula: （1） in , These are the low-pass and high-pass filter coefficients corresponding to the selected wavelet basis; index Implicit downsampling operation; Original signal ; Decomposition layer number Based on sampling frequency Power supply base frequency Number of valid sampling points during the fault period and wavelet filter length To determine common constraints, the following conditions must be met simultaneously: If there exists a positive integer that satisfies both constraints... Then the maximum value among them is selected as the final decomposition level. If there is no positive integer that simultaneously satisfies both constraints. Then select the one that satisfies The largest positive integer As the final decomposition level; Calculate the standard deviation, skewness, and kurtosis of the detail coefficients and approximation coefficients of each layer as time-frequency domain statistical features; for any layer coefficient sequence Its standard deviation skewness and kurtosis Calculate according to the following formulas respectively: in, This is the length of the coefficient for this layer; The coefficient mean is used; the calculated statistics are combined to construct a feature vector; Using the feature vector as input and the fault segment label as output label, a fully connected neural network is trained.
4. The method according to claim 1, characterized in that, Preset positive integer K The value is determined by the recall rate on the validation set: Calculate the number of different candidates corresponding Top-K Recall rate : in, The total number of samples in the validation set; For the first The neural network outputs a probability vector for each sample; To take the probability before Large segment index set functions; Labels for actual faulty sections; It is an indicator function; This is the recall rate threshold; This is the upper limit for the search.
5. The method according to claim 1, characterized in that, In step S3, calculating the voltage changes of each monitoring node before and after the fault specifically includes: For the n In the nth sample j Each monitoring node extracts the absolute value of the change in the three-phase voltage amplitude before and after the fault. , , The voltage change at the monitoring node is determined by taking the largest change among the three phases. For candidate segments Extract the voltage changes of all monitoring nodes downstream of this section and calculate the average downstream voltage change: Calculate the coefficient of variation: in Candidate segment The collection of all downstream monitoring nodes The number of elements in the set. A very small positive number set to prevent the denominator from being zero.
6. The method according to claim 1, characterized in that, The process of verifying the candidate fault segments of the fault sample under test using the first criterion and the second criterion, and selecting the candidate fault segment with the highest output probability from the candidate fault segments that simultaneously satisfy both criteria, specifically includes: For candidate segments Define the first verification result Second verification result : in: It is the monitoring node with the largest voltage change amplitude in the entire network. It is a collection of monitoring nodes across the entire network. Candidate segment The end node, Candidate segment The set of all downstream monitoring nodes; Final Failure Section Select according to the following formula: in For the candidate segment set, For neural networks to segment The output probability.
7. The method according to claim 1, characterized in that, In step S4, if there is no segment in the candidate fault segment of the current fault sample that simultaneously satisfies the first criterion and the second criterion, the location result is deemed to be unreliable and is transferred to manual review.
8. The method according to claim 3, characterized in that, The fully connected neural network contains multiple hidden layers, and the output layer uses the Softmax function to normalize the output values of each segment to obtain the fault probability vector of each segment.