A multi-branch active power distribution network fault locating method and system

Abstract

The present application belongs to the technical field of power distribution network fault positioning, and particularly relates to a multi-branch active power distribution network fault positioning method and system, comprising: obtaining a multi-dimensional fault feature vector of a multi-branch active power distribution network; constructing a fault feature space based on the obtained multi-dimensional fault feature vector; performing multi-task fault classification of the multi-branch active power distribution network according to the constructed fault feature space and a pre-set fault classification model to obtain a fault joint classification result; and performing regression prediction of a fault position according to the obtained fault joint classification result, the fault feature space and a regression model to complete fault positioning of the multi-branch active power distribution network.

Description

Technical Field

[0001] This invention belongs to the field of distribution network fault location technology, specifically relating to a fault location method and system for multi-branch active distribution networks. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of new power systems, the topology of distribution networks is becoming increasingly complex, exhibiting tree-like network characteristics with multi-level branches. With the integration of numerous distributed generation (DG) sources, distribution networks have evolved from single-source radial systems to multi-source systems. Fault information exhibits complex spatiotemporal phenomena such as bidirectional flow and multi-path superposition, leading to complex correlations of fault characteristics and increasing the difficulty of fault feature characterization and location accuracy. Given a limited number of measuring devices, conducting research on active distribution network fault location from the perspective of spatiotemporal feature characterization and correlation modeling is of great significance for improving location accuracy under complex operating conditions.

[0004] When a fault occurs in a distribution network, the fault characteristics typically exhibit non-stationary features such as abrupt changes, damped oscillations, and multi-scale transients. Currently, fault signal feature extraction mainly focuses on three aspects: time-domain features, frequency-domain features, and time-frequency-domain features. While time-domain features have clear physical meaning, their ability to characterize the spectral changes of non-stationary fault signals is limited. Existing methods can extract the time-domain, frequency-domain, time-frequency, and spatial features of fault signals, but they do not utilize the topological structure, making it difficult to fully characterize the spatiotemporal coupling characteristics of fault signals in multi-branch active distribution networks.

[0005] Traditional methods for fault location in distribution networks include impedance methods, traveling wave methods, and matrix methods. While impedance methods offer clear modeling, they are highly dependent on line parameters and load characteristics. Traveling wave methods, despite their high accuracy and fast response, suffer from distortion and reflection at multiple power source nodes after distributed generation integration, leading to decreased fault location performance. Matrix methods can explicitly incorporate distribution network topology information, but in multi-source distribution networks, the need to process high-dimensional information matrices makes them unsuitable for rapid location. Existing fault location methods exhibit good performance under certain conditions, but in complex active distribution network topologies and with multiple power sources, they struggle to balance accuracy, efficiency, and engineering applicability. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a fault location method and system for multi-branch active distribution networks. Based on a time-frequency-topology fusion Temporal Graph Convolutional Network (TGCN) – iFlowformer – fault location in multi-branch active distribution networks is achieved. This method extracts feature information containing fault sections and locations under complex operating conditions, effectively overcoming the limited representational ability of single features under complex conditions. Furthermore, the introduction of iFlowformer to learn long-sequence temporal relationships fully mines the discriminative information contained in multi-dimensional fault features, enabling joint classification of fault branches and types, as well as precise fault location, thus improving fault location accuracy under complex operating conditions.

[0007] According to some embodiments, the first solution of the present invention provides a fault location method for a multi-branch active distribution network, which adopts the following technical solution: A fault location method for a multi-branch active distribution network includes: Obtain multidimensional fault feature vectors for multi-branch active distribution networks; Based on the acquired multidimensional fault feature vectors, a fault feature space is constructed. Based on the constructed fault feature space and the preset fault classification model, multi-task fault classification is performed on the multi-branch active distribution network to obtain the joint fault classification result. Based on the obtained joint fault classification results, fault feature space and regression model, the fault location is predicted by regression, and the fault location of the multi-branch active distribution network is completed.

[0008] As a further technical limitation, the preset fault classification model adopts the iFlowformer classification model based on multi-task learning temporal graph convolutional neural network. The temporal graph convolutional neural network is used to characterize the spatial correlation characteristics and short-sequence temporal features of fault features in the distribution network topology, while the iFlowformer is used to capture the long-sequence temporal features of fault features, mine the discriminative information in multi-dimensional fault features, and combine the multi-task learning framework to perform joint modeling of fault type classification and fault branch classification, so as to realize the joint classification of fault branches and types.

[0009] Furthermore, the temporal graph convolutional neural network utilizes a graph convolutional network (GCN) to spatially aggregate node features without depending on the node arrangement order. It then uses a gated recurrent unit (GRU) to perform temporal modeling on the aggregated feature sequence, thereby achieving a joint characterization of the spatial correlation and temporal dynamics of fault features.

[0010] As a further technical limitation, a confusion matrix is ​​used to statistically analyze the classification results of fault branches and fault types. The classification performance of the fault classification model is evaluated by combining the classification accuracy and the total loss function. The location performance is comprehensively evaluated by using a fault distance prediction scatter plot and combining evaluation indicators such as the coefficient of determination, mean absolute error, and root mean square error, and the regression prediction of the fault location is completed.

[0011] As a further technical limitation, a distribution network topology model is constructed based on the node-branch connection relationship of the multi-branch active distribution network. The actual nodes of the distribution network are used as graph nodes and the actual branches of the distribution network are used as edges. The obtained multi-dimensional fault feature vectors are mapped to the constructed distribution network topology model to realize the fusion of multi-dimensional fault feature vectors and distribution network topology information, and to construct a fault feature space.

[0012] As a further technical limitation, in the process of obtaining the multidimensional fault feature vector of a multi-branch active distribution network, the three-phase voltage signal and zero-mode voltage component of the power supply node of the multi-branch active distribution network are collected, and the features of the three-phase voltage signal and zero-mode voltage component are extracted from the time domain and frequency domain respectively, and a multidimensional fault feature vector containing wavelet packet energy entropy and time domain statistical features is constructed.

[0013] According to some embodiments, the second aspect of the present invention provides a fault location system for a multi-branch active distribution network, employing the following technical solution: A fault location system for a multi-branch active distribution network includes: The acquisition module is configured to acquire multidimensional fault feature vectors of a multi-branch active distribution network; The module is configured to construct a fault feature space based on the acquired multidimensional fault feature vectors. The classification module is configured to perform multi-task fault classification of multi-branch active distribution networks based on the constructed fault feature space and the preset fault classification model, and obtain the joint fault classification result. The location module is configured to perform regression prediction on the fault location based on the obtained joint fault classification results, fault feature space and regression model, and complete the fault location of multi-branch active distribution network.

[0014] According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium, employing the following technical solution: A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the multi-branch active distribution network fault location method as described in the first aspect of the present invention.

[0015] According to some embodiments, the fourth aspect of the present invention provides an electronic device, which adopts the following technical solution: An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps in the multi-branch active distribution network fault location method as described in the first aspect of the present invention.

[0016] According to some embodiments, the fifth aspect of the present invention provides a computer program product, which adopts the following technical solution: A computer program product includes software code, wherein the program in the software code performs the steps of the multi-branch active distribution network fault location method as described in the first aspect of the present invention.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention employs wavelet packet transform to analyze the energy characteristics of voltage and its zero-mode component during faults, while integrating voltage time-domain statistical features. Based on the distribution network topology information, it constructs a multi-dimensional fault feature space that reflects the spatial correlation of nodes. This allows for the extraction of feature information containing fault sections and locations under complex operating conditions, overcoming the problem that single features have limited representation capabilities under complex operating conditions.

[0018] This invention, based on distribution network topology information, utilizes TGCN for feature propagation and aggregation regardless of node arrangement order to characterize the spatial correlation characteristics in multi-branch networks and model the short-sequence temporal dynamics of fault features. To enhance the model's ability to model long-sequence temporal relationships and global temporal feature interactions, iFlowformer is introduced to learn long-sequence temporal relationships across observation points and feature channels. This fully mines the discriminative information contained in multi-dimensional fault features, achieving joint classification of fault branches and types, as well as precise fault location, thus improving fault location accuracy under complex operating conditions. Attached Figure Description

[0019] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0020] Figure 1 This is a flowchart of the multi-branch active distribution network fault location method in Embodiment 1 of the present invention; Figure 2 This is a diagram of the overall architecture for time-frequency fault feature extraction and topology fusion in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the three-layer wavelet packet decomposition in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the TGCN structure in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the GCN structure in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the GRU structure in Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of the iFlowformer structure in Embodiment 1 of the present invention; Figure 8 This is a schematic diagram of the multi-task learning structure in Embodiment 1 of the present invention; Figure 9 This is a schematic diagram of the actual 33-node distribution network simulation model in Embodiment 1 of the present invention; Figure 10 This is a schematic diagram of the precise fault location results in Embodiment 1 of the present invention; Figure 11 This is a schematic diagram comparing the precise positioning of the four models in Embodiment 1 of the present invention; Figure 12 This is a structural block diagram of the multi-branch active power distribution network fault location system in Embodiment 2 of the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0025] In this invention, terms such as "fixed connection," "connected," and "linked" should be interpreted broadly, indicating a fixed connection, an integral connection, or a detachable connection; a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can determine the specific meaning of these terms in this invention based on the specific circumstances, and they should not be construed as limitations on the invention.

[0026] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0027] Example 1 Embodiment 1 of the present invention introduces a fault location method for multi-branch active distribution networks.

[0028] like Figure 1 The method for fault location in a multi-branch active distribution network, as shown, includes: Obtain multidimensional fault feature vectors for multi-branch active distribution networks; Based on the acquired multidimensional fault feature vectors, a fault feature space is constructed. Based on the constructed fault feature space and the preset fault classification model, multi-task fault classification is performed on the multi-branch active distribution network to obtain the joint fault classification result. Based on the obtained joint fault classification results, fault feature space and regression model, the fault location is predicted by regression, and the fault location of the multi-branch active distribution network is completed.

[0029] This embodiment introduces a fault location method for multi-branch active distribution networks based on the time-frequency-spatiotemporal joint modeling concept, using TGCN-iFlowformer and time-frequency-topology fusion. It employs wavelet packet transform to analyze the energy characteristics of voltage and its zero-mode component during a fault, while simultaneously fusing voltage time-domain statistical features. Based on the distribution network topology information, a multi-dimensional fault feature space reflecting the spatial relationships between nodes is constructed. This space can extract feature information about fault sections and locations under complex operating conditions, overcoming the limited representational ability of single features in complex situations. Based on the distribution network topology information, TGCN is used for feature propagation and aggregation without depending on the node arrangement order, characterizing the spatial correlation characteristics in the multi-branch network and modeling the short-sequence temporal dynamics of fault features. To further enhance the model's ability to model long-sequence temporal relationships and global temporal feature interactions, iFlowformer is introduced to learn long-sequence temporal relationships across observation points and feature channels. This fully mines the discriminative information contained in the multi-dimensional fault features, achieving joint classification of fault branches and types, and precise fault location, thus improving fault location accuracy under complex operating conditions.

[0030] Based on the time-frequency and spatiotemporal joint modeling concept, this embodiment constructs a model from time-domain features, frequency-domain features, and network topology as follows: Figure 2 The overall architecture for fault feature extraction and fusion is shown below, specifically: Given the limited number of measuring devices, this embodiment only configures measuring devices at power supply nodes for voltage acquisition. When a fault occurs in the distribution network, the three-phase voltage... u a , u b and u c It typically contains both reactive and interphase coupled linear mode components. and and the zero-mode component of the ground imbalance characteristics u 0. Direct analysis of fault signals makes it difficult to effectively separate fault-related information. To reduce the effects of inter-phase coupling and highlight fault-sensitive characteristic components, a Kelvin-Behr transform is performed on the voltage signal to extract the zero-mode component, which is highly sensitive to single-phase grounding and unbalanced faults. u 0; To extract the energy information contained in the signal in the frequency domain, this embodiment uses wavelet packet transform (WPT) to... u a , u b , u c and u 0. Multi-scale decomposition is performed to fully explore the energy distribution information in different frequency sub-bands; wavelet packet energy entropy is introduced as a frequency domain feature index to characterize the redistribution of energy in each frequency band after the fault occurs; To visually depict the fault u a , u b , u c and u The influence of amplitude variation and waveform shape is analyzed in this embodiment by using time-domain statistical methods such as root mean square (RMS), standard deviation (STD), and crest factor (CF) to extract the waveform shape. u a , u b , u c and u The time-domain characteristics of 0 are used to supplement the frequency-domain characteristics in describing amplitude and waveform changes. The power distribution network is composed of interconnected power sources, lines, loads and various control devices. Its operating status and electrical characteristics are significantly constrained by the network topology and naturally possess graph structure characteristics. In this embodiment, the power distribution network is abstracted as a graph structure model composed of nodes and edges to explicitly depict the electrical connection relationships and information transmission paths between different nodes. This embodiment embeds a time-frequency fault feature matrix, which includes frequency-domain energy characteristics and time-domain statistical characteristics, into the graph structure model as attribute information of the measurement nodes. Nodes without measurement devices retain only their topological structure information, thereby achieving an effective characterization of the overall operating status of the distribution network under limited measurement conditions. By fusing voltage time-frequency characteristics with the distribution network topology, a time-frequency-topology fault feature space is constructed, providing a unified data representation foundation for subsequent spatial correlation learning and fault localization based on graph time-series models.

[0031] In this embodiment, when a fault occurs in the distribution network, the following applies: u a , u b , u c and u By applying WPT to decompose the original non-stationary fault signal into different frequency band subspaces, a fine spectral division is obtained. The structure of the three-level wavelet packet decomposition is shown in the figure below. Figure 3 As shown.

[0032] Wavelet packet decomposition is defined as ;in, and Let be the scaling function and the wavelet function, representing the low-frequency and high-frequency components of the decomposed signal; and These are the coefficients of the low-pass and high-pass filters, used to extract low-frequency and high-frequency information from the signal; Indicates the position of the filter coefficients.

[0033] Perform wavelet packet decomposition on the signal, the first... j The first layer i Energy corresponding to each frequency band for ;in, For the first j The first layer i Each frequency band wavelet packet coefficient ; M For the first i The length of each frequency band.

[0034] With the j The ratio of total energy of layer signals for ;in, For the firstj Total energy of layer signal, .

[0035] No. j The wavelet energy entropy S of the layer is ; Thus, a combination of u a , u b , u c and u 0 frequency domain fault feature vector This provides an effective frequency domain representation for the subsequent construction of a multidimensional fault feature space.

[0036] This embodiment sets the analysis time window after the fault occurs, and a total of samples are obtained. B Each discrete voltage sampling point, for u a , u b , u c and u Calculate the following time-domain statistical characteristics respectively.

[0037] (1)RMS RMS reflects the effective amplitude of the voltage within the analysis window. After a fault occurs, the voltage amplitude usually decreases significantly or fluctuates due to short circuits or grounding, resulting in a significant change in RMS; that is... ;in, u ( t ) indicates that the voltage is at the 1st . t The instantaneous value at each sampling time.

[0038] (2) STD STD describes the dispersion of voltage within the analysis window and reflects the strength of voltage waveform fluctuations. During a fault, due to the introduction of transient and oscillating components, voltage waveform fluctuations intensify, and STD typically increases significantly. ;in, This is the average value of the voltage signal.

[0039] (3) CF The voltage flux density (CF) characterizes the relationship between the peak and effective voltage values, reflecting the degree of waveform distortion. Fault transients are often accompanied by spikes and impulse components, causing a significant increase in CF. ; The fault types in the distribution network are diverse and the imbalance characteristics are significant. Therefore, simultaneously utilizing u a , u b , uc and u 0 performs time-domain feature extraction and constructs a time-domain fault feature matrix O2. .

[0040] In the process of graph structure modeling, this embodiment uses graphs. Describe the network structure status of the distribution network, where, J To construct a corresponding graph node set for the nodes in the distribution network; E Let A be the set of graph edges determined by the actual topological connections of the distribution network, used to describe the electrical connections between nodes. Based on this, to further characterize the spatial association strength between adjacent nodes, a Gaussian kernel function based on physical distance is introduced to weight the graph edges, constructing a weighted adjacency matrix A. Let nodes... m and nodes n The length of the line between them is l mn Then the adjacency matrix element A mn Defined as ;in, This is a distance scale parameter used to control the degree of attenuation of spatial correlation. This is to enhance the robustness of adjacency weights to different line length distributions.

[0041] The weighted adjacency matrix constructed in this embodiment can quantitatively reflect the spatial proximity relationship between nodes while keeping the actual topology of the distribution network unchanged, providing reasonable spatial constraints for feature propagation in graph convolution operations.

[0042] Based on the constructed frequency domain fault feature vector O1 and time domain fault feature matrix O2, and graph structure modeling G Construct a time-frequency-topology fault feature space X; that is... .

[0043] After a fault occurs, the time-frequency characteristics of the voltage signal exhibit significant spatiotemporal coupling between different nodes. TGCN utilizes GCN to spatially aggregate node features regardless of node arrangement order, and then uses a Gated Recurrent Unit (GRU) to perform temporal modeling on the aggregated feature sequence, thereby achieving a joint characterization of the spatial correlation and temporal dynamics of fault features. Its structure is as follows: Figure 4 As shown.

[0044] GCN is a type of deep learning model oriented towards graph-structured data. It leverages the ability to aggregate spatial features to automatically learn the influence of neighborhood features on target nodes. Its structure is illustrated below. Figure 5 As shown.

[0045] Based on the constructed time-frequency-topology fusion fault feature space, using the weighted adjacency matrix A as a graph structure constraint, the multi-dimensional fault feature matrix corresponding to each node is used as the attribute information of the graph node, forming the node feature matrix C; that is... .

[0046] GCN in s The feature propagation and update process of the layer is as follows ;in, For the first s Layer output information; Let A be the degree matrix. For the first s Layer input information; For the first s Layer weight matrix; This is the ReLU activation function.

[0047] GRU is a type of deep learning model that focuses on the dynamics of short time sequences. It adaptively adjusts historical states through update and reset gates, achieving effective memorization and updating of short time sequence features with a relatively small parameter scale. Its structure is as follows: Figure 6 As shown.

[0048] GRU's update gate Reset door Candidate hidden output state and GRU output They are respectively ; ; ; ; in, Use the Sigmoid activation function; yes t Hidden layer output at time -1; It is the current t Input at any moment; , and They are respectively , and The weight matrix; , and They are respectively , and The bias vector; It is the hyperbolic tangent activation function; This indicates element-wise multiplication.

[0049] To effectively characterize the long-sequence temporal dependencies of multidimensional fault features, this embodiment, based on the spatial correlation and short-sequence temporal dynamics modeled by TGCN, introduces iFlowformer to further model the temporal features, the structure of which is as follows: Figure 7 As shown.

[0050] Temporal feature sequences output by TGCN Y for ;in, T This represents the length of the time series.

[0051] To construct the feature representation required for information flow modeling, the input sequence Y A linear mapping is performed at each time step to obtain the query vector. Key vector Sum value vector ;Right now ; ; ; in, For the present t Characteristic representation of time; , and They are respectively , and The trainable weight matrix.

[0052] The core idea of ​​iFlowformer is to view the time series modeling process as a process of information accumulating progressively over time. To characterize the importance of different time steps in long-sequence time series modeling, adaptive information flow weights are introduced. This enables iFlowformer to adaptively emphasize time segments that contribute more to the current fault identification, while suppressing the influence of redundant or noisy features during long-term learning; that is... ;in, It is a non-negative feature mapping function, which ensures the non-negativity of the information flow weights; For historical time steps The cumulative value represents the historical characteristic distribution up to the current moment; This indicates the degree of correlation between the current time step features and historical information; This is a normalization term used to satisfy the information flow conservation constraint.

[0053] The update rule for the state of the progressive information flow is defined as follows: ;in, This is a progressive information flow state, indicating the deadline. tThe cumulative results of historical characteristics; This represents the historical information flow state of the previous time step; this update mechanism enables the model to gradually accumulate historical information over time and control the contribution of current features to long-term memory through a weight adjustment mechanism, thereby avoiding the gradual decay of key information during long-sequence modeling.

[0054] at any time step t The output of the iFlowformer model is ;in, This represents a feedforward neural network used to map long-sequence temporal features to a specific task space.

[0055] In multi-branch active distribution networks, the fault branch determines the spatial propagation path of transient characteristics, while the fault type directly affects the amplitude distribution and time-frequency characteristic shape of the transient signal. The two have a significant coupling relationship at the characteristic level.

[0056] To fully leverage the complementary information between different tasks, this embodiment introduces a multi-task learning (MTL) framework to jointly model the two tasks of fault type classification and fault branch classification. By sharing feature representations, positive knowledge transfer between tasks is achieved, thereby improving the overall performance of the model under complex operating conditions. Its structure is as follows: Figure 8 As shown.

[0057] Based on the TGCN-iFlowformer model, through... Perform a mapping transformation to obtain shared features that serve as a multi-task learning framework. D ;Right now .

[0058] Dedicated output branches are set up for different tasks. For fault branch and fault type classification tasks, the predicted outputs are respectively ; ; in, and These represent the predicted probability distributions of the sample belonging to each candidate fault branch and fault type, respectively. and These are the weight matrix and bias vector for the fault branch classification task, respectively; and These are the weight matrix and bias vector for the fault type classification task, respectively; This is the activation function.

[0059] Loss function for fault branch and fault type classification task and Employing multi-class cross-entropy; that is ; ; in, and These represent the number of faulty branches and the number of fault types, respectively. and These are the one-hot codes for the faulty branch and the actual label of the fault type, respectively; and The model prediction samples for faulty branches and fault types belong to the first... The probability of a class.

[0060] In multi-task joint training, and Together they constitute the total loss function ;Right now ; in, and These are the weight coefficients for the fault branch and fault type classification tasks, respectively, used to balance the contributions of the two tasks during training and prevent one task from dominating the feature learning process. The shared feature extraction network is optimized through backpropagation of joint loss to learn discriminative features that balance fault branch localization and type identification.

[0061] To achieve accurate classification of faulty branches and fault types, as well as precise fault location in multi-branch active distribution networks, this embodiment employs the TGCN-iFlowformer fault location method based on time-frequency-topology feature fusion. The main steps are as follows: (A) Fault voltage feature extraction Three-phase voltage signals from power source nodes in a multi-branch active distribution network were acquired under limited measurement equipment conditions, and zero-mode voltage components were extracted using Kelenberger transform. Considering the non-stationary and abrupt characteristics of fault signals, features of the three-phase voltage and zero-mode voltage signals were extracted from both the frequency and time domains, constructing a multi-dimensional fault feature matrix that includes frequency domain wavelet packet energy entropy and time domain statistical features.

[0062] (B) Time-frequency-topology fusion A network topology graph model is constructed based on the node-branch connection relationship of the distribution network, with actual nodes as graph nodes and branches as edges, and a weighted adjacency matrix is ​​formed by combining line parameters. On this basis, multi-dimensional fault features are mapped to the graph structure to achieve deep integration of fault time-frequency features and distribution network topology information, and to construct a unified fault feature space.

[0063] (C) Multi-task fault classification In the task of identifying fault branches and fault types, a TGCN-iFlowformer classification model based on multi-task learning is constructed. TGCN is used to characterize the spatial correlation characteristics and short-sequence temporal characteristics of fault features in the distribution network topology, while iFlowformer is used to capture the long-sequence temporal characteristics of fault features. The joint classification of fault branches and fault types is achieved through a multi-task learning mechanism.

[0064] (D) Precise location of the fault In the task of precise fault location, the joint classification results are used as prior features and fused with the constructed time-frequency-topology fault feature space. They are then input into the TGCN-iFlowformer regression model to perform precise regression of the fault location.

[0065] (E) Results Analysis For the fault branch and fault type classification task, a confusion matrix is ​​used to statistically analyze the classification results, and the classification performance of the model is evaluated by combining classification accuracy and total loss function. For the fault precise location task, a fault distance prediction scatter plot is used, combined with the coefficient of determination. R 2 The positioning performance is comprehensively evaluated using evaluation metrics such as mean absolute error (MAE) and root mean square error (RMSE).

[0066] Case Analysis This embodiment uses a real 33-node distribution network in a certain area as the test system, and constructs a simulation system based on the PSCAD / EMTDC platform. Figure 9 The power distribution network simulation model shown has a sampling frequency of 20 kHz, and voltage data is acquired by placing measuring devices only at the power supply nodes.

[0067] The system has a rated voltage of 10.5 kV, a rated frequency of 50 Hz, and a total active power load of 4.694 MW. The distributed power supply adopts an inverter-type photovoltaic power generation model, with five photovoltaic power sources connected, injecting a total active power of 1.25 MW. Considering the distributed parameter characteristics of the line, a frequency-dependent phase domain model is used for the distribution line.

[0068] To fully simulate various fault conditions that may occur in actual power distribution networks, multiple fault scenarios are constructed, taking into account factors such as different transition resistances, initial phase angles of faults, fault types, and fault locations. Specific simulation parameters are shown in Table 1. Given the symmetry of the three-phase voltage in terms of structure and dynamic characteristics, for example, under single-phase ground fault conditions, the fault characteristics corresponding to different phases are mapped only to that phase, and their time-frequency evolution remains consistent. Therefore, in this embodiment, Ag represents a single-phase ground fault, AB represents a two-phase short-circuit fault, and ABg represents a two-phase ground fault in the fault scenario settings.

[0069] Table 1 Simulation Operating Condition Parameter Settings

[0070] As shown in Table 1, a total of 23,040 samples were generated through combination traversal, forming a multi-dimensional node feature matrix for subsequent graph time-series model input. The sample set was divided into a training set and a test set in a 7:3 ratio, used for model training and performance validation, respectively.

[0071] For the fault branch classification task and the fault type classification task, this embodiment uses a confusion matrix to evaluate the classification performance of the model under the single-task learning condition. The rows represent the true class of the sample and the columns represent the model's predicted classification results, as shown in Table 2.

[0072] Table 2. Binary Confusion Matrix

[0073] TP and TN represent the number of correctly identified positive and negative samples, respectively, while FN and FP represent the number of incorrectly identified positive and negative samples, respectively. In model performance evaluation, classification accuracy is... As a core metric, it is defined as the proportion of samples correctly identified by the model out of the total number of samples; that is... .

[0074] In the joint classification scenario based on multi-task learning, the fault branch classification and fault type classification tasks share feature representations and are optimized synchronously. In this embodiment, a weighted total loss function is used to evaluate the overall performance of the multi-task model.

[0075] For the task of accurately locating the fault, this embodiment uses the mean absolute error. Root mean square error and coefficient of determination A comprehensive evaluation of the model's regression performance is conducted; that is... ; ; ; in, n The number of samples; and The first i The true and predicted values ​​of the fault distance for each sample; This represents the average fault distance. Wherein, and The smaller the value, the lower the model's prediction error for the fault location and the higher the positioning accuracy. The larger the value, the stronger the model's ability to fit the fault distance.

[0076] To verify the performance advantages of the proposed TGCN-iFlowformer method (F4) in fault segment classification and fault type classification tasks, the CNN-BiGRU-ATT method (F1), CNN-Transformer method (F2), and GAT method (F3) were selected as comparison models, and their fault segment classification accuracy was evaluated. λ br Fault type classification accuracy λ type and weighted total loss function The classification performance of the four methods was comprehensively evaluated using three indicators, and the comparison results are shown in Table 3.

[0077] Table 3 Comparison of classification results from different models

[0078] Table 3 shows that the four models exhibit significant performance differences in the classification task, and all four models outperform single-task learning under multi-task learning. The classification accuracy of F1 and F2 is generally lower, mainly because their ability to uncover differences in fault characteristics under complex working conditions is still insufficient; under multi-task learning, F3... λ br and λ type The values ​​are 0.9867 and 0.9874 respectively. The score is 0.8913 because F3 significantly improves classification performance by introducing graph structure information, but its ability to model long-term features remains limited. Compared to F1, F2, and F3, F4 has the best overall performance under multi-task learning conditions. λ br and λ type They reached 0.9975 and 0.9971 respectively. The value is 0.7786. This is because the method in this embodiment can more effectively integrate time-frequency features and topological information, and improve the classification accuracy in complex fault scenarios through graph-time series joint modeling and multi-task collaborative learning.

[0079] In this embodiment, the fault branch classification results and fault type classification results are used as prior features, integrated into the time-frequency-topology fault feature space, and jointly input into the TGCN-iFlowformer regression model to accurately regress the fault location.

[0080] To verify the accuracy of the TGCN-iFlowformer regression model proposed in this embodiment, 6912 test samples were input into the trained model for prediction. The accurate fault location results are as follows: Figure 10 As shown, the data points are closely distributed near the fitted curve, and the deviation between the positioning distance and the actual distance is small, indicating that the TGCN-iFlowformer model proposed in this embodiment has high performance in accurate fault location.

[0081] To verify the localization accuracy of the TGCN-iFlowformer model (F4) proposed in this embodiment, three comparison models were set up: CNN-BiGRU-ATT model (F1), CNN-Transformer model (F2), and GAT model (F3). Fifty sets of samples were randomly selected from the test set for testing. The localization results of the four models were compared with the actual values. Figure 11 As shown, the four models exhibit significant differences in fault location. F1 and F2 show large deviations between predicted and actual values ​​for some samples. This is because they fail to adequately utilize the time-frequency characteristics of fault signals and the distribution network topology, making it difficult to fully characterize the spatial propagation characteristics and distance variation patterns of fault disturbances in multi-branch networks. F3 shows improved accuracy in predicting results, but some samples still fluctuate. This is because F3, by incorporating graph structure information, possesses a certain ability to characterize the spatial correlation characteristics of fault propagation.

[0082] Compared to F1, F2, and F3, the prediction results of F4 are closer to the true values ​​overall, and the curve trend is more consistent. This indicates that the method in this embodiment has better fitting ability and more stable results in accurate fault location. This is mainly because the collaborative modeling of time-frequency information and topology effectively enhances the expression of fault distance characteristics.

[0083] To compare the performance of different models in the task of accurate fault location, this embodiment inputs all test samples into four pre-trained models for prediction, and then... , and A comparative analysis was conducted, and the results are shown in Table 4.

[0084] Table 4 Comparison of localization results of the four models

[0085] As shown in Table 4, among the four localization models, F4 performs best in the task of precise fault localization. MAE and R MSE The m and m are 58.233 m and 75.111 m respectively. R 2 The value was 0.9942, indicating that its regression accuracy and fitting ability were superior to the other three comparison models. Compared to F1 and F2, M AE The reductions of 94.115 m and 50.453 m are mainly due to the fact that F4, in its modeling process, not only utilizes the time-frequency information of the fault signal but also introduces distribution network topology constraints, enabling it to more effectively characterize the propagation differences of fault disturbances in multi-branch networks; compared to F3, M AE The accuracy was reduced by 16.18 m because F4, combined with TGCN and iFlowformer, jointly modeled the spatial correlation, short-term dynamics, and long-sequence dependencies of fault features, thereby improving the accuracy of fault location regression.

[0086] To verify the synergistic effect of time-frequency features and topology information in fault feature representation, this embodiment designed an ablation comparison experiment of fault feature space construction methods. Four fault feature spaces were constructed: time-domain feature-frequency feature fusion (SP), time-domain feature-topology fusion (ST), frequency-domain feature-topology fusion (PT), and the time-frequency feature-topology fusion (SPT) proposed in this embodiment. λ br , M AE , R MSE and R 2 A comparative analysis was conducted, and the results are shown in Table 5.

[0087] Table 5 Comparison of Spatial Ablation Results for Four Fault Characteristics

[0088] Table 5 shows that there are significant differences in the performance of the four feature space models. SP method R 2 The error was only 0.9328, the largest among all methods, because signal characteristics alone cannot distinguish similar transient waveforms on different branches. Introducing topological information improved the performance of both the ST and PT methods, with PT outperforming ST. λ br It is 0.9949. R 2 The value is 0.9840 because frequency domain characteristics are more sensitive to line impedance parameters and naturally synergize with topological constraints, enabling more effective location of the correct branch. The SPT method proposed in this embodiment achieves the best results. λbr Increased to 0.9970. M AE It dropped to 58.233 m. R 2 Increased to 0.9942, compared to PT. M AE The decrease of 40.205 m is mainly due to the fact that time-domain features can statistically reflect the amplitude and waveform changes of voltage signals after a fault, frequency-domain features can reveal the distribution pattern of transient energy in different frequency bands, and topological information can describe the spatial propagation relationship of fault disturbances in the distribution network. The three features are highly complementary in their characterization, thereby improving the model's ability to identify fault locations and its positioning accuracy.

[0089] To evaluate the applicability of the model under different transition resistance conditions, this embodiment selects fault scenarios outside the training sample set for testing. Figure 9 In the distribution network shown, single-phase grounding is set at a distance of 1km from the first node on branches 8-14 and 16-24, and transition resistances of 0.5 Ω, 15 Ω, 150 Ω and 3000 Ω are configured for simulation analysis. The results are shown in Table 6.

[0090] Table 6 Comparison of results for four models under different transition resistances

[0091] As shown in Table 6, all four models can correctly classify fault branches under different transition resistance conditions, but the accuracy of localization varies significantly. F1 and F2 have relatively large overall localization errors, while F3 exhibits significant error fluctuations and limited adaptability to changes in transition resistance. In contrast, F4 maintains an accuracy localization error within 58 m under low resistance conditions (0.5Ω and 15Ω) and within 107 m under high resistance conditions (3000Ω). This indicates that the method in this embodiment has stronger generalization ability to varying transition resistance scenarios. This is mainly because F4 integrates time-frequency features and topological information, TGCN characterizes the spatial correlation of fault propagation, and iFlowformer extracts long-term dynamic features, thus maintaining high localization accuracy even when transient amplitudes are weakened by transition resistance.

[0092] To evaluate the applicability of the model under different initial phase angle conditions of the fault, this embodiment selects fault scenarios outside the training sample set for testing. Figure 9 In the distribution network shown, single-phase grounding was set at a distance of 1km from the first node on branch 8-14 and branch 16-24, and the initial phase angle of the fault was configured as 20°, 50°, 70° and 100° for simulation analysis. The results are shown in Table 7.

[0093] Table 7 Comparison of results for four models under different initial phase angles of the fault

[0094] As shown in Table 7, all four models can correctly classify the faulty branch under different initial phase angles of the fault, but the accuracy of the localization varies significantly. F1 and F3 have relatively large localization errors overall, while F2 exhibits significant error fluctuations and limited adaptability to changes in the initial phase angle. In contrast, F4's accuracy in localization under different initial phase angles is generally controlled within 55 m, indicating that the method in this embodiment has a stronger generalization ability to changes in the initial phase angle of the fault. This is mainly because F4 can simultaneously utilize the time-frequency information of the fault signal and network structure constraints to more fully characterize the changes in transient waveform amplitude and energy distribution under different initial phase angles, thereby reducing the impact of initial phase angle changes on the localization results.

[0095] To evaluate the model's applicability under different fault types, this embodiment selects fault scenarios outside the training sample set for testing. Figure 9 In the distribution network shown, fault points 1km away from the first node were set on branches 8-14 and 16-24, and the fault types were configured as Ag, AB, ABg, and ABC for simulation analysis. The results are shown in Table 8.

[0096] Table 8 Comparison of results for the four models under different fault types

[0097] As shown in Table 8, all four models can correctly classify fault branches under different fault types, but their accurate location varies significantly. F1, F2, and F3 have relatively large overall location errors, with average errors of 92.27 m, 107.27 m, and 109.97 m, respectively. In contrast, F4's average error is only within 43.1 m, indicating that the method in this embodiment has strong scenario generalization ability for different fault types. This is mainly because the F4 model can comprehensively explore the differences in waveform characteristics, frequency band energy distribution, and propagation paths of transient signals under different fault types, and enhances the identification ability of complex fault modes through graph-time series joint modeling. Therefore, it can maintain relatively stable location performance even when the fault type changes.

[0098] To evaluate the robustness of the method in this embodiment under noise interference conditions, this embodiment superimposes Gaussian noise with signal-to-noise ratios of 20 dB, 30 dB, 40 dB, and 50 dB onto the test data in the test set, and then... λ br , M AE , R MSE and R 2 A comparative analysis was conducted, and the results are shown in Table 9.

[0099] Table 9 Comparison of results for the four models under noise conditions

[0100] As shown in Table 9, when the noise level increases from 50 dB to 20 dB, the four models... λ br , M AE , R MSE and R 2 All showed a decreasing trend, with F4 being the best overall in this embodiment. At 20 dB, F1 and F2... λ br They decreased to 0.7193 and 0.7377 respectively. R 2 The values ​​dropped to 0.7064 and 0.8263 respectively. This is because the F1 and F2 models do not make sufficient use of topological information and are unable to effectively maintain the spatial correlation and temporal stability of fault characteristics under strong noise interference; F3 at 20 dB, its λ br and R 2 The noise immunity of F4 decreased to 0.8553 and 0.8458 respectively, because the introduction of the graph structure in F3 enhanced its noise immunity; compared with F1, F2 and F3, F4 at 20 dB... λ br and R 2 The scores can still reach 0.8799 and 0.8655, indicating that F4 has better robustness. This is mainly because TGCN enhances the effective feature propagation capability between nodes through topological constraints, which can reduce the impact of local noise disturbance on spatial feature extraction. iFlowformer can model long-sequence time-series features, retain relatively stable fault information and suppress the masking of key time-frequency features by random Gaussian noise. Therefore, it can maintain high classification and localization performance under noisy conditions.

[0101] To evaluate the applicability of the method in this embodiment under conditions of limited training sample size, this embodiment constructs training sample sets of different sizes and compares and analyzes the model's performance. While keeping the test set unchanged, the number of training samples is reduced to 30%, 50%, 70%, and 90% of the original sample set, respectively, and then... λ br , M AE , R MSE and R 2 A comparative analysis was conducted, and the results are shown in Table 10.

[0102] Table 10 Comparison of results for the four models under different training samples

[0103] As shown in Table 10, as the training samples are gradually reduced, the four models... λ br , M AE , R MSE and R 2 All showed a downward trend, with F4 being the best overall in this embodiment. When the training samples were only 30%, F1, F2, and F3... λ br and R 2 All fell below 0.74 because F1 and F2 models lacked topological correlation with the distribution network, and F3 model's ability to co-characterize multi-dimensional fault features was insufficient, making it difficult to fully learn the inherent distribution patterns of fault features under small sample conditions; F4... λ br and R 2 The values ​​are 0.8004 and 0.8138, respectively. Compared with the other three comparison models, the model performance is better, indicating that the method in this embodiment has stronger robustness to changes in sample size. This is because the fusion of time-frequency features and topological information enhances the separability of input features and reduces the model's dependence on large-scale samples. In addition, TGCN and iFlowformer jointly model spatial correlation, short-term dynamics and long-term dependence, enabling the model to learn a relatively stable fault discrimination mechanism even when there are insufficient samples.

[0104] To evaluate the robustness of the method in this embodiment under conditions of measurement data loss, this embodiment constructs a data loss scenario to test the model performance. While keeping the test set unchanged, 8%, 6%, 4%, and 2% of the measurement data in each sample of the test set are randomly set to zero to simulate data loss. λ br , M AE , R MSE and R 2 A comparative analysis was conducted, and the results are shown in Table 11.

[0105] Table 11 Comparison of results for four models under missing data conditions

[0106] As shown in Table 11, as the data missing rate increases from 2% to 8%, the four models... λbr , M AE , R MSE and R 2 All showed a downward trend, and in this embodiment, F4 was generally better than F1, F2, and F3. F1 and F2... λ br and R 2 The value dropped to around 0.60 because F1 and F2 do not make sufficient use of topological information, making it difficult to effectively characterize the spatial correlation of fault features; F3's... λ br and R 2 The F3 model's performance dropped to around 0.81 because it incorporates graph structure information, enhancing its ability to model relationships between nodes and thus exhibiting some resilience against missing data. The F4 model, however, performed better with a missing data rate of 8%. λ br and R 2 The scores were 0.8424 and 0.8478 respectively, which were significantly better than the other three comparison models. This is because the topological information provides neighborhood association compensation for missing nodes, and the model can recover some fault propagation information by using the features of neighboring nodes. In addition, the time-frequency features characterize fault disturbances from different dimensions and have a certain degree of complementarity, which enables the model to maintain relatively stable performance when data is missing.

[0107] To address the challenges of complex spatiotemporal coupling of fault features, insufficient fusion of time-frequency features and topology structure, and limited accuracy and generalization ability in fault location within multi-branch active distribution networks under limited measurement conditions, this embodiment proposes a TGCN-iFlowformer multi-branch active distribution network fault location method based on time-frequency-topology fusion. The time-frequency-topology fusion feature space constructed in this embodiment exhibits optimal performance. λ br It is 0.9970. M AE , R MSE and R 2 The values ​​are 58.233 m, 75.111 m, and 0.9942, respectively, indicating that this feature space can effectively improve fault identification and location accuracy. This embodiment constructs a multi-task fault location model based on TGCN-iFlowformer, achieving fault branch classification, fault type classification, and accurate fault distance location. Results show that under multi-task learning conditions... λ br and λ typeThe values ​​are 0.9975 and 0.9971 respectively, with a total loss function of 0.7786. In the fault precise location task, the overall performance is better than the comparison models. At the same time, under different conditions such as transition resistance, fault initial phase angle, fault type, noise interference, limited training samples and missing data, the method of this embodiment still shows good stability, scene generalization ability and robustness.

[0108] Example 2 Embodiment 2 of the present invention introduces a fault location system for a multi-branch active distribution network.

[0109] like Figure 12 The multi-branch active distribution network fault location system shown includes: The acquisition module is configured to acquire multidimensional fault feature vectors of a multi-branch active distribution network; The module is configured to construct a fault feature space based on the acquired multidimensional fault feature vectors. The classification module is configured to perform multi-task fault classification of multi-branch active distribution networks based on the constructed fault feature space and the preset fault classification model, and obtain the joint fault classification result. The location module is configured to perform regression prediction on the fault location based on the obtained joint fault classification results, fault feature space and regression model, and complete the fault location of multi-branch active distribution network.

[0110] The detailed steps are the same as those of the multi-branch active distribution network fault location method provided in Example 1, and will not be repeated here.

[0111] Example 3 Embodiment 3 of the present invention provides a computer-readable storage medium.

[0112] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the multi-branch active distribution network fault location method as described in Embodiment 1 of the present invention.

[0113] The detailed steps are the same as those of the multi-branch active distribution network fault location method provided in Example 1, and will not be repeated here.

[0114] Example 4 Embodiment 4 of the present invention provides an electronic device.

[0115] An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements the steps in the multi-branch active distribution network fault location method as described in Embodiment 1 of the present invention.

[0116] The detailed steps are the same as those of the multi-branch active distribution network fault location method provided in Example 1, and will not be repeated here.

[0117] Example 5 Embodiment 5 of the present invention provides a computer program product.

[0118] A computer program product includes software code, wherein the program in the software code performs the steps of the multi-branch active distribution network fault location method as described in Embodiment 1 of the present invention.

[0119] The detailed steps are the same as those of the multi-branch active distribution network fault location method provided in Example 1, and will not be repeated here.

[0120] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0121] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0122] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0123] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0124] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0125] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0126] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A fault location method for a multi-branch active distribution network, characterized in that, include: Obtaining a multidimensional fault feature vector for a multi-branch active distribution network includes: collecting three-phase voltage signals and zero-mode voltage components from the power supply nodes of the multi-branch active distribution network; extracting features of the three-phase voltage signals and zero-mode voltage components from the time and frequency domains respectively; and constructing a multidimensional fault feature vector containing wavelet packet energy entropy and time-domain statistical features. Specifically, wavelet packet decomposition is defined as: ； in, and Let be the scaling function and the wavelet function, representing the low-frequency and high-frequency components of the decomposed signal; t is a continuous time variable; and These are the coefficients of the low-pass and high-pass filters, used to extract low-frequency and high-frequency information from the signal; Indicates the position of the filter coefficients; Perform wavelet packet decomposition on the signal, the first... j The first layer i Energy corresponding to each frequency band for: ； in, For the first j The first layer i Each frequency band wavelet packet coefficient ; M For the first i The length of each frequency band; With the j The ratio of total energy of layer signals for: ； in, For the first j Total energy of layer signal, ; No. j The wavelet energy entropy S of the layer is ; The time-domain statistical characteristics are specifically calculated as follows: RMS reflects the effective amplitude of the voltage within the analysis window, that is: ； in, u ( t ) indicates that the voltage is at the 1st . t The instantaneous value at each sampling moment, where B is the number of discrete voltage sampling points; STD describes the degree of dispersion of voltage within the analysis window, that is: ； in, The average value of the voltage signal; CF characterizes the relationship between the peak voltage and the effective voltage, i.e.: ； The time-domain fault feature matrix O2 is constructed as follows: in, u a , u b and u c It is a three-phase voltage. u 0 represents the zero-modulus component; Based on the acquired multidimensional fault feature vectors, a time-frequency-topology fault feature space is constructed, including: constructing a distribution network topology graph model with actual distribution network nodes as graph nodes and actual distribution network branches as edges according to the node-branch connection relationship of the multi-branch active distribution network; mapping the acquired multidimensional fault feature vectors to the constructed distribution network topology graph model to realize the fusion of multidimensional fault feature vectors and distribution network topology information, and constructing a fault feature space. Based on the constructed fault feature space and the preset fault classification model, multi-task fault classification of multi-branch active distribution networks is performed to obtain joint fault classification results. The preset fault classification model adopts a time-series graph convolutional neural network-iFlowformer classification model based on multi-task learning. The time-series graph convolutional neural network is used to characterize the spatial correlation characteristics and short-series time-series features of fault features in the distribution network topology, while iFlowformer is used to capture the long-series time-series features of fault features and mine the discriminative information in multi-dimensional fault features. Combined with the multi-task learning framework, joint modeling of fault type classification and fault branch classification is performed to achieve joint classification of fault branches and types. Based on the obtained joint fault classification results, fault feature space and regression model, the fault location is predicted by regression, and the fault location of the multi-branch active distribution network is completed.

2. The fault location method for a multi-branch active distribution network as described in claim 1, characterized in that, The temporal graph convolutional neural network utilizes graph convolutional neural networks to spatially aggregate node features without depending on the node arrangement order. By performing temporal modeling on the aggregated feature sequence, it achieves a joint characterization of the spatial correlation and temporal dynamics of fault features.

3. The fault location method for a multi-branch active distribution network as described in claim 1, characterized in that, The classification results of fault branches and fault types are statistically analyzed using a confusion matrix. The classification performance of the fault classification model is evaluated by combining classification accuracy and total loss function. The location performance is comprehensively evaluated by using a fault distance prediction scatter plot and combining evaluation indicators such as coefficient of determination, mean absolute error and root mean square error, and the regression prediction of fault location is completed.

4. A fault location system for a multi-branch active distribution network, characterized in that, include: The acquisition module is configured to acquire a multi-dimensional fault feature vector of a multi-branch active distribution network, including: collecting three-phase voltage signals and zero-mode voltage components of the power supply nodes of the multi-branch active distribution network; extracting features of the three-phase voltage signals and zero-mode voltage components from the time domain and frequency domain, respectively; and constructing a multi-dimensional fault feature vector containing wavelet packet energy entropy and time-domain statistical features. Specifically, wavelet packet decomposition is defined as: ； in, and Let be the scaling function and the wavelet function, representing the low-frequency and high-frequency components of the decomposed signal; t is a continuous time variable; and These are the coefficients of the low-pass and high-pass filters, used to extract low-frequency and high-frequency information from the signal; Indicates the position of the filter coefficients; Perform wavelet packet decomposition on the signal, the first... j The first layer i Energy corresponding to each frequency band for: ； in, For the first j The first layer i Each frequency band wavelet packet coefficient ; M For the first i The length of each frequency band; With the j The ratio of total energy of layer signals for: ； in, For the first j Total energy of layer signal, ; No. j The wavelet energy entropy S of the layer is ; The time-domain statistical characteristics are specifically calculated as follows: RMS reflects the effective amplitude of the voltage within the analysis window, that is: ； in, u ( t ) indicates that the voltage is at the 1st . t The instantaneous value at each sampling moment, where B is the number of discrete voltage sampling points; STD describes the degree of dispersion of voltage within the analysis window, that is: ； in, The average value of the voltage signal; CF characterizes the relationship between the peak voltage and the effective voltage, i.e.: ； The time-domain fault feature matrix O2 is constructed as follows: in, u a , u b and u c It is a three-phase voltage. u 0 represents the zero-modulus component; The construction module is configured to construct a time-frequency-topology fault feature space based on the acquired multidimensional fault feature vectors. This includes: constructing a distribution network topology graph model with actual distribution network nodes as graph nodes and actual distribution network branches as edges based on the node-branch connection relationship of the multi-branch active distribution network; mapping the acquired multidimensional fault feature vectors to the constructed distribution network topology graph model; and realizing the fusion of multidimensional fault feature vectors and distribution network topology information to construct the fault feature space. The classification module is configured to perform multi-task fault classification of multi-branch active distribution networks based on the constructed fault feature space and the preset fault classification model, and obtain joint fault classification results. The preset fault classification model adopts the iFlowformer classification model based on multi-task learning temporal graph convolutional neural network. The temporal graph convolutional neural network is used to characterize the spatial correlation characteristics and short-sequence temporal features of fault features in the distribution network topology, and the iFlowformer is used to capture the long-sequence temporal features of fault features, mine the discriminative information in multi-dimensional fault features, and combine the multi-task learning framework to perform joint modeling of fault type classification and fault branch classification, so as to realize the joint classification of fault branches and types. The location module is configured to perform regression prediction on the fault location based on the obtained joint fault classification results, fault feature space and regression model, and complete the fault location of multi-branch active distribution network.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the multi-branch active distribution network fault location method as described in any one of claims 1-3.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-branch active distribution network fault location method as described in any one of claims 1-3.

7. A computer program product, comprising software code, characterized in that, The program in the software code performs the steps of the fault location method for multi-branch active distribution networks as described in any one of claims 1-3.

Images