A method and system for locating faults in a smart power distribution network

Abstract

The present invention discloses a method and system for locating faults in a smart distribution network, and relates to the technical field of distribution networks. The method comprises: collecting voltage amplitude and frequency signals in real time by deploying distributed acquisition devices at each node, calculating the spatial distribution law to generate a feature template library; combining the topological structure of the distribution network and the amplitude-frequency distribution characteristics, generating a node state label that matches the fault mode; designing a fault location model based on a graph convolutional network, introducing a time-space convolution and distribution law comparison module, extracting the fault mode and realizing node fault state prediction; collecting real-time data after the fault occurs, calculating the matching similarity between the distribution characteristics and the template library, and comprehensively judging the fault type and specific location; dynamically updating the distribution template library and optimizing the model algorithm by testing the performance of the model in different scenarios. The present invention is applicable to complex distribution network scenarios, and can achieve efficient fault diagnosis, especially in the case of distributed power supply access, dynamic topology and noise interference.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution networks, and in particular to a method and system for locating faults in a smart power distribution network. Background Art

[0002] As an important part of the power system, the distribution network's main task is to transmit electric energy from the high-voltage transmission network to end users. With the continuous growth of power demand and the widespread access of distributed power sources, the operating environment of the distribution network has become more complex. In the existing technology, fault location methods mainly include technologies based on physical models, data-driven and hybrid methods.

[0003] In recent years, with the development of artificial intelligence technology, especially the rise of deep learning and graph neural networks, fault location methods based on graph networks have received widespread attention. This type of method uses the topological information of the distribution network to build a graph model, takes the associations and features between nodes as input, and extracts spatiotemporal features through neural networks to achieve efficient diagnosis of complex faults. In addition, the introduction of dynamic topology adjustment and multi-dimensional feature extraction technology makes this type of method more adaptable when dealing with dynamic scenarios such as distributed power access and load fluctuations.

[0004] In response to the above problems, the present invention provides a smart distribution network fault location method and system, which improves the accuracy and adaptability of fault location when facing dynamic topology changes, data noise interference and multiple types of faults by combining the spatial distribution laws of the amplitude-frequency characteristics of the distribution network. Summary of the invention

[0005] In view of the above problems, the present invention provides a smart distribution network fault location method and system to solve the problems of insufficient positioning accuracy and robustness in the prior art when facing dynamic topology changes, data noise interference and multiple types of faults.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a method and system for locating faults in a smart power distribution network, comprising the following steps:

[0008] Step S1, by deploying distributed acquisition devices at each node, collecting voltage amplitude and frequency signals in real time, extracting instantaneous features and calculating spatial distribution laws, and generating a feature template library;

[0009] Wherein step S1 also includes the following sub-steps:

[0010] S1-1, by deploying distributed acquisition devices at each node, real-time voltage amplitude and frequency data are collected, equipped with anti-interference modules to reduce the impact of external noise, and equipped with data loss compensation mechanisms to ensure that effective features can still be obtained when the data is incomplete;

[0011] S1-2, using the improved signal decomposition algorithm ICEEMDAN, decompose the voltage signal and obtain the intrinsic mode functions IMFs. The decomposed voltage signal can be expressed as the superposition of each order IMFs and the residual term, as shown in formula (1):

[0012] Formula (1)

[0013] in, is the voltage signal, are the i-th order IMFs, is the residual term;

[0014] Since the order and frequency range of the decomposed IMFs are not fixed, IMFs cannot be directly used as fault features in the fault location model. The Pearson correlation coefficient is used to select IMFs with a strong correlation with the original signal, as shown in formula (2):

[0015] Formula (2)

[0016] in, is the Pearson correlation coefficient between the i-th order IMF and the original signal, is the covariance operator, is the covariance of IMF and original signal, and Respectively and The standard deviation of

[0017] The instantaneous amplitude and instantaneous frequency characteristics of the fault voltage are extracted by Hilbert transform, as shown in formula (3):

[0018] Formula (3)

[0019] in, is the reconstructed voltage signal, and are the instantaneous amplitude and instantaneous frequency respectively;

[0020] S1-3, using the amplitude and frequency fluctuation amplitude of the adjacent nodes, calculate the spatial distribution characteristics, analyze the characteristic spatial distribution patterns corresponding to different fault types and locations, generate a distribution feature template library, and define the amplitude-frequency feature change rate of each node, as shown in formula (4):

[0021] Formula (4)

[0022] in, For Node The amplitude-frequency characteristic change rate, For Node The amplitude and frequency of For Node The amplitude and frequency of For Node The geographical or topological distance.

[0023] Step S2, combining the topological structure and amplitude-frequency distribution characteristics of the distribution network, dynamically adjusting the adjacency matrix, constructing a graph network that can reflect the operating status in real time, and generating node status labels that match the fault mode;

[0024] Wherein step S2 also includes the following sub-steps:

[0025] S2-1, distribution pattern mapping, combines the spatial distribution characteristics with the distribution network topology to construct a feature map based on the amplitude-frequency distribution law. Each node contains the amplitude, frequency and fluctuation characteristics of its adjacent nodes, as shown in formula (5):

[0026] Formula (5)

[0027] in, is the instantaneous amplitude matrix, is the instantaneous frequency matrix, is the adjacency matrix, which describes the topological relationship between nodes in the graph;

[0028] S2-2, introduces a dynamic topology adjustment mechanism, which dynamically adjusts the nodes and edge relationships of the graph network to reflect the current operating status by monitoring the changes in the operating topology of the distribution network, such as the addition of nodes, load transfer, and branch disconnection; and changes in electrical parameters, such as voltage amplitude and frequency fluctuations;

[0029] S2-3, graph label generation, based on the feature template library of distribution rules, generate corresponding graph labels for each fault mode, and mark the fault status of the nodes in the graph according to the distribution feature template , generate state labels, as shown in formula (6):

[0030] Formula (6)

[0031] in, For Node i Status label, 1 means normal and 0 means fault.

[0032] Step S3, designing a fault location model based on a graph convolutional network, introducing a spatiotemporal convolution and distribution law comparison module, using node features and spatial distribution laws to extract fault modes and achieve node fault state prediction;

[0033] Wherein step S3 also includes the following sub-steps:

[0034] S3-1, model structure definition, design a distribution law driven graph convolutional network (DG-GCN), combine node features and spatial distribution features to extract fault signal patterns, and introduce a distribution comparison module based on spatiotemporal convolution to enhance the recognition of specific distribution laws;

[0035] The temporal features are extracted through one-dimensional convolution, as shown in equations (7) and (8):

[0036] Formula (7)

[0037] Formula (8)

[0038] in, and Respectively The frequency and amplitude feature matrices of the layers, and Respectively The frequency and amplitude feature matrices of the layers; and These are two one-dimensional convolution kernel functions used to extract time domain features; is the activation function; Indicates the multiplication of corresponding elements;

[0039] The spatial features are extracted by spectral convolution, as shown in formula (9):

[0040] Formula (9)

[0041] in, For the The frequency feature matrix of the layer, W is the adjacency matrix, D is the degree matrix of W, is the convolution kernel;

[0042] The node failure index is calculated through the fully connected layer, as shown in formula (10):

[0043] Formula (10)

[0044] in, The predicted node i The failure index, is the activation function, and is the weight matrix in the fully connected layer, and are the instantaneous amplitude characteristics and instantaneous frequency characteristics of node i respectively;

[0045] S3-2, define the distribution law matching loss function, minimize the difference between the true distribution and the predicted distribution, and use the multi-label classification loss function to optimize the model, as shown in formula (11):

[0046] Formula (11)

[0047] in, The actual node state label vector, is the node state vector predicted by the model, Loss function value, n is the total number of nodes, is the fault state of node i, is the failure index of node i;

[0048] S3-3, use the constructed distribution feature template library and historical fault data to train the model, use the cross-validation method to evaluate the model performance, divide the training set and the test set, optimize the model parameters with the multi-label classification loss function during the training process, and learn the relationship between the amplitude-frequency characteristics of the nodes and the spatial distribution law.

[0049] Step S4, collect real-time data after the fault occurs, calculate the matching similarity between the distribution characteristics and the template library, and combine the model reasoning results to comprehensively judge the fault type and specific location;

[0050] Wherein step S4 also includes the following sub-steps:

[0051] S4-1, after a fault occurs, collect the real-time voltage signals of each node in the distribution network, and use the instantaneous amplitude and frequency characteristic data obtained by the deployed distributed acquisition devices. The real-time data is quickly processed by the signal processing module, including noise filtering and abnormal point correction;

[0052] S4-2, distribution pattern matching, calculate the spatial distribution characteristics of real-time data, match it with the distribution template library, determine the fault type and location, calculate the spatial distribution characteristics of real-time data, compare it with the samples in the distribution template library, and match the similarity, as shown in formula (12):

[0053] Formula (12)

[0054] in, is the matching similarity between real-time data and template, is the amplitude-frequency characteristic distribution of real-time data, is the amplitude-frequency characteristic template in the distribution template library;

[0055] S4-3, input the amplitude and frequency characteristic data collected in real time into the distribution law driven graph convolutional network (DG-GCN) model, and generate the fault index of each node through the joint analysis of node characteristics and spatial distribution laws by the model. Combined with the similarity matching results of the distribution feature template library, the specific node and branch location where the fault occurred is comprehensively judged.

[0056] Step S5, by testing the performance of the model in different scenarios, dynamically updating the distribution template library and optimizing the model algorithm structure, the accuracy of fault location is improved.

[0057] Wherein, in step S5, the following sub-steps are also included:

[0058] S5-1, as the distribution network operation data accumulates, the distribution law template library is dynamically updated using the newly collected fault data, and the incremental learning mechanism is used to adapt to the changes in the distribution network operation environment, focusing on generating new distribution templates for new scenarios such as distributed power access, load transfer, and topology adjustment;

[0059] S5-2, testing the positioning accuracy of the model under different fault scenarios and topology change conditions, and testing the real-time response capability of the model in simulation and actual scenarios;

[0060] S5-3, based on the performance evaluation results, the model is optimized and adjusted, including the introduction of a new distribution law mining algorithm to improve the feature extraction capability of complex faults, improving the structure of the graph convolutional network (DG-GCN) to enhance the expressiveness of the model, and adjusting the loss function to better adapt to multi-label classification tasks.

[0061] A smart distribution network fault location system, comprising:

[0062] Data acquisition module, signal processing and feature extraction module, graph network construction module, fault location model module, distribution pattern template library and matching module, optimization module;

[0063] The data acquisition module collects voltage signals in real time at each node of the distribution network through a distributed acquisition device. The collected data includes instantaneous voltage amplitude and frequency characteristics, supports high sampling rate, and is equipped with anti-interference and data loss compensation mechanisms;

[0064] The signal processing and feature extraction module uses an improved ICEEMDAN algorithm to decompose the voltage signal, extract the intrinsic mode function and remove noise, selects the IMF component with strong correlation through the Pearson correlation coefficient for signal reconstruction, applies the Hilbert transform to extract the instantaneous amplitude and frequency characteristics, calculates the amplitude-frequency change rate of the adjacent nodes, and generates the spatial distribution characteristics of the nodes;

[0065] The graph network construction module constructs a graph network based on the amplitude-frequency distribution law, defines node and edge features, introduces a dynamic topology adjustment mechanism, updates the adjacency matrix according to real-time voltage fluctuations, and generates corresponding graph labels for each fault mode based on a template library;

[0066] The fault location model module uses the distribution law driven graph convolution network (DG-GCN) model to extract spatiotemporal features, adds a distribution comparison module based on spatiotemporal convolution, strengthens the recognition of specific distribution laws, and calculates the fault index of the node through the fully connected layer;

[0067] The distribution rule template library and matching module maintains the distribution rule template library, stores the spatial distribution characteristics under different fault modes, calculates the similarity between real-time data and the template, and further confirms the fault type and location through template matching;

[0068] The optimization module dynamically updates the distribution template library to adapt to changes in the operating environment, tests the performance of the model in different scenarios, including dynamic topology and noise interference, optimizes the model structure and algorithm, and improves the accuracy and robustness of fault location.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] The present invention extracts the spatial distribution characteristics of the voltage amplitude and frequency of the distribution network, generates a distribution law template library, and combines it with a dynamic topology adjustment mechanism to more accurately describe the characteristic distribution when a fault occurs, thereby greatly improving the accuracy of fault location.

[0071] The present invention designs a dynamic topology adjustment mechanism, which can adjust the node and edge relationships of the graph network according to the real-time changes in the operating status of the distribution network; at the same time, it uses a fault location model based on a graph convolutional network (DG-GCN) to achieve a deep fusion analysis of the spatiotemporal characteristics and spatial distribution laws of the nodes, and adapt to complex scenarios such as distributed power supply access, load fluctuations and topology changes.

[0072] The present invention performs similarity matching between real-time collected data and a distribution pattern template library, and combines the inference results of a graph network model to quickly determine faulty nodes and branches, significantly shortening the fault location time and meeting the real-time response requirements of the distribution network.

[0073] The present invention adopts an improved decomposition algorithm and noise suppression technology in the signal processing process, which improves the adaptability to noise interference in complex electrical environments; in addition, through the update mechanism of the dynamic template library, the model performance is continuously optimized to enhance the reliability of the system under conditions of data loss or interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It is understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0075] Figure 1 is a flow chart of the method of the present invention;

[0076] Figure 2 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but is only for selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0078] Please refer to Figure 1-Figure 2 The present invention provides a method and system diagram for locating faults in a smart distribution network, which includes the following steps:

[0079] Step S1, by deploying distributed acquisition devices at each node, collecting voltage amplitude and frequency signals in real time, extracting instantaneous features and calculating spatial distribution laws, and generating a feature template library;

[0080] Wherein step S1 also includes the following sub-steps:

[0081] S1-1, by deploying distributed acquisition devices at each node, real-time voltage amplitude and frequency data are collected, equipped with anti-interference modules to reduce the impact of external noise, and equipped with data loss compensation mechanisms to ensure that effective features can still be obtained when the data is incomplete;

[0082] S1-2, using the improved signal decomposition algorithm ICEEMDAN, decompose the voltage signal and obtain the intrinsic mode functions IMFs. The decomposed voltage signal can be expressed as the superposition of each order IMFs and the residual term, as shown in formula (1):

[0083] Formula (1)

[0084] in, is the voltage signal, are the i-th order IMFs, is the residual term;

[0085] Since the order and frequency range of the decomposed IMFs are not fixed, IMFs cannot be directly used as fault features in the fault location model. The Pearson correlation coefficient is used to select IMFs with a strong correlation with the original signal, as shown in formula (2):

[0086] Formula (2)

[0087] in, is the Pearson correlation coefficient between the i-th order IMF and the original signal, is the covariance operator, is the covariance of IMF and original signal, and Respectively and The standard deviation of

[0088] The instantaneous amplitude and instantaneous frequency characteristics of the fault voltage are extracted by Hilbert transform, as shown in formula (3):

[0089] Formula (3)

[0090] in, is the reconstructed voltage signal, and are the instantaneous amplitude and instantaneous frequency respectively;

[0091] S1-3, using the amplitude and frequency fluctuation amplitude of the adjacent nodes, calculate the spatial distribution characteristics, analyze the characteristic spatial distribution patterns corresponding to different fault types and locations, generate a distribution feature template library, and define the amplitude-frequency feature change rate of each node, as shown in formula (4):

[0092] Formula (4)

[0093] in, For Node The amplitude-frequency characteristic change rate, For Node The amplitude and frequency of For Node The amplitude and frequency of For Node The geographical or topological distance.

[0094] It should be noted that different fault types have an impact on spatial distribution. For a single-phase ground short circuit fault, where only one phase is short-circuited to ground, the amplitude drop and frequency fluctuation of the voltage signal will be significant near the fault point, while the impact on nodes far away from the fault point will be small, and the spatial distribution will be localized.

[0095] In the case of two-phase short circuit or two-phase ground short circuit, the current between the two phases changes more dramatically and the impact range is wider than that of single-phase short circuit. The voltage signal will show obvious disturbances at multiple nodes, and the spatial distribution has a certain symmetry.

[0096] Three-phase short circuit or three-phase ground short circuit is a serious fault. The voltage of all nodes in the system will be significantly affected. The amplitude drop and frequency fluctuation are large, the spatial distribution is relatively uniform, and the coverage area is wide.

[0097] The impact of different fault locations on spatial distribution: Faults close to the main transformer have a larger impact range, because the main transformer is connected to multiple important branches, and the fault will cause drastic changes in voltage signals at more nodes; faults close to the end load have a smaller impact range, and the abnormalities of voltage signals are mainly concentrated at the fault point and its adjacent nodes, and nodes far away from the fault point are less affected.

[0098] Step S2, combining the topological structure and amplitude-frequency distribution characteristics of the distribution network, dynamically adjusting the adjacency matrix, constructing a graph network that can reflect the operating status in real time, and generating node status labels that match the fault mode;

[0099] Wherein step S2 also includes the following sub-steps:

[0100] S2-1, distribution pattern mapping, combines the spatial distribution characteristics with the distribution network topology to construct a feature map based on the amplitude-frequency distribution law. Each node contains the amplitude, frequency and fluctuation characteristics of its adjacent nodes, as shown in formula (5):

[0101] Formula (5)

[0102] in, is the instantaneous amplitude matrix, is the instantaneous frequency matrix, is the adjacency matrix, which describes the topological relationship between nodes in the graph;

[0103] S2-2, introduces a dynamic topology adjustment mechanism, which dynamically adjusts the nodes and edge relationships of the graph network to reflect the current operating status by monitoring the changes in the operating topology of the distribution network, such as the addition of nodes, load transfer, and branch disconnection; and changes in electrical parameters, such as voltage amplitude and frequency fluctuations;

[0104] S2-3, graph label generation, based on the feature template library of distribution rules, generate corresponding graph labels for each fault mode, and mark the fault status of the nodes in the graph according to the distribution feature template

[0105] , generate state labels, as shown in formula (6):

[0106] Formula (6)

[0107] in, For Node i Status label, 1 means normal and 0 means fault.

[0108] It should be noted that the instantaneous amplitude and frequency characteristics of the nodes are combined with the adjacency relationship to construct a feature graph, and the adjacency matrix is ​​updated in real time through a dynamic topology adjustment mechanism to adapt to dynamically changing scenarios such as load transfer and distributed power supply access. By monitoring electrical parameters including voltage fluctuations and structural changes including new nodes and branch disconnections, adaptive adjustment of the graph network is achieved. Each node also generates a status label based on the distribution feature template, further enhancing the graph model's ability to express complex operating states.

[0109] Step S3, designing a fault location model based on a graph convolutional network, introducing a spatiotemporal convolution and distribution law comparison module, using node features and spatial distribution laws to extract fault modes and achieve node fault state prediction;

[0110] Wherein step S3 also includes the following sub-steps:

[0111] S3-1, model structure definition, design a distribution law driven graph convolutional network (DG-GCN), combine node features and spatial distribution features to extract fault signal patterns, and introduce a distribution comparison module based on spatiotemporal convolution to enhance the recognition of specific distribution laws;

[0112] The temporal features are extracted through one-dimensional convolution, as shown in equations (7) and (8):

[0113] Formula (7)

[0114] Formula (8)

[0115] in, and Respectively The frequency and amplitude feature matrices of the layers, and Respectively The frequency and amplitude feature matrices of the layer; and These are two one-dimensional convolution kernel functions used to extract time domain features; is the activation function; Indicates the multiplication of corresponding elements;

[0116] The spatial features are extracted by spectral convolution, as shown in formula (9):

[0117] Formula (9)

[0118] in, For the The frequency feature matrix of the layer, W is the adjacency matrix, D is the degree matrix of W, is the convolution kernel;

[0119] The node failure index is calculated through the fully connected layer, as shown in formula (10):

[0120] Formula (10)

[0121] in, The predicted node i The failure index, is the activation function, and is the weight matrix in the fully connected layer, and are the instantaneous amplitude characteristics and instantaneous frequency characteristics of node i respectively;

[0122] S3-2, define the distribution law matching loss function, minimize the difference between the true distribution and the predicted distribution, and use the multi-label classification loss function to optimize the model, as shown in formula (11):

[0123] Formula (11)

[0124] in, The actual node state label vector, is the node state vector predicted by the model, is the loss function value, n is the total number of nodes, is the fault state of node i, is the failure index of node i;

[0125] S3-3, use the constructed distribution feature template library and historical fault data to train the model, use the cross-validation method to evaluate the model performance, divide the training set and the test set, optimize the model parameters with the multi-label classification loss function during the training process, and learn the relationship between the amplitude-frequency characteristics of the nodes and the spatial distribution law.

[0126] It should be noted that by comparing the feature distribution in the template library with the predicted distribution, it is ensured that the model can learn the spatial distribution law of the amplitude-frequency characteristics, thereby improving the accuracy of the prediction. The weight of the node state is introduced into the loss function, and the key nodes are given a larger optimization weight to improve the ability to capture the distribution law.

[0127] Since multiple nodes in the distribution network may fail at the same time, a multi-label classification loss function is designed so that the model can handle complex scenarios where multiple faults occur at the same time. The predicted value of each node is calculated independently, and the global loss is adjusted based on the overall distribution law. During the optimization process, the model gradually narrows the difference between the predicted results and the actual distribution, thereby improving the model's adaptability to complex fault distribution patterns.

[0128] Step S4, collect real-time data after the fault occurs, calculate the matching similarity between the distribution characteristics and the template library, and combine the model reasoning results to comprehensively judge the fault type and specific location;

[0129] Wherein step S4 also includes the following sub-steps:

[0130] S4-1, after a fault occurs, collect the real-time voltage signals of each node in the distribution network, and use the instantaneous amplitude and frequency characteristic data obtained by the deployed distributed acquisition devices. The real-time data is quickly processed by the signal processing module, including noise filtering and abnormal point correction;

[0131] S4-2, distribution pattern matching, calculate the spatial distribution characteristics of real-time data, match it with the distribution template library, determine the fault type and location, calculate the spatial distribution characteristics of real-time data, compare it with the samples in the distribution template library, and match the similarity, as shown in formula (12):

[0132] Formula (12)

[0133] in, is the matching similarity between real-time data and template, is the amplitude-frequency characteristic distribution of real-time data, is the amplitude-frequency characteristic template in the distribution template library;

[0134] S4-3, input the amplitude and frequency characteristic data collected in real time into the distribution law driven graph convolutional network (DG-GCN) model, and generate the fault index of each node through the joint analysis of node characteristics and spatial distribution laws by the model. Combined with the similarity matching results of the distribution feature template library, the specific node and branch location where the fault occurred is comprehensively judged.

[0135] It should be noted that after a fault occurs, the system first generates a fault index for each node through a distribution-driven graph convolutional network (DG-GCN) model to describe the possibility of a node failure. The model uses the real-time amplitude and frequency characteristics of the node, combined with the adjacency relationship in the graph network, and conducts a comprehensive analysis through spatiotemporal feature extraction and distribution comparison modules to output the fault index of each node. The closer the similarity value is to 1, the higher the degree of match between the real-time distribution feature and a certain fault mode in the template library. By combining the fault index and similarity matching results, the specific node where the fault occurs and the location of its branch are comprehensively judged.

[0136] Step S5, by testing the performance of the model in different scenarios, dynamically updating the distribution template library and optimizing the model algorithm structure, the accuracy of fault location is improved.

[0137] Wherein, in step S5, the following sub-steps are also included:

[0138] S5-1, as the distribution network operation data accumulates, the distribution law template library is dynamically updated using the newly collected fault data, and the incremental learning mechanism is used to adapt to the changes in the distribution network operation environment, focusing on generating new distribution templates for new scenarios such as distributed power access, load transfer, and topology adjustment;

[0139] S5-2, testing the positioning accuracy of the model under different fault scenarios and topology change conditions, and testing the real-time response capability of the model in simulation and actual scenarios;

[0140] S5-3, based on the performance evaluation results, the model is optimized and adjusted, including the introduction of a new distribution law mining algorithm to improve the feature extraction capability of complex faults, improving the structure of the graph convolutional network (DG-GCN) to enhance the expressiveness of the model, and adjusting the loss function to better adapt to multi-label classification tasks.

[0141] It should be noted that the access of distributed power sources will change the electrical parameter characteristics and power flow of the original distribution network, causing the fluctuation characteristics of voltage amplitude and frequency to present a new distribution pattern. Real-time data is used to extract the amplitude-frequency characteristics of the nodes after access, and combined with the power access location and its output characteristics, a new distribution template that adapts to the distributed power access scenario is generated.

[0142] During the load transfer process, the power demand and voltage distribution in the distribution network change significantly. By monitoring the dynamic changes of node characteristics during the load transfer process and recalculating the amplitude-frequency change rate between nodes, a distribution pattern template that can adapt to the load transfer scenario is constructed.

[0143] When the topology of the distribution network is adjusted, such as adding new branches or nodes or disconnecting certain lines, the topological relationship between the nodes and the electrical distribution rules will change accordingly. At this time, the adjacency matrix is ​​updated in real time, and the node characteristics under the new topology are remodeled to generate a distribution template suitable for the adjusted topology.

[0144] A smart distribution network fault location system, comprising:

[0145] Data acquisition module, signal processing and feature extraction module, graph network construction module, fault location model module, distribution pattern template library and matching module, optimization module;

[0146] The data acquisition module collects voltage signals in real time at each node of the distribution network through a distributed acquisition device. The collected data includes instantaneous voltage amplitude and frequency characteristics, supports high sampling rate, and is equipped with anti-interference and data loss compensation mechanisms;

[0147] The signal processing and feature extraction module uses an improved ICEEMDAN algorithm to decompose the voltage signal, extract the intrinsic mode function and remove noise, selects the IMF component with strong correlation through the Pearson correlation coefficient for signal reconstruction, applies the Hilbert transform to extract the instantaneous amplitude and frequency characteristics, calculates the amplitude-frequency change rate of the adjacent nodes, and generates the spatial distribution characteristics of the nodes;

[0148] The graph network construction module constructs a graph network based on the amplitude-frequency distribution law, defines node and edge features, introduces a dynamic topology adjustment mechanism, updates the adjacency matrix according to real-time voltage fluctuations, and generates corresponding graph labels for each fault mode based on a template library;

[0149] The fault location model module uses the distribution law driven graph convolution network (DG-GCN) model to extract spatiotemporal features, adds a distribution comparison module based on spatiotemporal convolution, strengthens the recognition of specific distribution laws, and calculates the fault index of the node through the fully connected layer;

[0150] The distribution rule template library and matching module maintains the distribution rule template library, stores the spatial distribution characteristics under different fault modes, calculates the similarity between real-time data and the template, and further confirms the fault type and location through template matching;

[0151] The optimization module dynamically updates the distribution template library to adapt to changes in the operating environment, tests the performance of the model in different scenarios, including dynamic topology and noise interference, optimizes the model structure and algorithm, and improves the accuracy and robustness of fault location.

[0152] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention has various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for locating faults in a smart distribution network, characterized in that: The following steps are involved: Step S1, by deploying distributed acquisition devices at each node, collecting voltage amplitude and frequency signals in real time, extracting instantaneous features and calculating spatial distribution laws, and generating a feature template library; Step S2, combining the topological structure and amplitude-frequency distribution characteristics of the distribution network, dynamically adjusting the adjacency matrix, constructing a graph network that can reflect the operating status in real time, and generating node status labels that match the fault mode; Step S3, designing a fault location model based on a graph convolutional network, introducing a spatiotemporal convolution and distribution law comparison module, using node features and spatial distribution laws to extract fault modes and achieve node fault state prediction; Step S4, collect real-time data after the fault occurs, calculate the matching similarity between the distribution characteristics and the template library, and combine the model reasoning results to comprehensively judge the fault type and specific location; Step S5, by testing the performance of the model in different scenarios, dynamically updating the distribution template library and optimizing the model algorithm structure, so as to improve the accuracy of fault location; Wherein step S1 also includes the following sub-steps: S1-1, by deploying distributed acquisition devices at each node, real-time voltage amplitude and frequency data are collected, equipped with anti-interference modules to reduce the impact of external noise, and equipped with data loss compensation mechanisms to ensure that effective features can still be obtained when the data is incomplete; S1-2, using the improved signal decomposition algorithm ICEEMDAN, decompose the voltage signal to obtain the intrinsic mode functions IMFs. The decomposed voltage signal is expressed as the superposition of each order IMFs and the residual term, as shown in formula (1): Formula (1) in, is the voltage signal, are the i-th order IMFs, is the residual term; Since the order and frequency range of the decomposed IMFs are not fixed, IMFs cannot be directly used as fault features in the fault location model. The Pearson correlation coefficient is used to select IMFs with a strong correlation with the original signal, as shown in formula (2): Formula (2) in, is the Pearson correlation coefficient between the i-th order IMF and the original signal, is the covariance operator, is the covariance of IMF and original signal, and Respectively and The standard deviation of The instantaneous amplitude and instantaneous frequency characteristics of the fault voltage are extracted by Hilbert transform, as shown in formula (3): Formula (3) in, is the reconstructed voltage signal, and are the instantaneous amplitude and instantaneous frequency respectively; S1-3, using the amplitude and frequency fluctuation amplitude of the adjacent nodes, calculate the spatial distribution characteristics, analyze the characteristic spatial distribution patterns corresponding to different fault types and locations, generate a distribution feature template library, and define the amplitude-frequency feature change rate of each node, as shown in formula (4): Formula (4) in, For Node The amplitude-frequency characteristic change rate, For Node The amplitude and frequency of For Node The amplitude and frequency of For Node geographical or topological distance; Wherein step S2 also includes the following sub-steps: S2-1, distribution pattern mapping, combines the spatial distribution characteristics with the distribution network topology to construct a feature map based on the amplitude-frequency distribution law. Each node contains the amplitude, frequency and fluctuation characteristics of its adjacent nodes, as shown in formula (5): Formula (5) in, is the instantaneous amplitude matrix, is the instantaneous frequency matrix, is the adjacency matrix, which describes the topological relationship between nodes in the graph; S2-2, introduces a dynamic topology adjustment mechanism, which dynamically adjusts the nodes and edge relationships of the graph network by monitoring the changes in the operating topology structure and electrical parameters of the distribution network, so that it can reflect the current operating status; S2-3, graph label generation, based on the feature template library of distribution rules, generate corresponding graph labels for each fault mode, and mark the fault status of the nodes in the graph according to the distribution feature template , generate state labels, as shown in formula (6): Formula (6) in, For Node i Status label, 1 means normal and 0 means fault.

2. A smart distribution network fault location method according to claim 1, characterized in that: Wherein step S3 also includes the following sub-steps: S3-1, model structure definition, design a distribution law driven graph convolutional network (DG-GCN), combine node features and spatial distribution features to extract fault signal patterns, and introduce a distribution comparison module based on spatiotemporal convolution to enhance the recognition of specific distribution laws; The temporal features are extracted through one-dimensional convolution, as shown in equations (7) and (8): Formula (7) Formula (8) in, and Respectively The frequency and amplitude feature matrices of the layers, and Respectively The frequency and amplitude feature matrices of the layer; and These are two one-dimensional convolution kernel functions used to extract time domain features; is the activation function; Indicates the multiplication of corresponding elements; The spatial features are extracted by spectral convolution, as shown in formula (9): Formula (9) in, For the The frequency feature matrix of the layer, W is the adjacency matrix, D is the degree matrix of W, is the convolution kernel; The node failure index is calculated through the fully connected layer, as shown in formula (10): Formula (10) in, The predicted node i The failure index, is the activation function, and is the weight matrix in the fully connected layer, and are the instantaneous amplitude characteristics and instantaneous frequency characteristics of node i respectively; S3-2, define the distribution law matching loss function, minimize the difference between the true distribution and the predicted distribution, and use the multi-label classification loss function to optimize the model, as shown in formula (11): Formula (11) in, The actual node state label vector, is the node state vector predicted by the model, is the loss function value, n is the total number of nodes, is the fault state of node i, is the failure index of node i; S3-3, use the constructed distribution feature template library and historical fault data to train the model, use the cross-validation method to evaluate the model performance, divide the training set and the test set, optimize the model parameters with the multi-label classification loss function during the training process, and learn the relationship between the amplitude-frequency characteristics of the nodes and the spatial distribution law.

3. A smart distribution network fault location method according to claim 1, characterized in that: Wherein step S4 also includes the following sub-steps: S4-1, after a fault occurs, collect the real-time voltage signals of each node in the distribution network, and use the instantaneous amplitude and frequency characteristic data obtained by the deployed distributed acquisition devices. The real-time data is quickly processed by the signal processing module, including noise filtering and abnormal point correction; S4-2, distribution pattern matching, calculate the spatial distribution characteristics of real-time data, match it with the distribution template library, determine the fault type and location, calculate the spatial distribution characteristics of real-time data, compare it with the samples in the distribution template library, and match the similarity, as shown in formula (12): Formula (12) in, is the matching similarity between real-time data and template, is the amplitude-frequency characteristic distribution of real-time data, is the amplitude-frequency characteristic template in the distribution template library; S4-3, input the amplitude and frequency characteristic data collected in real time into the distribution law driven graph convolutional network (DG-GCN) model, and generate the fault index of each node through the joint analysis of node characteristics and spatial distribution laws by the model. Combined with the similarity matching results of the distribution feature template library, the specific node and branch location where the fault occurred is comprehensively judged.

4. A smart distribution network fault location method according to claim 1, characterized in that: Wherein, in step S5, the following sub-steps are also included: S5-1, as the distribution network operation data accumulates, the distribution law template library is dynamically updated using the newly collected fault data, and the incremental learning mechanism is used to adapt to the changes in the distribution network operation environment, focusing on generating new distribution templates for new scenarios such as distributed power access, load transfer, and topology adjustment; S5-2, testing the positioning accuracy of the model under different fault scenarios and topology change conditions, and testing the real-time response capability of the model in simulation and actual scenarios; S5-3, based on the performance evaluation results, the model is optimized and adjusted, including the introduction of a new distribution law mining algorithm to improve the feature extraction capability of complex faults, improving the structure of the graph convolutional network (DG-GCN) to enhance the expressiveness of the model, and adjusting the loss function to better adapt to multi-label classification tasks.

5. A smart distribution network fault location system, applied to any of the location methods described in claims 1-4, characterized in that: include: Data acquisition module, signal processing and feature extraction module, graph network construction module, fault location model module, distribution pattern template library and matching module, optimization module; The data acquisition module collects voltage signals in real time at each node of the distribution network through a distributed acquisition device. The collected data includes instantaneous voltage amplitude and frequency characteristics, supports high sampling rate, and is equipped with anti-interference and data loss compensation mechanisms; The signal processing and feature extraction module uses an improved ICEEMDAN algorithm to decompose the voltage signal, extract the intrinsic mode function and remove noise, selects the IMF component with strong correlation through the Pearson correlation coefficient for signal reconstruction, applies the Hilbert transform to extract the instantaneous amplitude and frequency characteristics, calculates the amplitude-frequency change rate of the adjacent nodes, and generates the spatial distribution characteristics of the nodes; The graph network construction module constructs a graph network based on the amplitude-frequency distribution law, defines node and edge features, introduces a dynamic topology adjustment mechanism, updates the adjacency matrix according to real-time voltage fluctuations, and generates corresponding graph labels for each fault mode based on a template library; The fault location model module uses the distribution law driven graph convolution network (DG-GCN) model to extract spatiotemporal features, adds a distribution comparison module based on spatiotemporal convolution, strengthens the recognition of specific distribution laws, and calculates the fault index of the node through the fully connected layer; The distribution rule template library and matching module maintains the distribution rule template library, stores the spatial distribution characteristics under different fault modes, calculates the similarity between real-time data and the template, and further confirms the fault type and location through template matching; The optimization module dynamically updates the distribution template library to adapt to changes in the operating environment, tests the performance of the model in different scenarios, including dynamic topology and noise interference, optimizes the model structure and algorithm, and improves the accuracy and robustness of fault location.

Images