Power grid intelligent power transmission line fault positioning system

Abstract

The invention belongs to the technical field of power grid power transmission, and discloses a power grid intelligent power transmission line fault positioning system, which comprises the steps of collecting power operation data, spatial data and environmental data; correcting the power operation data based on the constructed power correction model to obtain corrected power data; performing topology establishment on the corrected power data and the spatial data by adopting a neural network algorithm to obtain a power grid topological graph, and performing graph decomposition on the power grid topological graph by adopting an ant colony algorithm to obtain B topological sub-graphs; constructing a fault prediction model for predicting a fault type; performing fault prediction on the B topology sub-graphs through a fault prediction model, marking the topology sub-graphs with predicted fault types as fault sub-graphs, positioning fault points in the fault sub-graphs through a particle swarm algorithm, and determining final fault point positions; and the positioning speed of the fault point location is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid transmission, and more specifically, to a power grid intelligent transmission line fault location system. Background Art

[0002] The patent application with the publication number CN118209817A discloses a method and system for identifying transmission line faults based on environmental features; the method includes: real-time monitoring and data preprocessing of multiple first transmission line nodes in the power system network to obtain voltage waveform features and current waveform features; abnormal node identification to obtain multiple second transmission line nodes; data fusion to obtain multi-dimensional fusion data; feature extraction to obtain initial multi-dimensional fusion features and feature dimension reduction processing to obtain target multi-dimensional fusion features; transmission line fault identification through a transmission line fault identification model, and outputting the transmission line fault type of each second transmission line node; determining the initial transmission line fault handling plan for each second transmission line node and performing global optimization analysis to obtain the global transmission line fault handling plan of the power system network, thereby improving the accuracy of transmission line fault identification.

[0003] Current fault detection is mainly based on single electrical signals, such as current mutations and voltage fluctuations. However, in a changing natural environment, these single signals are easily interfered by external factors, resulting in false alarms or missed alarms. There is a lack of multi-dimensional sensor data fusion to comprehensively judge the fault type and location. Although some transmission lines have begun to use digital equipment for data monitoring and collection, existing data processing systems mostly use traditional fault analysis methods, such as Fourier transform or simple statistical analysis. These methods can often only process limited input data. When faced with massive data collected by large-scale sensor networks, the analysis efficiency is low and it is impossible to make accurate fault location judgments in a short time. In addition, the limitations of data processing capabilities also lead to inaccurate identification of complex fault types.

[0004] In view of this, the present invention proposes a power grid intelligent transmission line fault location system to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A power grid intelligent transmission line fault location system, comprising: a data acquisition module: collecting power operation data, spatial data and environmental data;

[0006] Data correction module: corrects the power operation data based on the constructed power correction model to obtain corrected power data;

[0007] Graph construction module: including a graph construction unit and a graph decomposition unit. The graph construction unit uses a neural network algorithm to establish topology for the corrected power data and spatial data to obtain a power grid topology map. The graph decomposition unit decomposes the power grid topology map to obtain B topology sub-graphs.

[0008] Fault model building module: collects F groups of historical power fault data, and builds a fault prediction model for predicting fault types based on the historical power fault data;

[0009] Fault location module: Fault prediction is performed on B topological subgraphs through the fault prediction model, and the topological subgraph with predicted fault type is marked as the fault subgraph. The fault point in the fault subgraph is located through the particle swarm algorithm to determine the final fault point.

[0010] Furthermore, the power operation data includes: electric field data, magnetic field data, mechanical stress data and vibration data; the spatial data includes: accuracy, latitude and proximity; and the environmental data includes: temperature, wind force and humidity.

[0011] Furthermore, the method for obtaining the corrected power data includes:

[0012] Based on the constructed power correction model, the power operation data and environmental data are used as the power correction model for prediction to obtain corrected power data. The corrected power data includes corrected electric field data, corrected magnetic field data, corrected mechanical stress data and corrected vibration data.

[0013] Furthermore, the grid topology diagram is constructed in the following manner:

[0014] The precision and latitude in each corrected power data and spatial data are used as the attributes of a node to construct a power grid node. The neighboring relationship in the spatial node is used as the basis for the connection between nodes to construct a power grid node matrix. The value in the power grid node matrix is ​​determined based on the neighboring relationship. When the neighboring relationship is "yes", the value in the power grid node matrix is ​​1, and when the neighboring relationship is "no", the value in the power grid node matrix is ​​1; an associated index set of each node is constructed based on the power grid node matrix. The method of constructing the associated index set of each node includes: based on the power grid node JD represented by each row in the power grid node matrix, the power grid node with a value of 1 in each row of the matrix is ​​the associated node of the power grid node JD, and the indexes of all associated nodes constitute an associated index set; the attribute of each node is used as the input of the graph convolutional network input layer, and the graph convolutional network is used to perform feature aggregation on each node to obtain node embedding features, and the node embedding features are used as topological nodes. The neighboring relationship in the spatial node is used as the basis for the connection between nodes. The similarity of the connected topological nodes is calculated by the cosine similarity formula to obtain the node similarity value, and the node similarity value is used as the weight of the edge to construct a power grid topology map.

[0015] Furthermore, the formula for feature aggregation is: in, represents the node embedding feature of the vth power grid node after convolution by the i-th convolutional layer, N v is the associated index set of the vth power grid node, d v represents the degree of the vth grid node, d u represents the degree of the u-th grid node, W l represents the weight matrix of the i-th convolutional layer, b i represents the bias term of the i-th convolutional layer, and σ represents the nonlinear activation function.

[0016] Furthermore, the method of decomposing the power grid topology diagram includes:

[0017] Starting from each topological node in the power grid topology map, the breadth-first algorithm is used to search for the shortest path between each pair of topological nodes in the power grid topology map; based on all the shortest paths, G ants are initialized, where the G value is the total number of shortest paths, and a shortest path is assigned to each ant. Each ant represents the execution of a shortest path, and each ant is assigned to a different topological node in the power grid topology map; pheromone is initialized: the pheromone concentration is initialized for each edge, and the initial pheromone concentration is assigned to each ant; each ant starts from the starting topological node and moves in the power grid topology map with the assigned shortest path as the basis for each ant's movement. The pheromone concentration of all edges in the shortest path assigned to each ant is updated according to the weight of all edges in the shortest path, and the formula for updating the pheromone concentration is: Among them, τ m,n represents the updated value of the pheromone concentration of the edge connecting the mth node and the nth node in the power grid topology graph, ρ represents the pheromone volatility coefficient, S represents the initial pheromone concentration assigned to the ants, which is used to adjust the pheromone increment, and Sum represents the weighted sum of all edges through which the selected path passes; when all ants reach the end of the path, the edge with the highest pheromone concentration is selected as the intermediate edge according to the pheromone concentration of each edge in the power grid topology graph, and the intermediate edge is deleted from the power grid topology graph to form a new power grid topology graph. When the power grid topology graph is divided into N unconnected topological subgraphs, each topological subgraph is evaluated for module to obtain the modularity of each topological subgraph, and this is repeated until the modularity value of each topological subgraph no longer changes, and B topological subgraphs are obtained.

[0018] Furthermore, the formula for module evaluation is: Among them, Q j represents the modularity of the jth topological subgraph, a j represents the total number of edges in the jth topological subgraph, represents the degree of the cth topological node in the jth topological subgraph, represents the degree of the dth topological node in the jth topological subgraph, q represents the weight of the edge connecting the cth topological node and the dth topological node, SIM j represents the combination of all topological nodes with connected edges in the jth topological subgraph, c and d represent the indexes of the topological nodes in the jth topological subgraph, and c≠d.

[0019] Furthermore, the fault prediction model is constructed in the following manner:

[0020] F groups of historical power fault data are collected, and the historical power fault data include fault categories, electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data. Based on the fault category, the historical power fault data with the same fault category are divided into a group to obtain R groups of fault data groups, and anomaly detection is performed on the data in each group of fault data groups. The method of performing anomaly detection includes: the electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data in each group of data in the fault data group are combined into a C×1 fault vector, where C represents the total number of fault data categories; the similarity value between each fault vector in the fault vector and other fault vectors is calculated by a similarity calculation formula, the calculated similarity values ​​are averaged to obtain the average similarity of each fault vector, and the zero score algorithm is used to calculate the similarity of each fault vector based on the average similarity. The discreteness of the quantity is determined, a discreteness threshold is preset, the power fault data with a discreteness greater than or equal to the discreteness threshold is marked as abnormal data, and the historical power fault data marked as abnormal data in each group of fault data is eliminated to obtain a valid fault data group; the valid fault data group is used as a training sample set, the GRU model is used as an initial model, the GRU model is trained using the training sample set, the electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data in the valid fault data group are used as input data of the fault prediction model, and the predicted fault category is used as output data of the fault prediction model; minimizing the error between the actual fault category and the predicted fault category is used as the training goal, the recall rate function is used as the loss function of the stability prediction model, and when the loss function reaches convergence, the training is stopped to obtain the fault prediction model.

[0021] Furthermore, the method for determining the fault subgraph includes:

[0022] According to the power operation data of each topological node in the topological subgraph, the global representative data of each topological subgraph is calculated. The global representative data includes the mean of the electric field data, the mean of the magnetic field data, the mean of the mechanical stress data and the mean of the vibration data. The global representative data is used as the input of the fault prediction model to perform prediction, and the predicted fault category is obtained. The topological subgraph corresponding to the global representative data is marked as the predicted fault category, and all marked topological subgraphs are recorded as fault subgraphs.

[0023] Furthermore, the method of locating the fault point in the fault subgraph includes:

[0024] Each topological node in the fault subgraph is regarded as a particle. Each particle represents a possible fault point. Based on the number of topological nodes in the fault subgraph, the number of particle swarms is initialized to half of the total number of topological nodes. The number of particles in each particle swarm is initialized to num. The initial speed of each particle is initialized to a very small positive value to ensure that the initial state has a certain randomness. The fitness function formula of the topological node is defined as Representative The fitness value of a topological node, The calculation formula is: Among them, μ, α, β and γ represent weight coefficients, and the weight coefficients satisfy the weight condition μ+α+β+γ=1. Representative Corrected electric field data of topological nodes, Representative Corrected magnetic field data of topological nodes, Representative Corrected mechanical stress data of topological nodes, Representative The corrected vibration data of a topological node is obtained, and the fitness values ​​of all connected topological nodes at the particle's location are calculated based on the fitness value formula. The topological node with the largest fitness value of all connected topological nodes is recorded as the update node. The position of the particle is updated based on the update node. When the fitness value of the update node is greater than the fitness value of the particle's current position, the position of the updated particle is the update node. When the fitness value of the update node is less than or equal to the fitness value of the particle's current position, the particle no longer moves. When the fitness values ​​of all particle positions no longer change, the topological node corresponding to the position where the particles are most concentrated is the fault point.

[0025] The technical effects and advantages of the present invention are as follows:

[0026] The present invention improves the correction accuracy of power operation data by training based on the difference between historical measured data and predicted data, and optimizing the obtained power correction model by minimizing the loss function; the graph decomposition process adopts the ant colony algorithm, which effectively reduces the computational complexity of complex power grids and facilitates efficient fault analysis of local subgraphs; the GRU model is used to predict fault categories, capture the time dependency in historical fault data, and thus improve the accuracy of fault prediction; anomaly detection and elimination of historical power fault data ensures that the data used for model training is more effective and accurate, thereby improving the robustness and reliability of the fault prediction model; the particle swarm algorithm is used to simulate the group intelligent behavior in nature, and efficient search and positioning are performed in the power grid topology subgraph, ensuring the accuracy of the final positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of a power grid intelligent transmission line fault location system of the present invention;

[0028] Figure 2 A schematic diagram of a method for locating faults in an intelligent power transmission line of a power grid according to the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 creative work are within the scope of protection of the present invention.

[0030] Example 1

[0031] See also Figure 1 As shown, the intelligent power transmission line fault location system of the power grid described in this embodiment includes: a data acquisition module: collecting power operation data, spatial data and environmental data;

[0032] Data correction module: corrects the power operation data based on the constructed power correction model to obtain corrected power data;

[0033] Graph construction module: including a graph construction unit and a graph decomposition unit. The graph construction unit uses a neural network algorithm to establish topology for the corrected power data and spatial data to obtain a power grid topology map. The graph decomposition unit decomposes the power grid topology map to obtain B topology sub-graphs.

[0034] Fault model building module: collects F groups of historical power fault data, and builds a fault prediction model for predicting fault types based on the historical power fault data;

[0035] Fault location module: Fault prediction is performed on B topological subgraphs through the fault prediction model, and the topological subgraph with predicted fault type is marked as a fault subgraph. The fault point in the fault subgraph is located through the particle swarm algorithm to determine the final fault point.

[0036] Each module is connected by wired and / or wireless means to achieve data transmission between modules;

[0037] Power operation data includes: electric field data, magnetic field data, mechanical stress data and vibration data; spatial data includes: accuracy, latitude and proximity; environmental data includes: temperature, wind speed and humidity; among them, electric field data is the field generated by charged bodies or currents, and its distribution will change with the line status and environmental conditions. Electric field data is obtained through electric field sensors, usually capacitive or inductive sensors, and the sensors are placed at a certain distance to avoid high voltage from causing damage to the equipment; magnetic field data is generated by current, and the magnetic field strength reflects the current situation in the line. Currents of different phases form different magnetic field distributions. Magnetic field data is obtained through electric field sensors, usually capacitive or inductive sensors, and the sensors are placed at a certain distance to avoid high voltage from causing damage to the equipment. The mechanical stress data refers to the internal force acting on the object. The mechanical stress is mainly caused by the gravity of the wire itself, wind force and electromagnetic force. The mechanical stress data is obtained by the fiber Bragg grating sensor. The fiber Bragg grating sensor can measure stress by using the fiber Bragg grating effect. It has strong anti-interference ability and is suitable for harsh environments. The vibration data is obtained by the acceleration sensor, which is usually a piezoelectric accelerometer that can sense the vibration frequency and amplitude. The proximity relationship is the connection relationship between each substation, which is a definite value, "yes" or "no". "yes" means that the substations are connected, and "no" means that the substations are not connected.

[0038] Collecting a group of historical power operation data and historical environmental data, the historical power operation data includes: historical measured power data and historical predicted power data. Based on the historical power operation data and historical environmental data, the historical predicted power data and historical environmental data are used as inputs of the power correction model, the predicted corrected power data are used as outputs of the power correction model, and the historical measured power data are used as prediction targets of the power correction model for model training. The power correction model is constructed. The power correction model is a linear regression model. The target loss function of the power correction model is a mean square error function. Minimizing the value of the target loss function is used as the training target, and a soil viscosity prediction model with the minimum value of the target loss function is obtained.

[0039] Based on the constructed power correction model, the power operation data and environmental data are used as the power correction model for prediction to obtain corrected power data, which includes corrected electric field data, corrected magnetic field data, corrected mechanical stress data and corrected vibration data;

[0040] The data categories included in the historical measured power data and the historical estimated power data are consistent, both of which are historical electric field data, historical magnetic field data, historical mechanical stress data and historical vibration data, and are consistent with the data categories and types included in the power operation data; the data categories and types included in the historical environmental data and the environmental data are consistent, and the historical environmental data include historical temperature, historical wind speed and historical humidity;

[0041] The historical predicted power data are data obtained under normal temperature, normal wind speed and normal humidity, and the historical measured power data are data obtained under abnormal temperature, abnormal wind speed and abnormal humidity. The historical predicted power data and the historical measured power data are obtained by electric field sensors, Hall effect sensors, fiber Bragg grating sensors and acceleration sensors respectively;

[0042] The precision and latitude in each corrected power data and spatial data are used as the attributes of a node to construct a power grid node. The neighboring relationship in the spatial node is used as the basis for the connection between nodes to construct a power grid node matrix. The value in the power grid node matrix is ​​determined based on the neighboring relationship. When the neighboring relationship is "yes", the value in the power grid node matrix is ​​1, and when the neighboring relationship is "no", the value in the power grid node matrix is ​​1; based on the power grid node matrix, an associated index set of each node is constructed. The method of constructing the associated index set of each node includes: based on the power grid node JD represented by each row in the power grid node matrix, the power grid node with a value of 1 in each row of the matrix is ​​the associated node of the power grid node JD, and the indexes of all associated nodes constitute an associated index set; the attribute of each node is used as the input of the graph convolutional network input layer, and the graph convolutional network is used to perform feature aggregation on each node. The formula for feature aggregation is:

[0043] in, represents the node embedding feature of the vth power grid node after convolution by the i-th convolutional layer. The node embedding feature is the low-dimensional vector of the power grid node after mapping by the convolutional layer, which is convenient for efficient calculation and analysis. v is the associated index set of the vth power grid node, d v represents the degree of the vth grid node (i.e., the number of edges connected to the vth grid node), d u represents the degree of the u-th grid node, W l represents the weight matrix of the i-th convolutional layer, which is used to extract local features of the input data, b i represents the bias term of the i-th convolutional layer, which is used to adjust the output of the convolution operation. σ represents a nonlinear activation function, such as the ReLU function. Represents the input features of the i-th convolutional layer, where i represents the index of the convolutional layer;

[0044] The node embedding feature is used as the topological node, and the neighboring relationship in the spatial node is used as the basis for the connection between nodes. The similarity of the connected topological nodes is calculated by the cosine similarity formula to obtain the node similarity value. The power grid topology graph is constructed with the node similarity value as the edge weight. The number of vertices in the power grid topology graph is the number of point embedding features, and the number of connected edges in the power grid topology graph is the number of values ​​1 in the power grid node matrix.

[0045] The power grid topology graph is decomposed, and the graph decomposition method includes: starting from each topological node in the power grid topology graph, using the breadth-first algorithm to search for the shortest path between each pair of topological nodes in the power grid topology graph; based on all the shortest paths, G ants are initialized, where the G value is the total number of shortest paths, a shortest path is assigned to each ant, each ant represents the execution of a shortest path, and each ant is assigned to a different topological node in the power grid topology graph; pheromone is initialized: the pheromone concentration is initialized for each edge, and an initial pheromone concentration is assigned to each ant; each ant starts from the starting topological node and moves in the power grid topology graph with the assigned shortest path as the basis for each ant's movement, and the pheromone concentration of all edges in the shortest path corresponding to each ant is updated, and the formula for updating the pheromone concentration is: Among them, τ m,n represents the updated value of the pheromone concentration of the edge connecting the mth node and the nth node in the power grid topology graph, ρ represents the pheromone volatility coefficient, S represents the initial pheromone concentration assigned to the ant, which is used to adjust the pheromone increment, and Sum represents the weighted sum of all edges through which the selected path passes; when all ants reach the end of the path, the edge with the highest pheromone concentration is selected as the intermediate edge according to the pheromone concentration of each edge in the power grid topology graph, and the intermediate edge is deleted from the power grid topology graph to form a new power grid topology graph. When the power grid topology graph is divided into N unconnected topological subgraphs, each topological subgraph is evaluated by module, N≥2; the formula for module evaluation is: Among them, Q j represents the modularity of the jth topological subgraph, a j represents the total number of edges in the jth topological subgraph, represents the degree of the cth topological node in the jth topological subgraph, represents the degree of the dth topological node in the jth topological subgraph, q represents the weight of the edge connecting the cth topological node and the dth topological node, SIM j represents the combination of all topological nodes with connected edges in the jth topological subgraph, c and d represent the indexes of the topological nodes in the jth topological subgraph, and c≠d; repeat until the modularity value of each topological subgraph no longer changes, and obtain B topological subgraphs;

[0046] F groups of historical power fault data are collected, and the historical power fault data include fault categories, electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data. Based on the fault category, the historical power fault data with the same fault category are divided into a group to obtain R groups of fault data groups, and anomaly detection is performed on the data in each group of fault data groups. The method of performing anomaly detection includes: the electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data in each group of data in the fault data group are combined into a C×1 fault vector, where C represents the total number of fault data categories; the similarity value between each fault vector in the fault vector and other fault vectors is calculated by a similarity calculation formula, the calculated similarity values ​​are averaged to obtain the average similarity of each fault vector, and the zero score algorithm is used to calculate the similarity of each fault vector based on the average similarity. The discreteness of the quantity is determined, a discreteness threshold is preset, and the power fault data with a discreteness greater than or equal to the discreteness threshold is marked as abnormal data, and the historical power fault data marked as abnormal data in each group of fault data is eliminated to obtain a valid fault data group; the valid fault data group is used as a training sample set, the GRU model is used as an initial model, and the GRU model is trained using the training sample set, and the electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data in the valid fault data group are used as input data of the fault prediction model, and the predicted fault category is used as output data of the fault prediction model; minimizing the error between the actual fault category and the predicted fault category is used as the training goal, and the recall rate function is used as the loss function of the stability prediction model. When the loss function reaches convergence, the training is stopped to obtain the fault prediction model;

[0047] Fault categories are currently known fault conditions in power grid transmission lines, for example: Short circuit fault: refers to conductive connection between phases or between phases and ground, resulting in rapid increase in current. Common short circuit faults include: single-phase ground short circuit (single-phase to ground fault), two-phase short circuit (two-phase short circuit), two-phase ground short circuit, three-phase short circuit and three-phase ground short circuit; Open circuit fault: refers to disconnection of conductors in transmission lines, usually caused by mechanical damage or equipment aging, resulting in current failure to flow normally; Ground fault: refers to accidental contact between the conductor and the ground, usually accompanied by high current leakage to the ground, which may cause fire or equipment damage in severe cases;

[0048] According to the power operation data of each topological node in the topological subgraph, the global representative data of each topological subgraph is calculated. The global representative data includes the mean of the electric field data, the mean of the magnetic field data, the mean of the mechanical stress data and the mean of the vibration data. The global representative data is used as the input of the fault prediction model to perform prediction, and the predicted fault category is obtained. The topological subgraph corresponding to the global representative data is marked as the predicted fault category, and all the marked topological subgraphs are recorded as fault subgraphs.

[0049] Each topological node in the fault subgraph is regarded as a particle. Each particle represents a possible fault point. Based on the number of topological nodes in the fault subgraph, the number of particle swarms is initialized to half of the total number of topological nodes. The number of particles in each particle swarm is initialized to num. The initial speed of each particle is initialized to a very small positive value to ensure that the initial state has a certain randomness. The fitness function formula of the topological node is defined as Representative The fitness value of a topological node, The calculation formula is: Among them, μ, α, β and γ represent weight coefficients, and the weight coefficients satisfy the weight condition μ+α+β+γ=1. Representative Corrected electric field data of topological nodes, Representative Corrected magnetic field data of topological nodes, Representative Corrected mechanical stress data of topological nodes, Representative The corrected vibration data of a topological node is used to calculate the fitness values ​​of all connected topological nodes at the particle's location based on the fitness value formula. The topological node with the largest fitness value among all connected topological nodes is recorded as the update node. The position of the particle is updated based on the update node. When the fitness value of the update node is greater than the fitness value of the particle's current position, the position of the updated particle is the update node. When the fitness value of the update node is less than or equal to the fitness value of the particle's current position, the particle no longer moves. When the fitness values ​​of all particle positions no longer change, the topological node corresponding to the position where the particles are most concentrated is the fault point. The number of particles in the particle swarm is determined by multiplying the data latitude of the topological node in the fault subgraph by the particle constant. The particle constant is a positive integer, and the value range of the particle constant is [1,5].

[0050] This embodiment improves the correction accuracy of power operation data by training based on the difference between historical measured data and predicted data, and optimizing the obtained power correction model by minimizing the loss function; the graph decomposition process adopts the ant colony algorithm, which effectively reduces the computational complexity of complex power grids and facilitates efficient fault analysis of local subgraphs; the GRU model is used to predict fault categories, capture the time dependency in historical fault data, and thus improve the accuracy of fault prediction; anomaly detection and elimination of historical power fault data ensures that the data used for model training is more effective and accurate, thereby improving the robustness and reliability of the fault prediction model; the particle swarm algorithm is used to simulate the group intelligence behavior in nature, and efficient search and positioning are performed in the power grid topology subgraph, ensuring the accuracy of the final positioning.

[0051] Example 2

[0052] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, and a method for locating faults of intelligent power transmission lines in a power grid is provided, including:

[0053] S1. Collect power operation data, spatial data and environmental data;

[0054] S2. Correct the power operation data based on the constructed power correction model to obtain corrected power data;

[0055] S3, using a neural network algorithm to establish topology for the corrected power data and spatial data, obtaining a power grid topology map, performing graph decomposition on the power grid topology map, and obtaining B topology sub-maps;

[0056] S4, collecting F groups of historical power failure data, and building a fault prediction model for predicting the fault type based on the historical power failure data;

[0057] S5. Fault prediction is performed on B topological subgraphs through the fault prediction model, and the topological subgraph with the predicted fault type is marked as a fault subgraph. The fault point in the fault subgraph is located through the particle swarm algorithm to determine the final fault point.

[0058] Example 3

[0059] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned intelligent power transmission line fault location system for a power grid is implemented.

[0060] Since the electronic device introduced in this embodiment is an electronic device used to implement a power grid intelligent transmission line fault location system in the embodiment of this application, based on the power grid intelligent transmission line fault location system introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the power grid intelligent transmission line fault location system in the embodiment of this application, it belongs to the scope of protection of this application.

[0061] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0062] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A power grid intelligent transmission line fault location system, characterized in that: include: Data acquisition module: collects power operation data, spatial data and environmental data; Data correction module: corrects the power operation data based on the constructed power correction model to obtain corrected power data; Graph construction module: including a graph construction unit and a graph decomposition unit. The graph construction unit uses a neural network algorithm to establish topology for the corrected power data and spatial data to obtain a power grid topology map. The graph decomposition unit decomposes the power grid topology map to obtain B topology sub-graphs. Fault model building module: collects F groups of historical power fault data, and builds a fault prediction model for predicting fault types based on the historical power fault data; Fault location module: Fault prediction is performed on B topological subgraphs through the fault prediction model, and the topological subgraph with predicted fault type is marked as the fault subgraph. The fault point in the fault subgraph is located through the particle swarm algorithm to determine the final fault point.

2. A power grid intelligent transmission line fault location system according to claim 1, characterized in that: The power operation data includes: electric field data, magnetic field data, mechanical stress data and vibration data; the spatial data includes: accuracy, latitude and proximity; the environmental data includes: temperature, wind force and humidity.

3. A power grid intelligent transmission line fault location system according to claim 2, characterized in that: The method for obtaining the corrected power data includes: Based on the constructed power correction model, the power operation data and environmental data are used as the power correction model for prediction to obtain corrected power data. The corrected power data includes corrected electric field data, corrected magnetic field data, corrected mechanical stress data and corrected vibration data.

4. A power grid intelligent transmission line fault location system according to claim 3, characterized in that: The construction method of the power grid topology diagram includes: The precision and latitude in each corrected power data and spatial data are used as the attributes of a node to construct a power grid node. The neighboring relationship in the spatial node is used as the basis for the connection between nodes to construct a power grid node matrix. The value in the power grid node matrix is ​​determined based on the neighboring relationship. When the neighboring relationship is "yes", the value in the power grid node matrix is ​​1, and when the neighboring relationship is "no", the value in the power grid node matrix is ​​1; the associated index set of each node is constructed based on the power grid node matrix. The method of constructing the associated index set of each node includes: based on the power grid node JD represented by each row in the power grid node matrix, the power grid node with a value of 1 in each row of the matrix is ​​the associated node of the power grid node JD, and the indexes of all associated nodes constitute the associated index set; the attribute of each node is used as the input of the graph convolutional network input layer, and the graph convolutional network is used to perform feature aggregation on each node to obtain node embedding features, and the node embedding features are used as topological nodes. The neighboring relationship in the spatial node is used as the basis for the connection between nodes. The similarity of the connected topological nodes is calculated by the cosine similarity formula to obtain the node similarity value, and the node similarity value is used as the weight of the edge to construct the power grid topology map.

5. A power grid intelligent transmission line fault location system according to claim 4, characterized in that: The formula for feature aggregation is: in, represents the node embedding feature of the vth power grid node after convolution by the i-th convolutional layer, N v is the associated index set of the vth power grid node, d v represents the degree of the vth grid node, d u represents the degree of the u-th grid node, W l represents the weight matrix of the i-th convolutional layer, b i represents the bias term of the i-th convolutional layer, and σ represents the nonlinear activation function.

6. A power grid intelligent transmission line fault location system according to claim 5, characterized in that: The method of decomposing the power grid topology diagram includes: Starting from each topological node in the power grid topology map, the breadth-first algorithm is used to search for the shortest path between each pair of topological nodes in the power grid topology map; based on all the shortest paths, G ants are initialized, where the G value is the total number of shortest paths, and a shortest path is assigned to each ant. Each ant represents the execution of a shortest path, and each ant is assigned to a different topological node in the power grid topology map; pheromone is initialized: the pheromone concentration is initialized for each edge, and the initial pheromone concentration is assigned to each ant; each ant starts from the starting topological node and moves in the power grid topology map with the assigned shortest path as the basis for each ant's movement. The pheromone concentration of all edges in the shortest path assigned to each ant is updated according to the weight of all edges in the shortest path, and the formula for updating the pheromone concentration is: Among them, τ m,n represents the updated value of the pheromone concentration of the edge connecting the mth node and the nth node in the power grid topology graph, ρ represents the pheromone volatility coefficient, S represents the initial pheromone concentration assigned to the ants, which is used to adjust the pheromone increment, and Sum represents the weighted sum of all edges through which the selected path passes; when all ants reach the end of the path, the edge with the highest pheromone concentration is selected as the intermediate edge according to the pheromone concentration of each edge in the power grid topology graph, and the intermediate edge is deleted from the power grid topology graph to form a new power grid topology graph. When the power grid topology graph is divided into N unconnected topological subgraphs, each topological subgraph is evaluated for module to obtain the modularity of each topological subgraph, and this is repeated until the modularity value of each topological subgraph no longer changes, and B topological subgraphs are obtained.

7. A power grid intelligent transmission line fault location system according to claim 6, characterized in that: The formula for module evaluation is: Among them, Q j represents the modularity of the jth topological subgraph, a j represents the total number of edges in the jth topological subgraph, represents the degree of the cth topological node in the jth topological subgraph, represents the degree of the dth topological node in the jth topological subgraph, q represents the weight of the edge connecting the cth topological node and the dth topological node, SIM j represents the combination of all topological nodes with connected edges in the jth topological subgraph, c and d represent the indexes of the topological nodes in the jth topological subgraph, and c≠d.

8. A power grid intelligent transmission line fault location system according to claim 7, characterized in that: The fault prediction model is constructed in the following manner: F groups of historical power fault data are collected, and the historical power fault data include fault categories, electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data. Based on the fault category, the historical power fault data with the same fault category are divided into a group to obtain R groups of fault data groups, and anomaly detection is performed on the data in each group of fault data groups. The method of performing anomaly detection includes: the electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data in each group of data in the fault data group are combined into a C×1 fault vector, where C represents the total number of fault data categories; the similarity value between each fault vector in the fault vector and other fault vectors is calculated by a similarity calculation formula, the calculated similarity values ​​are averaged to obtain the average similarity of each fault vector, and the zero score algorithm is used to calculate the similarity of each fault vector based on the average similarity. The discreteness of the quantity is determined, a discreteness threshold is preset, the power fault data with a discreteness greater than or equal to the discreteness threshold is marked as abnormal data, and the historical power fault data marked as abnormal data in each group of fault data is eliminated to obtain a valid fault data group; the valid fault data group is used as a training sample set, the GRU model is used as an initial model, the GRU model is trained using the training sample set, the electric field fault data, magnetic field fault data, mechanical stress fault data and vibration fault data in the valid fault data group are used as input data of the fault prediction model, and the predicted fault category is used as output data of the fault prediction model; minimizing the error between the actual fault category and the predicted fault category is used as the training goal, the recall rate function is used as the loss function of the stability prediction model, and when the loss function reaches convergence, the training is stopped to obtain the fault prediction model.

9. A power grid intelligent transmission line fault location system according to claim 8, characterized in that: The method for determining the fault subgraph includes: According to the power operation data of each topological node in the topological subgraph, the global representative data of each topological subgraph is calculated. The global representative data includes the mean of the electric field data, the mean of the magnetic field data, the mean of the mechanical stress data and the mean of the vibration data. The global representative data is used as the input of the fault prediction model to perform prediction, and the predicted fault category is obtained. The topological subgraph corresponding to the global representative data is marked as the predicted fault category, and all marked topological subgraphs are recorded as fault subgraphs.

10. A power grid intelligent transmission line fault location system according to claim 9, characterized in that: The method for locating the fault point in the fault subgraph includes: Each topological node in the fault subgraph is regarded as a particle. Each particle represents a possible fault point. Based on the number of topological nodes in the fault subgraph, the number of particle swarms is initialized to half of the total number of topological nodes. The number of particles in each particle swarm is initialized to num. The initial speed of each particle is initialized to a very small positive value to ensure that the initial state has a certain randomness. The fitness function formula of the topological node is defined as Representative The fitness value of a topological node, The calculation formula is: Among them, μ, α, β and γ represent weight coefficients, and the weight coefficients satisfy the weight condition μ+α+β+γ=1. Representative Corrected electric field data of topological nodes, Representative Corrected magnetic field data of topological nodes, Representative Corrected mechanical stress data of topological nodes, Representative The corrected vibration data of a topological node is obtained, and the fitness values ​​of all connected topological nodes at the particle's location are calculated based on the fitness value formula. The topological node with the largest fitness value of all connected topological nodes is recorded as the update node. The position of the particle is updated based on the update node. When the fitness value of the update node is greater than the fitness value of the particle's current position, the position of the updated particle is the update node. When the fitness value of the update node is less than or equal to the fitness value of the particle's current position, the particle no longer moves. When the fitness values ​​of all particle positions no longer change, the topological node corresponding to the position where the particles are most concentrated is the fault point.

Images