A distribution network fault location method and system based on limited µPMU data drive

Through a method based on finite µPMU data-driven, topology diagram and neural network technology are used to solve the problems of multiple estimation and high cost in distribution network fault location, achieving more efficient fault location and lower cost fault detection.

CN118707256BActive Publication Date: 2024-11-19HEFEI UNIV OF TECH
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411195611.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-19
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

Among the existing fault positioning methods for distribution networks, the impedance method has multiple estimation problems, while the measurement cost of the traveling wave method is high, making it difficult to achieve fast and accurate fault positioning.

Method used

Using a method based on finite µPMU data driver, by constructing the distribution network topology diagram, obtaining the shortest distance matrix and adjacency matrix, determining the placement position of the µPMU device and smart meter, collecting voltage waveform sequences for discrete wavelet transformation, and combining with graph convolution neural network and artificial neural networks perform fault feature extraction and location prediction.

Benefits of technology

Improves the accuracy and economicality of fault location, reduces the number of installations of µPMU devices, reduces measurement costs, and enhances the data perception capability of the distribution network.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118707256B_ABST
    Figure CN118707256B_ABST
Patent Text Reader

Abstract

The present invention discloses a distribution network fault distance measurement method and system based on limited µPMU data drive, the method includes: constructing a topological map of the distribution network, obtaining the adjacency matrix and the shortest distance matrix of the topological map; determining the placement position of the µPMU device and the smart meter in the distribution network node according to the shortest distance matrix; collecting the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when the distribution network fails; decomposing the voltage waveform sequence by discrete wavelet transform to obtain detail components and approximate components; combining the approximate components with the voltage value measured by the smart meter to obtain the original voltage input feature matrix of the fault selection line and the adjacency matrix as input, inputting it into the convolutional neural network of the fault selection line map to obtain the optimal fault line feature; inputting the detail component and the optimal fault line feature into the fault distance measurement artificial neural network to obtain the location information of the fault. The present invention can solve the multiple estimation problem of impedance method fault location in the traditional method.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault distance measurement, and in particular to a distribution network fault distance measurement method and system driven by limited µPMU data. Background Art

[0002] The distribution system is directly connected to users, and is a key link in ensuring the quality of power supply and user service and improving the operating efficiency of the power system. Due to the wide coverage, numerous lines and complex topology of the distribution network, the probability of failure in the distribution network is greater than that of the transmission line. According to data, 88% of the total power outage time of users in my country is caused by faults in the medium and low voltage distribution network. Therefore, achieving fast and accurate distribution network fault location and effectively reducing fault troubleshooting and power outage time are of great significance to improving the reliability and stability of the distribution network power supply and promoting social and economic development.

[0003] Conventional methods for fault location currently include impedance method and traveling wave method: Generally speaking, the main disadvantage of impedance-based methods is that multiple estimates are required due to the presence of multiple possible fault points at the same distance. Therefore, the impedance method provides accurate but uncertain fault location. The most important disadvantage of the traveling wave method is that it requires measurement equipment with a very high sampling rate (MHz).

[0004] In the prior art, the invention patent with the patent publication number CN112162176A discloses a method for locating phase-to-phase short-circuit faults in distribution networks based on µPMU measurement data. This method only configures micro-synchronous phasor measurements (µPMU) at both ends of the main network of the radial distribution line, and uses the bus voltage and line current data before and after the fault measured by the µPMU for accurate positioning. This method can accurately locate faults in different fault locations, transition resistances of different sizes, and different fault types. However, the comparative document is suitable for locating phase-to-phase short-circuit faults, which provides a basis for troubleshooting and restoring power supply after a distribution network fault. Summary of the invention

[0005] The technical problem to be solved by the present invention is to solve the problem of multiple estimation in the impedance method and the expensive measurement cost of the traveling wave method.

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

[0007] A distribution network fault location method based on limited µPMU data drive, comprising:

[0008] Construct a topological graph of the distribution network and obtain the adjacency matrix and shortest distance matrix of the topological graph;

[0009] According to the normalized shortest distance matrix, determine the placement of the µPMU device in the distribution network node, and place smart meters at other nodes of the distribution network;

[0010] Collect the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when a fault occurs in the distribution network;

[0011] The voltage waveform sequence measured by the µPMU device is decomposed by discrete wavelet transform to obtain detail components and approximate components;

[0012] The approximate component and the voltage value measured by the smart meter are combined to obtain the original voltage input feature matrix for fault line selection; the detail component is used as the original voltage input feature matrix for fault location;

[0013] The original voltage input feature matrix of the fault line selection and the adjacency matrix are input into the fault line selection graph convolutional neural network as input to obtain the optimal fault line features;

[0014] The original voltage input feature matrix of fault location and the optimal fault line feature are taken as input and input into the fault location artificial neural network to obtain the fault location information.

[0015] Advantages: The present invention uses a few µPMUs and smart meters to provide fault data through hybrid measurement, which can solve the problem of multiple estimations in the impedance method and the disadvantage of the expensive measurement cost of the traveling wave method, thereby improving the accuracy and economy of fault location.

[0016] In one embodiment of the present invention, obtaining an adjacency matrix and a shortest distance matrix includes:

[0017] All busbar units in the distribution network are regarded as nodes of a topological graph, and the lines between the busbar units are regarded as edges between the nodes in the topological graph, so as to obtain a topological graph of the distribution network;

[0018] Jordi i Nodes and j Nodes If there is an edge connecting them, let the element in the i-th row and j-th column of the adjacency matrix is 1, otherwise, let is 0, get the adjacency matrix;

[0019] Using Dijkstra's algorithm, we calculate i Nodes and j Nodes The shortest distance between them, get the shortest distance matrix.

[0020] In one embodiment of the present invention, determining the placement position of the µPMU device in the distribution network node includes:

[0021] Normalize the rows and columns of the shortest distance matrix to obtain the normalized shortest distance matrix;

[0022] According to the normalized shortest distance matrix, get the i Nodes The average shortest distance to other nodes;

[0023] Assume that a K The µPMU device with the smallest average shortest distance is selected. K A µPMU device is arranged in each bus unit.

[0024] In one embodiment of the present invention, obtaining the original voltage input characteristic matrix of the fault selection line includes:

[0025] Obtain the voltage waveform sequence measured by the µPMU device and the voltage RMS value measured by the smart meter when a fault occurs in the distribution network;

[0026] Order u The voltage waveform sequence of a bus unit with a µPMU device when a fault occurs is: , , Indicates u Place µ The bus unit of the PMU device collects m The fault voltage value of the sampling point, M Represents the total number of voltage samples. w The RMS value of a bus unit with a smart meter when a fault occurs is , ;

[0027] The voltage waveform sequence is transformed by discrete wavelet transform. Perform 5-layer discrete wavelet decomposition to obtain the u Place µ Five detailed components of the busbar unit of the PMU device and 1 approximate component ,in, Represents approximate component The zth sample value in , Z represents the total number of samples of the approximate component;

[0028] Get the u Place µ Approximate components of the busbar unit of a PMU device The root mean square value of ;

[0029] Depend on NK The voltage RMS value and K Place µ The root mean square of the approximate components of the busbar unit of the PMU device constitutes the original voltage input characteristic matrix of the fault selection line , T represents transpose, N is the total number of nodes.

[0030] In one embodiment of the present invention, obtaining the original voltage input characteristic matrix of fault location measurement includes:

[0031] Depend on K The five detail components of the busbar unit where the µPMU device is placed constitute the original voltage input characteristic matrix of the fault location ;in, Indicates k Five detail components of a busbar unit where a µPMU device is placed.

[0032] In one embodiment of the present invention, obtaining the optimal fault line feature includes:

[0033] Construct a convolutional neural network for fault line selection, including: L Layer graph convolutional network, a softmax layer;

[0034] Define the current number of layers of the graph convolutional layer as , and initialize ; Definition The input features of the layer graph convolutional network are , and initialize ;in, The normalized voltage input characteristic matrix of the original voltage input characteristic matrix of the fault selection line;

[0035] Use the following formula to get Output features of layer graph convolutional network , and as the first Input features of the layer graph convolutional network;

[0036] ;

[0037] In the formula, I represents the identity matrix, represents the adjacency matrix, Indicates The weight matrix to be trained in the layer graph convolutional network, represents the activation function;

[0038] make Assign to Then return to the previous step until you get theL Output features of layer graph convolutional network Then proceed to the next step;

[0039] Will The input is processed in the softmax layer, and the probability of failure of all edges in the topology graph is output. The edge with the highest probability is selected as the predicted fault line. Y ;

[0040] Based on real fault lines and predicted fault lines Y Calculate the cross entropy loss function of the convolutional neural network for fault line selection graph , and use the gradient descent algorithm to continuously update the training network parameters until convergence, so as to obtain the optimal fault line selection convolutional neural network, and use it to output the optimal fault line characteristics .

[0041] In one embodiment of the present invention, obtaining the location information of the fault includes:

[0042] The artificial neural network for fault location measurement includes: an input layer, a hidden layer, an output layer and a softmax layer;

[0043] The normalized original voltage of the fault line is input into the characteristic matrix and optimal fault line characteristics Input into the input layer, and then use the following formula to obtain the output matrix of the hidden layer P ;

[0044] ;

[0045] In the formula, , The normalized original voltage input feature matrix of the fault line selection and the optimal fault line feature matrix in the input layer to the hidden layer of the fault location artificial neural network are respectively Two weight matrices of ; Represents the threshold matrix from the input layer to the hidden layer;

[0046] The output matrix of the output layer of the fault location artificial neural network is obtained by the following formula: O :

[0047] ;

[0048] In the formula, Represents the weight matrix from the input layer to the hidden layer; Represents the threshold matrix from hidden layer to output layer; Represents the sigmoid activation function;

[0049] The output matrix of the output layer O After being processed in the softmax layer, the optimal fault line features are obtained. Characteristics of the upper fault point to the optimal fault line The distance between the left busbar unit and the optimal fault line characteristics The ratio between the lengths of and is used as the predicted fault location;

[0050] A loss function is established according to the error between the predicted fault location and the actual fault location, and the fault location artificial neural network is trained using the error back propagation algorithm to adjust the weight matrix and threshold matrix of the fault location artificial neural network until the loss function converges, thereby obtaining the optimal fault location artificial neural network for realizing the location prediction of the fault point on the line.

[0051] In one embodiment of the present invention, the average shortest distance is obtained by the following formula:

[0052] ;

[0053] In the formula, Indicates i Nodes The average shortest distance to other nodes, Indicates the total number of nodes, Represents the normalized shortest distance matrix The element at row i and column j in .

[0054] In one embodiment of the present invention, the probability of failure of all edges in the output topology graph is obtained by the following formula:

[0055] ;

[0056] In the formula, is the probability of edge failure in the topology graph, is the input of the classification function, An edge is a connection between two nodes.

[0057] The present invention also provides a system for applying the above-mentioned distribution network fault location method based on limited µPMU data drive, comprising:

[0058] The topology map module is used to construct the topology map of the distribution network and obtain the adjacency matrix and the shortest distance matrix of the topology map;

[0059] A voltage measurement device location confirmation module is used to determine the placement location of the µPMU device in the distribution network node and to place smart meters at other nodes of the distribution network based on the normalized shortest distance matrix;

[0060] A data acquisition module is used to collect the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when a fault occurs in the distribution network;

[0061] A data decomposition module is used to decompose the voltage waveform sequence measured by the µPMU device through discrete wavelet transform to obtain detail components and approximate components;

[0062] The voltage input characteristic matrix module is used to combine the approximate component with the voltage value measured by the smart meter to obtain the original voltage input characteristic matrix for fault line selection; the detail component is used as the original voltage input characteristic matrix for fault location;

[0063] The optimal fault line feature module is used to input the original voltage input feature matrix of the fault line selection and the adjacency matrix as input to the fault line selection graph convolutional neural network to obtain the optimal fault line feature;

[0064] The fault location information prediction module is used to input the fault distance original voltage input feature matrix and the optimal fault line feature as input into the fault distance artificial neural network to obtain the fault location information.

[0065] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention proposes to use the approximate components of the voltage measured by the µPMU device and the root mean square value of the voltage measured by the smart meter for fault line selection, and the detailed components of the voltage measured by the µPMU device for fault distance measurement, which gives full play to the characteristics of high-frequency and low-frequency signals, solves the multiple estimation problem, and improves the accuracy of fault location.

[0066] The present invention constructs a shortest distance matrix for determining a bus unit suitable for installing a µPMU device, thereby reducing the number of µPMU devices to be installed, improving the measurement cost of fault location, and improving the data perception capability of the distribution network.

[0067] The present invention introduces a graph convolutional neural network for fault line selection, which can not only extract the voltage information of the node, but also make good use of the spatial topology information of the distribution network, thereby increasing the accuracy of fault line selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The present invention is a flow chart of a distribution network fault location method driven by limited µPMU data according to an embodiment of the present invention.

[0069] Figure 2 This is a structural diagram of a distribution network according to an embodiment of the present invention.

[0070] Figure 3 This is a structural block diagram of a fault line selection convolutional neural network and a fault distance artificial neural network according to an embodiment of the present invention.

[0071] Figure 4 The figure is a block diagram of a distribution network fault location system driven by limited µPMU data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0073] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0074] See also Figure 1 As shown, the present invention provides a distribution network fault location method driven by limited µPMU data, comprising:

[0075] S10, constructing a topological graph of the distribution network, and obtaining an adjacency matrix and a shortest distance matrix of the topological graph.

[0076] See also Figure 2 As shown, in one embodiment of the present invention, obtaining the adjacency matrix and the shortest distance matrix of the topological graph includes:

[0077] S11, all busbar units in the distribution network are regarded as nodes of the topological graph, and the lines between the busbar units are regarded as edges between the nodes in the topological graph, so as to obtain the topological graph of the distribution network G= ( V, E ),in, A collection of nodes representing a topological graph, Indicates i nodes, N Indicates the total number of nodes, represents the edge set of the topological graph, Indicates i Nodes and j Nodes There are edges with connecting relationships between them. Assume that the weight of each edge with connecting relationship is 1.

[0078] S12, if i Nodes and j Nodes There are edges connecting each other, so let the adjacency matrix A The element in row i and column j in is 1, otherwise, let is 0, get the adjacency matrix A .

[0079] S13, Dijkstra algorithm can be used to calculate the shortest path from one vertex to the remaining vertices, solving the shortest path problem in the topological graph. The main feature of this algorithm is that it starts from the starting point and adopts the strategy of greedy algorithm, traversing the adjacent nodes of the vertex closest to the starting point and not visited each time until it reaches the end point. i Nodes and j Nodes The shortest distance between , get the shortest distance matrix .

[0080] S20, determining the placement location of the µPMU device in the distribution network node according to the normalized shortest distance matrix, and placing smart meters at other nodes of the distribution network.

[0081] In one embodiment of the present invention, determining the placement position of the µPMU device in the distribution network node includes:

[0082] S21, using formula (1) for the shortest distance matrix D Normalize rows and columns to get the normalized shortest distance matrix ;

[0083] , (1);

[0084] In formula (1), and is a diagonal matrix, Each element on the diagonal is the shortest distance matrix D The sum of the elements in the corresponding rows; Each element on the diagonal is the shortest distance matrix D The sum of the corresponding column elements in ; and Respectively Sum of square roots of inverse matrices The square root of the inverse matrix.

[0085] S22, using formula (2) to calculate the i Nodes The average shortest distance to other nodes .

[0086] , (2);

[0087] In formula (2), Represents the normalized shortest distance matrix The element at row i and column j in .

[0088] S23, assuming that a K The µPMU device with the smallest average shortest distance is selected. K Each bus unit is equipped with a µPMU device, and the remaining N - K Each bus unit is equipped with a smart meter. K =3, the µPMU device is arranged in the following position: Figure 2 shown.

[0089] S30, collects the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when a fault occurs in the distribution network.

[0090] In one embodiment of the present invention, the collection K The voltage waveform sequence of the bus unit with µPMU device when the fault occurs and NK The voltage root mean square value of the bus unit where the smart meter is placed when a fault occurs. u Place µ The voltage waveform sequence of the bus unit of the PMU device when a fault occurs is: , , Indicates u The first m The fault voltage value of the sampling point, M Represents the total number of voltage samples. w The RMS value of a bus unit with a smart meter when a fault occurs is , .

[0091] S40, decomposes the voltage waveform sequence measured by the µPMU device through discrete wavelet transform to obtain detail components and approximate components.

[0092] In one embodiment of the present invention, formula (3) is used to Perform 5-layer discrete wavelet decomposition to obtain the u Five detailed components of a busbar unit where a µPMU device is placed and 1 approximate component ,in, Represents approximate component The zth sample value in , Z represents the total number of samples of the approximate component.

[0093] , (3);

[0094] In formula (3), The coefficients representing the detail or approximate components of the decomposition levels 1 to 5. Specifically, the discrete wavelet decomposition uses the db4 wavelet basis function. Daubecies-4 (db4) is one of the most widely used wavelets in power system disturbance analysis. DB4 is selected as the wavelet function due to its orthogonality, compact time domain support, and good performance in power system research.

[0095] S50, combining the approximate component with the voltage value measured by the smart meter to obtain the original voltage input feature matrix for fault line selection; and using the detail component as the original voltage input feature matrix for fault location.

[0096] In one embodiment of the present invention, formula (4) is used to calculate the u Approximate components of busbar units where µPMU devices are placed The root mean square value of .

[0097] , (4);

[0098] Depend on NK The RMS value and K The RMS values ​​of the approximate components of the busbar units where the µPMU devices are placed constitute the original voltage input characteristic matrix of the fault selection line , T Indicates transpose.

[0099] Input characteristic matrix of original voltage of fault selection line After normalization, the normalized voltage input characteristic matrix of the original voltage input characteristic matrix of the fault selection line is obtained: ,in, It represents the normalized result of the RMS value of the w-th bus unit where the smart meter is placed when a fault occurs. Indicates u The normalized result of the approximate components of the busbar unit where the µPMU device is placed.

[0100] In one embodiment of the present invention, K The five detail components of the busbar unit where the µPMU device is placed constitute the original voltage input characteristic matrix of the fault location ;in, Indicates k Five detail components of a busbar unit where a µPMU device is placed.

[0101] The original voltage input characteristic matrix of fault location After normalization, the normalized original voltage input characteristic matrix of the fault selection line is obtained: ;in, Indicates k Normalized values ​​of the five detail components of a busbar unit where a µPMU device is placed.

[0102] S60, inputting the original voltage input feature matrix of the fault line selection and the adjacency matrix as input into the fault line selection graph convolutional neural network to obtain the optimal fault line feature.

[0103] In one embodiment of the present invention, obtaining the optimal fault line feature includes:

[0104] S61, constructing a convolutional neural network for fault line selection graph, including: L A layer graph convolutional network, a softmax layer; used to implement fault line selection. In this embodiment, L =4.

[0105] S62, define the current number of layers of the graph convolutional layer as , and initialize ; Definition The input features of the layer graph convolutional network are , and initialize .

[0106] S63, using formula (5) to get Output features of layer graph convolutional network , and as the first Input features of the layer graph convolutional network.

[0107] , (5);

[0108] In formula (5), I represents the identity matrix, the adjacency matrix Plus the identity matrix represents the influence of the node on itself, Indicates l The weight matrix to be trained in the layer graph convolutional network, Represents the activation function.

[0109] S64, order Assign to Then, return to S63 until the first L Output features of layer graph convolutional network Then, execute S65.

[0110] S65, The input is processed in the softmax layer, and the probability of failure of all edges in the topology graph is output. The edge with the highest probability is selected as the predicted fault line. Y Specifically, the classification function in the Softmax layer is shown in formula (6):

[0111] , (6);

[0112] In formula (6), is the input of the classification function, which can be understood as L Output features of layer graph convolutional network , is the output of the classification function, i.e., the probability of edge failure in the topology graph, is an edge that connects two nodes. By using formula (6), the output feature is mapped to the interval [0, 1].

[0113] S66, based on real fault lines and predicted fault lines Y Calculating the cross entropy loss function for graph convolutional neural networks , and use the gradient descent algorithm to continuously update the training network parameters until convergence, so as to obtain the optimal fault line selection convolutional neural network, and use it to output the optimal fault line characteristics .

[0114] S70, inputting the fault distance original voltage input feature matrix and the optimal fault line feature into the fault distance artificial neural network as input to obtain the fault location information.

[0115] In one embodiment of the present invention, obtaining the location information of the fault includes:

[0116] S71, construct a fault location artificial neural network including: an input layer, a hidden layer, an output layer and a softmax layer for fault location.

[0117] S72, input the normalized original voltage of the fault selection line into the characteristic matrix and optimal fault line characteristics Input into the input layer, and then use formula (7) to get the output matrix of the hidden layer P :

[0118] , (7);

[0119] In formula (7), , The normalized original voltage input feature matrix of the fault line selection and the optimal fault line feature matrix in the input layer to the hidden layer of the fault location artificial neural network are respectively Two weight matrices of ; Represents the threshold matrix from the input layer to the hidden layer.

[0120] S73, using formula (8) the output matrix of the output layer of the fault location artificial neural network O.

[0121] , (8);

[0122] In formula (8), Represents the weight matrix from the input layer to the hidden layer; Represents the threshold matrix from hidden layer to output layer; represents the activation function; in this embodiment, is the sigmoid function.

[0123] S74, the output matrix of the output layer O After being processed in the softmax layer, the optimal fault line feature is obtained. Characteristics of the upper fault point to the optimal fault line The distance between the left busbar unit and the optimal fault line characteristics and used as the predicted fault location.

[0124] S75, establish a loss function based on the error between the predicted fault location and the actual fault location, and use the error back propagation algorithm to train the fault distance artificial neural network to adjust the weight matrix and threshold matrix of the fault distance artificial neural network until the loss function converges, thereby obtaining the optimal fault distance artificial neural network for realizing the location prediction of the fault point on the line. The structure diagram of the convolutional neural network of the fault selection line diagram and the fault distance artificial neural network is shown in the figure. Figure 3 shown.

[0125] See also Figure 4 As shown, the present invention also provides a system for applying the above-mentioned distribution network fault location method based on limited µPMU data drive, including:

[0126] The topology map module is used to construct the topology map of the distribution network and obtain the adjacency matrix and shortest distance matrix of the topology map.

[0127] The voltage measurement device location confirmation module is used to determine the placement location of the µPMU device in the distribution network node according to the normalized shortest distance matrix, and to place smart meters at other nodes of the distribution network.

[0128] The data acquisition module is used to collect the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when a fault occurs in the distribution network.

[0129] The data decomposition module is used to decompose the voltage waveform sequence measured by the µPMU device through discrete wavelet transform to obtain detail components and approximate components.

[0130] The voltage input characteristic matrix module is used to combine the approximate component with the voltage value measured by the smart meter to obtain the original voltage input characteristic matrix for fault line selection; and use the detail component as the original voltage input characteristic matrix for fault location.

[0131] The optimal fault line feature module is used to input the original voltage input feature matrix of the fault line selection and the adjacency matrix as input into the fault line selection graph convolutional neural network to obtain the optimal fault line features.

[0132] The fault location information prediction module is used to input the fault distance original voltage input feature matrix and the optimal fault line feature as input into the fault distance artificial neural network to obtain the fault location information.

[0133] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0134] The above-described embodiments merely represent implementation methods of the invention. The protection scope of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements may be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A distribution network fault location method driven by limited µPMU data, It is characterized in that include: Construct a topological graph of the distribution network and obtain the adjacency matrix and shortest distance matrix of the topological graph; According to the normalized shortest distance matrix, determine the placement of the µPMU device in the distribution network node, and place smart meters at other nodes of the distribution network; Collect the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when a fault occurs in the distribution network; The voltage waveform sequence measured by the µPMU device is decomposed by discrete wavelet transform to obtain detail components and approximate components; The approximate components are combined with the voltage values ​​measured by the smart meter to obtain the original voltage input characteristic matrix of the fault selection line; The detail component is input into the characteristic matrix as the original voltage of fault location; The original voltage input feature matrix of the fault line selection and the adjacency matrix are input into the fault line selection graph convolutional neural network as input to obtain the optimal fault line features; The original voltage input feature matrix of fault location and the optimal fault line feature are input into the fault location artificial neural network to obtain the fault location information; The acquisition of fault location information includes: The artificial neural network for fault location measurement includes: an input layer, a hidden layer, an output layer and a softmax layer; The normalized original voltage of the fault line is input into the characteristic matrix and optimal fault line characteristics Input into the input layer, and then use the following formula to obtain the output matrix of the hidden layer P ; ; In the formula, , The normalized original voltage input feature matrix of the fault line selection and the optimal fault line feature matrix in the input layer to the hidden layer of the fault location artificial neural network are respectively Two weight matrices of ; Represents the threshold matrix from the input layer to the hidden layer; The output matrix of the output layer of the fault location artificial neural network is obtained by the following formula: O : ; In the formula, Represents the weight matrix from the input layer to the hidden layer; Represents the threshold matrix from hidden layer to output layer; Represents the sigmoid activation function; The output matrix of the output layer O After being processed in the softmax layer, the optimal fault line features are obtained. Characteristics of the upper fault point to the optimal fault line The distance between the left busbar unit and the optimal fault line characteristics The ratio between the lengths of and is used as the predicted fault location; A loss function is established according to the error between the predicted fault location and the actual fault location, and the fault location artificial neural network is trained using the error back propagation algorithm to adjust the weight matrix and threshold matrix of the fault location artificial neural network until the loss function converges, thereby obtaining the optimal fault location artificial neural network for realizing the location prediction of the fault point on the line.

2. The distribution network fault location method based on limited µPMU data drive according to claim 1, It is characterized in that Get the adjacency matrix and the shortest distance matrix, including: All busbar units in the distribution network are regarded as nodes of a topological graph, and the lines between the busbar units are regarded as edges between the nodes in the topological graph, so as to obtain a topological graph of the distribution network; Jordi i Nodes and j Nodes If there is an edge connecting them, let the element in the i-th row and j-th column of the adjacency matrix is 1, otherwise, let is 0, get the adjacency matrix; Using Dijkstra's algorithm, we calculate i Nodes and j Nodes The shortest distance between them, get the shortest distance matrix.

3. The distribution network fault location method based on limited µPMU data drive according to claim 1, It is characterized in that Determine the placement of µPMU devices in the distribution network nodes, including: Normalize the rows and columns of the shortest distance matrix to obtain the normalized shortest distance matrix; According to the normalized shortest distance matrix, get the i Nodes The average shortest distance to other nodes; Assume that a K µPMU devices, select the front with the smallest average shortest distance K A µPMU device is arranged in each bus unit.

4. The distribution network fault location method based on limited µPMU data drive according to claim 1, It is characterized in that Obtain the original voltage input feature matrix of the fault selection line, including: Obtain the voltage waveform sequence measured by the µPMU device and the voltage RMS value measured by the smart meter when a fault occurs in the distribution network; Order u The voltage waveform sequence of a bus unit with a µPMU device when a fault occurs is: , , Indicates u Place µ The bus unit of the PMU device collects m The fault voltage value of the sampling point, M Represents the total number of voltage samples. w The RMS value of a bus unit with a smart meter when a fault occurs is , ; The voltage waveform sequence is transformed by discrete wavelet transform. Perform 5-layer discrete wavelet decomposition to obtain the u Place µ Five detailed components of the busbar unit of the PMU device and 1 approximate component ,in, Represents approximate component The zth sample value in , Z represents the total number of samples of the approximate component; Get the u Place µ Approximate components of the busbar unit of a PMU device The root mean square value of ; Depend on NK The voltage RMS value and K Place µ The root mean square of the approximate components of the busbar unit of the PMU device constitutes the original voltage input characteristic matrix of the fault selection line , T represents transpose, N is the total number of nodes.

5. The distribution network fault location method based on limited µPMU data drive according to claim 4, It is characterized in that Obtain the original voltage input feature matrix for fault location, including: Depend on K The five detail components of the busbar unit where the µPMU device is placed constitute the original voltage input characteristic matrix of the fault location ;in, Indicates k Five detail components of a busbar unit where a µPMU device is placed.

6. The distribution network fault location method based on limited µPMU data drive according to claim 1, It is characterized in that Obtain optimal fault line characteristics, including: Construct a convolutional neural network for fault line selection, including: L Layer graph convolutional network, a softmax layer; Define the current number of layers of the graph convolutional layer as , and initialize ; Definition The input features of the layer graph convolutional network are , and initialize ;in, The normalized voltage input characteristic matrix of the original voltage input characteristic matrix of the fault selection line; Use the following formula to get Output features of layer graph convolutional network , and as the first Input features of the layer graph convolutional network; ; In the formula, I represents the identity matrix, represents the adjacency matrix, Indicates The weight matrix to be trained in the layer graph convolutional network, represents the activation function; make Assign to Then return to the previous step until you get the L Output features of layer graph convolutional network Then proceed to the next step; Will The input is processed in the softmax layer, and the probability of failure of all edges in the topology graph is output. The edge with the highest probability is selected as the predicted fault line. Y ; Based on real fault lines and predicted fault lines Y Calculate the cross entropy loss function of the convolutional neural network for fault line selection graph , and use the gradient descent algorithm to continuously update the training network parameters until convergence, so as to obtain the optimal fault line selection convolutional neural network, and use it to output the optimal fault line characteristics .

7. The distribution network fault location method based on limited µPMU data drive according to claim 3, It is characterized in that The average shortest distance is obtained by the following formula: ; In the formula, Indicates i Nodes The average shortest distance to other nodes, Indicates the total number of nodes, Represents the normalized shortest distance matrix The element at row i and column j in .

8. The distribution network fault location method based on limited µPMU data drive according to claim 6, It is characterized in that The probability of failure of all edges in the output topology graph is obtained by the following formula: ; In the formula, is the probability of edge failure in the topology graph, is the input of the classification function, An edge is a connection between two nodes.

9. A system using the distribution network fault location method based on limited µPMU data drive as described in any one of claims 1 to 8, It is characterized in that include: The topology map module is used to construct the topology map of the distribution network and obtain the adjacency matrix and the shortest distance matrix of the topology map; A voltage measurement device location confirmation module is used to determine the placement location of the µPMU device in the distribution network node and to place smart meters at other nodes of the distribution network based on the normalized shortest distance matrix; A data acquisition module is used to collect the voltage waveform sequence measured by the µPMU device and the voltage value measured by the smart meter when a fault occurs in the distribution network; A data decomposition module is used to decompose the voltage waveform sequence measured by the µPMU device through discrete wavelet transform to obtain detail components and approximate components; The voltage input characteristic matrix module is used to combine the approximate components with the voltage values ​​measured by the smart meter to obtain the original voltage input characteristic matrix of the fault line selection; The detail component is input into the characteristic matrix as the original voltage of fault location; The optimal fault line feature module is used to input the original voltage input feature matrix of the fault line selection and the adjacency matrix as input to the fault line selection graph convolutional neural network to obtain the optimal fault line feature; The fault location information prediction module is used to input the fault distance original voltage input feature matrix and the optimal fault line feature as input into the fault distance artificial neural network to obtain the fault location information.

Citation Information

Patent Citations

  • Power distribution network phase-to-phase short circuit fault positioning method based on muPMU measurement data

    CN112162176A

  • Single-pole grounding fault distance measurement method based on wavelet decomposition and DBN

    CN112114232A

  • Distribution line fault positioning method based on spatial data fusion

    CN114624551A

  • Power distribution network fault identification method based on wavelet optimization and graph convolutional neural network

    CN116298699A

  • Active power distribution network fault positioning method based on measurement data

    CN117269666A