Distributed power transmission line fault accurate positioning system and method

The neural network model processes the graph structure data and wave recording data of distributed transmission lines, which solves the problem of insufficient fault positioning accuracy in the existing technology, and achieves higher fault positioning accuracy and interference information stripping capability.

CN118914744AActive Publication Date: 2024-11-08MAANSHAN POWER SUPPLY COMPANY STATE GRID ANHUI ELECTRIC POWER +1
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410951974.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-08
Estimated Expiration
2044-07-16

AI Technical Summary

Technical Problem

The existing distributed transmission line fault positioning system fails to make full use of the data collected by the distributed system, resulting in insufficient fault positioning accuracy and difficulty in stripping off interfering information.

Method used

Using neural network model combined with graph structure data and wave recording data, the location and type classification results of fault points are generated through the self-encoding layer, tensor layer and graph structure learning layer, and deep learning is used to organize the recording of distributed transmission line fault positioning system. Wave data, fused with transient wave recording data of multiple line segments.

Benefits of technology

It improves the accuracy of fault positioning, can effectively strip off interfering information, and improves the accuracy and reliability of positioning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914744A_ABST
    Figure CN118914744A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of power transmission line distributed fault positioning, and discloses a distributed power transmission line fault accurate positioning system and method, and the method comprises the steps: collecting the recording data of a power transmission line fault; constructing graph structure data based on the collected recording data; inputting the graph structure data into a neural network model, wherein the neural network model outputs a result representing the position of a fault point of the power transmission line; according to the fault distance ratio, calculating the distance between the fault point and the end point at which the fault transient recording is firstly started, extracting the position of the end point at which the fault transient recording is firstly started, and then positioning the fault point; according to the method, recording data of a distributed power transmission line fault positioning system is organized by means of deep learning, information abundance capable of stripping interference information is provided by fusion of global transient recording data, and the positioning precision is improved to a certain extent.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

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

[0002] Traveling wave location technology is a fault diagnosis method based on measuring the propagation time of transient traveling waves generated by faults. It has the characteristics of high accuracy and fast response. This technology has been widely used in transmission lines of voltage levels of 110kV and above, and has gradually been applied in distribution networks. Traveling wave location technology includes current traveling wave location and voltage traveling wave location, as well as single-end measurement method and double-end measurement method.

[0003] According to the structural method, it can be divided into: distributed and integrated.

[0004] The distributed system adopts the same structure as the transmission system, and the sensor is installed on the line. The fusion type is to integrate the traveling wave sensor with the pole switch, which is built-in or external to the pole;

[0005] Although the fault location system of the distributed system structure has a distributed layout of data acquisition terminals, it still uses traditional methods when locating faults and does not fully utilize the data collected by the distributed system. Summary of the invention

[0006] The present invention provides a distributed power transmission line fault accurate positioning system and method, which solves the technical problems in related technologies.

[0007] The present invention provides a method for accurately locating a distributed power transmission line fault, comprising:

[0008] Step 101, collecting recorded wave data when a transmission line fails;

[0009] Divide the transmission line into multiple line segments, and collect waveform data at the end points of each data segment;

[0010] The recording data includes the fault transient recording data of the endpoints of all line sections when a transmission line fails;

[0011] Step 102, constructing graph structure data based on the collected recorded wave data;

[0012] The graph structure data includes nodes and edges connecting the nodes. The nodes include trunk nodes and branch nodes. A trunk node represents a line segment of a trunk, and a branch node represents a line segment of a branch.

[0013] The generation rules of edges between nodes are as follows:

[0014] If the line segments represented by two trunk nodes are connected, an edge is generated between the two trunk nodes;

[0015] If the line segments represented by the trunk node and the branch node are connected, an edge is generated between the trunk node and the branch node;

[0016] If two branch nodes belong to the same branch, then there is an edge between the two branch nodes;

[0017] Step 103, inputting the graph structure data into the neural network model, and the neural network model outputs a result indicating the location of the fault point of the transmission line;

[0018] The neural network model includes:

[0019] The self-encoding layer is used to input a fault transient recording to output a self-encoding feature. When the fault transient recording of the starting point of the line segment represented by the node is input, the self-encoding feature corresponding to the node is output;

[0020] The tensor layer is used to quantize the node's autoencoder feature tensor into a vector form and then input it into the graph structure learning layer;

[0021] The graph structure learning layer is used to input the self-encoded features of the nodes and the graph structure data, and output the global features;

[0022] A fully connected layer, which is used to input global features and then output a result representing the distance ratio of the fault point;

[0023] Step 104, calculating the distance between the fault point and the endpoint of the first fault transient recording according to the fault distance ratio, extracting the position of the endpoint of the first fault transient recording, and then locating the fault point.

[0024] If the fault point distance ratio is a positive value, it means that the fault point is on the line segment after the endpoint that first starts the fault transient recording, otherwise it means that the fault point is on the line segment before the endpoint that first starts the fault transient recording.

[0025] Furthermore, the fault transient recording is collected synchronously by all line segments. When the endpoint of any line segment triggers the fault transient recording, the endpoints of all line segments synchronously perform the fault transient recording.

[0026] Furthermore, the line segment includes a trunk line segment and a branch line segment.

[0027] Furthermore, the fault point distance ratio is expressed as follows:

[0028]

[0029] Among them, L 0,+Indicates the length of the line section before the endpoint where the fault transient recording is first started, L 0,- Indicates the length of the line section after the endpoint that first starts the fault transient recording, L 0,b Indicates the distance between the fault point and the endpoint that first starts the fault transient recording.

[0030] Further, the waveform data of segments A, B, C, and D of the fault transient recording data are input from the encoding layer, and then the weighted features of segment A, segment B, segment C, and segment D are output;

[0031] Each waveform segment A, B, C, D is itself a time series, containing waveform sampling values ​​of multiple time steps. They can be regarded as four independent input sequences, denoted by X A , X B , X C , X D .

[0032] For each sequence, add the corresponding segment type embedding. For example, for X A , copy and expand the embedding vector of segment A to be the same as X A Matrices of equal length, denoted by E A Then X A and E A Concatenate on the feature dimension to get X′ A .

[0033] The four enhanced input sequences X′ A , X′ B , X′ C , X′ D Splice them into a long sequence X′=[X′ A , X′ B , X′ C , X′ D ].

[0034] Finally, X′ is input into the attention layer;

[0035] The calculation formula of the attention layer is as follows:

[0036] Perform a linear transformation on X′ to obtain the Query matrix Q, Key matrix K and Value matrix V.

[0037] Q=X′W Q

[0038] K=X′W K

[0039] V=X′W V

[0040] in is the learnable weight matrix, d k is the dimension of the attention head.

[0041] Then calculate the attention weight matrix

[0042]

[0043] Finally, the attention weight matrix is ​​multiplied by the Value matrix and linearly transformed to obtain the output autoencoder feature

[0044] Z=AVW O

[0045] in is the output layer weight matrix.

[0046] Furthermore, the graph structure learning layer includes 3 hidden layers and one linear layer. The calculation formula of the first hidden layer is as follows:

[0047]

[0048] in is the attention weight of node u to node v, N(v) represents the set of nodes that have edges with node v, Represents the first weight parameter of layer 1, which is calculated by the following formula:

[0049]

[0050] a is a learnable attention vector, || represents the vector concatenation operation, represents the second weight parameter of the first layer, and LeakyReLU represents the LeakyReLU function;

[0051]

[0052] where x v and x u Represent the feature representation of the vth and uth nodes respectively

[0053] The linear layer is calculated as follows:

[0054] Where Q t Represents global features, It represents the feature representation of node c in the third layer, N(all) represents the set of all nodes, and Concat represents vector concatenation.

[0055] Furthermore, the tensor quantization of the tensor layer is to expand the matrix row by row into vectors and then concatenate them.

[0056] Furthermore, the neural network model also includes a second fully connected layer, which is used to input global features and then output a classification result representing the fault type.

[0057] The present invention provides a distributed power transmission line fault accurate positioning system, comprising:

[0058] Signal detection terminals installed on the trunk and branches of the transmission line;

[0059] a communication terminal for receiving data from a signal detection terminal, the communication terminal being used to send the data to a system master station;

[0060] The system master station is used to execute steps 102-104 of the aforementioned distributed power transmission line fault accurate location method.

[0061] Furthermore, the signal detection terminal performs time synchronization of recording according to Beidou or GPS timing. The signal detection terminal can record according to the recording instruction. The recording start time included in the recording instruction can obtain synchronized fault transient recording when the instruction is executed.

[0062] The present invention provides a storage medium storing computer-readable instructions. When the computer-readable instructions are read, the above-mentioned method for accurately locating a distributed power transmission line fault can be executed.

[0063] The beneficial effects of the present invention are as follows: the present invention uses deep learning to organize the recorded data of the distributed transmission line fault location system, and utilizes the transmission characteristics of transient waves on the transmission line to fuse the transient recorded data on multiple line segments, thereby solving the problem that it is difficult to remove interference information from a single transient recorded data when performing fault location. The global fusion of transient recorded data provides information richness that can remove interference information, thereby improving the positioning accuracy to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a flow chart of a method for accurately locating faults in distributed power transmission lines according to the present invention;

[0065] Figure 2 This is a schematic diagram of the structure of the neural network model. Figure 1 ;

[0066] Figure 3 This is a schematic diagram of the structure of the neural network model. Figure 2 ;

[0067] Figure 4-1 Schematic diagram of single-line configuration of overhead lines;

[0068] Figure 4-2 Overhead line 220kV line site diagram;

[0069] Figure 4-3 Schematic diagram of single-line configuration of cable line;

[0070] Figure 4-4 Schematic diagram of overhead and cable hybrid line configuration;

[0071] Figure 4-5 Schematic diagram of overhead T-connected line configuration;

[0072] Figure 4-6 Schematic diagram of cable line branch box configuration;

[0073] Figure 5-1 Typical waveform of lightning shielding failure;

[0074] Figure 5-2 Typical waveform of lightning strike failure;

[0075] Figure 5-3 Typical waveform of tree barrier fault;

[0076] Figure 5-4 Typical waveform of wildfire fault;

[0077] Figure 5-5 Typical waveform of hanging object failure;

[0078] Figure 5-6 Typical waveform of icing fault;

[0079] Figure 5-7 Typical waveform of wind deviation fault;

[0080] Figure 5-8 Typical waveform diagram of construction machinery hitting the line failure. DETAILED DESCRIPTION

[0081] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0082] At least one embodiment of the present invention discloses a method for accurately locating a distributed power transmission line fault. Figure 1 As shown, including:

[0083] Step 101, collecting recorded wave data when a transmission line fails;

[0084] The transmission line is divided into multiple line segments, and the recorded data is collected at the end point of each data segment.

[0085] The fault transient recording is collected synchronously by all line segments. When the endpoint of any line segment triggers the fault transient recording, the endpoints of all line segments synchronously record the fault transient recording.

[0086] Therefore, the recording data includes fault transient recording data of the endpoints of all line sections when a transmission line fails.

[0087] Of course, in order to ensure the time synchronization of the starting point of the recording, it is necessary to perform time synchronization of the recording according to Beidou or GPS timing. Even if there is a delay in sending the recording instruction, the recording start time contained in the instruction can obtain synchronized fault transient recording when the instruction is executed.

[0088] The fault transient recording trigger is generally initiated based on abnormal fluctuations in current and voltage. Generally, a current or voltage threshold is set as the trigger condition. The threshold here can be the magnitude or phase of the current or voltage.

[0089] The line section includes the line section of the trunk line and the line section of the branch line;

[0090] Step 102, constructing graph structure data based on the collected recorded wave data;

[0091] The graph structure data includes nodes and edges connecting the nodes. The nodes include trunk nodes and branch nodes. A trunk node represents a line segment of a trunk, and a branch node represents a line segment of a branch.

[0092] The generation rules of edges between nodes are as follows:

[0093] If the line segments represented by two trunk nodes are connected, an edge is generated between the two trunk nodes;

[0094] If the line segments represented by the trunk node and the branch node are connected, an edge is generated between the trunk node and the branch node;

[0095] If two branch nodes belong to the same branch, then there is an edge between the two branch nodes;

[0096] This is because the fluctuation caused by the branch fault has a smaller impact on the main line, but a greater impact on the same branch line. Therefore, the fully connected edge operation is performed on the internal nodes of the branch line. For the main line itself, the transient wave transmission of the main line fault is more regular, so only the edge operation is performed on the nodes of the adjacent line segments.

[0097] Step 103, inputting the graph structure data into the neural network model, and the neural network model outputs a result indicating the location of the fault point of the transmission line;

[0098] The neural network model includes:

[0099] The self-encoding layer is used to input a fault transient recording to output a self-encoding feature. When the fault transient recording of the starting point of the line segment represented by the node is input, the self-encoding feature corresponding to the node is output;

[0100] The tensor layer is used to quantize the node's autoencoder feature tensor into a vector form and then input it into the graph structure learning layer;

[0101] The graph structure learning layer is used to input the self-encoded features of the nodes and the graph structure data, and output the global features;

[0102] A fully connected layer, which is used to input global features and then output a result representing the distance ratio of the fault point;

[0103] The fault point distance ratio is expressed as follows:

[0104]

[0105] Among them, L 0,+ Indicates the length of the line section before the endpoint where the fault transient recording is first started, L 0,- Indicates the length of the line section after the endpoint that first starts the fault transient recording, L 0,b Indicates the distance between the fault point and the endpoint that first starts the fault transient recording;

[0106] Step 104, calculating the distance between the fault point and the endpoint of the first fault transient recording according to the fault distance ratio, extracting the position of the endpoint of the first fault transient recording, and then locating the fault point.

[0107] If B0 is a positive value, it means that the fault point is on the line segment after the endpoint that first starts the fault transient recording, otherwise it means that the fault point is on the line segment before the endpoint that first starts the fault transient recording.

[0108] The self-encoding layer inputs the waveform data of segments A, B, C, and D of the fault transient recording data, and then outputs the weighted features of segment A, segment B, segment C, and segment D;

[0109] Each waveform segment A, B, C, D is itself a time series, containing waveform sampling values ​​of multiple time steps. They can be regarded as four independent input sequences, denoted by X A , X B , X C , X D .

[0110] For each sequence, add the corresponding segment type embedding. For example, for X A , copy and expand the embedding vector of segment A to be the same as X A Matrices of equal length, denoted by E A Then XA and E A Concatenate on the feature dimension to get X′ A .

[0111] The four enhanced input sequences X′ A , X′ B , X′ C , X′ D Splice them into a long sequence X′=[X′ A , X′ B , X′ C , X′ D ].

[0112] Finally, X′ is input into the attention layer;

[0113] The calculation formula of the attention layer is as follows:

[0114] Perform a linear transformation on X′ to obtain the Query matrix Q, Key matrix K and Value matrix V.

[0115] Q=X′W Q

[0116] K=X′W K

[0117] V=X′W V

[0118] in is the learnable weight matrix, d k is the dimension of the attention head.

[0119] Then calculate the attention weight matrix

[0120]

[0121] Finally, the attention weight matrix is ​​multiplied by the Value matrix and linearly transformed to obtain the output autoencoder feature

[0122] Z=AVW O

[0123] in is the output layer weight matrix.

[0124] The graph structure learning layer includes 3 hidden layers and one linear layer. The calculation formula of the lth hidden layer is as follows:

[0125]

[0126] in is the attention weight of node u to node v, N(v) represents the set of nodes that have edges with node v, Represents the first weight parameter of the lth layer, which is calculated by the following formula:

[0127]

[0128] a is a learnable attention vector, ∥ represents the vector concatenation operation, represents the second weight parameter of the lth layer, and LeakyReLU represents the LeakyReLU function;

[0129]

[0130] where x v and x u Represent the feature representation of the vth and uth nodes respectively

[0131] The linear layer is calculated as follows:

[0132] Where Q t Represents global features, It represents the feature representation of node c in the third layer, N(all) represents the set of all nodes, and Concat represents vector concatenation.

[0133] In one embodiment of the present invention, tensor quantization of the tensor layer is to expand the matrix row by row into vectors and then concatenate them.

[0134] The fully connected layer uses a general linear mapping to output the value of the fault point distance ratio.

[0135] In the prior art, the waveform of the fault recording is similar to Figure 5-1 to Figure 5-8 The fault type can be identified by comparing the typical waveforms of the faults shown. Although these typical waveforms of faults are representative, they are generally obtained through tests under power-off conditions. The fault waveforms of online monitoring are far more complex than the typical waveforms and are difficult to identify, which can easily lead to errors in fault type identification.

[0136] In one embodiment of the present invention, the neural network model further includes a second fully connected layer, which is used to input global features and then output a classification result representing the fault type.

[0137] In addition, the cross entropy loss function of the classification result output by the second fully connected layer can be weighted and summed with the prediction loss function of the fault point distance ratio to calculate the comprehensive loss to train the neural network model, which helps to improve the convergence speed of the neural network model and the recognition accuracy of the fault point distance ratio.

[0138] At least one embodiment of the present invention provides a distributed power transmission line fault accurate positioning system, comprising:

[0139] Signal detection terminals installed on the trunk and branches of the transmission line;

[0140] a communication terminal for receiving data from a signal detection terminal, the communication terminal being used to send the data to a system master station;

[0141] The system master station is used to execute steps 102-104 of the aforementioned distributed power transmission line fault accurate location method.

[0142] The main functions of the signal detection terminal provided in this embodiment include:

[0143] Real-time monitoring function

[0144] The signal detection terminal collects the transmission line conductor current, altitude, longitude and latitude in real time, and monitors their changes in real time.

[0145] Scheduled data collection and upload

[0146] The signal detection terminal regularly uploads the measurement data of conductor current, altitude, longitude and latitude, conductor temperature, humidity, etc. to the communication terminal;

[0147] The communication terminal regularly uploads measurement data such as conductor current, altitude, longitude and latitude, conductor temperature, humidity, etc. to the system main station.

[0148] Fault signal collection and upload

[0149] After the condition is triggered, the signal detection terminal completes high-speed sampling of the current, records the fault current waveform and the current traveling wave waveform, and uploads them to the communication terminal;

[0150] After receiving the fault waveform data from the signal detection terminal, the communication terminal screens the data and uploads the valid data to the background master station.

[0151] Data storage and historical data upload

[0152] The communication terminal supports up to 2G Nand Flash and 4G DDRII storage. The standard configuration is 128M Nand Flash and 2G DDRII storage. The standard configuration can realize 100,000*12 SOE event records and 333*12 complete waveform data storage (each waveform recording file is 32*12KB);

[0153] The communication terminal can not only actively upload monitoring data when a line fault occurs, but also record and measure in real time, and supports the function of remote manual retrieval of data by the background master station.

[0154] Equipment self-test

[0155] The signal detection terminal can regularly collect and upload the induction current, GPS online status, and battery / capacitor voltage;

[0156] The signal detection terminal has alarm and information upload functions such as low battery / capacitor voltage, GPS offline, and low short-distance wireless communication signal strength;

[0157] The communication terminal can regularly collect and upload parameters such as solar power supply current (optional), battery voltage, GPS online status, short-distance wireless communication signal strength, GPRS receiving signal strength (quality), etc.

[0158] The communication terminal has alarm and information uploading functions such as low battery voltage, GPS offline, low short-range wireless communication signal strength, and low GPRS reception signal strength (poor quality).

[0159] The communication terminal has software and hardware self-checking functions to prevent the program from "running away" or "deadlocking".

[0160] Overhead line single line configuration scheme

[0161] For overhead lines, a monitoring point is set up at each power source, and the location and number of monitoring points can be flexibly configured on the line. The distance between two adjacent detection points is not less than 5km and not more than 30km. Each monitoring point is equipped with a group of overhead signal detection terminals, with 3 signal detection terminals in each group, which are installed on three-phase conductors respectively.

[0162] Cable line single line configuration scheme

[0163] For cable lines, a monitoring point is set at each end of the cable line, and the distance between the two monitoring points is not less than 5km. A cable-type signal detection terminal is installed at each monitoring point, and the cable-type signal detection terminal can be installed on the wire pole or in the branch box. Each signal detection terminal can be connected to 12 Rogowski coil current transformers, and can simultaneously measure 3 cable lines.

[0164] Overhead and cable hybrid line configuration scheme

[0165] For overhead and cable hybrid lines, configure according to 4.1 and 4.2 respectively.

[0166] Overhead T-connection configuration scheme

[0167] For T-connected lines of overhead lines, both the trunk line and each branch are configured as single-line lines according to the configuration scheme in 4.1.

[0168] Cable line branch box configuration plan

[0169] In the cable line branch box, one cable-type signal detection terminal is installed, which can be equipped with up to 12 Rogowski coil current transformers.

[0170] At least one embodiment of the present invention provides a storage medium storing computer-readable instructions. When the computer-readable instructions are read, the aforementioned method for accurately locating a distributed power transmission line fault can be executed.

[0171] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for accurately locating faults in distributed power transmission lines, characterized in that: include: Step 101, collecting recorded wave data when a transmission line fails; Divide the transmission line into multiple line segments, and collect waveform data at the end points of each data segment; The recording data includes the fault transient recording data of the endpoints of all line sections when a transmission line fails; Step 102, constructing graph structure data based on the collected recorded wave data; The graph structure data includes nodes and edges connecting the nodes. The nodes include trunk nodes and branch nodes. A trunk node represents a line segment of a trunk, and a branch node represents a line segment of a branch. Step 103, inputting the graph structure data into the neural network model, and the neural network model outputs a result indicating the location of the fault point of the transmission line; The neural network model includes: The self-encoding layer is used to input a fault transient recording to output a self-encoding feature. When the fault transient recording of the starting point of the line segment represented by the node is input, the self-encoding feature corresponding to the node is output; The tensor layer is used to quantize the node's autoencoder feature tensor into a vector form and then input it into the graph structure learning layer; The graph structure learning layer is used to input the self-encoded features of the nodes and the graph structure data, and output the global features; A fully connected layer, which is used to input global features and then output a result representing the distance ratio of the fault point; Step 104, calculating the distance between the fault point and the endpoint of the first fault transient recording according to the fault distance ratio, extracting the position of the endpoint of the first fault transient recording, and then locating the fault point.

2. A method for accurately locating a distributed power transmission line fault according to claim 1, characterized in that: The fault transient recording is collected synchronously by all line segments. When the endpoint of any line segment triggers the fault transient recording, the endpoints of all line segments synchronously perform the fault transient recording.

3. A method for accurately locating a distributed power transmission line fault according to claim 1, characterized in that: The generation rules of edges between nodes are as follows: If the line segments represented by two trunk nodes are connected, an edge is generated between the two trunk nodes; If the line segments represented by the trunk node and the branch node are connected, an edge is generated between the trunk node and the branch node; If two branch nodes belong to the same branch, then there is an edge between the two branch nodes.

4. A method for accurately locating a distributed power transmission line fault according to claim 1, characterized in that: The fault point distance ratio is expressed as follows: Among them, L0, + Indicates the length of the line segment before the endpoint where the fault transient recording is first started, L0, - It indicates the length of the line section after the endpoint where the fault transient recording is first started, L0, b indicates the distance between the fault point and the endpoint where the fault transient recording is first started.

5. A method for accurately locating a distributed power transmission line fault according to claim 1, characterized in that: The self-encoding layer inputs the waveform data of segments A, B, C, and D of the fault transient recording data, and then outputs the weighted features of segment A, segment B, segment C, and segment D; Each waveform segment A, B, C, D is itself a time series, denoted by X A , X B , X C , X D ; For each sequence, add the corresponding segment type embedding; The four enhanced input sequences X′ A , X′ B , X′ C , X′ D Splice them into a long sequence X=[X′ A , X′ B , X′ C , X′ D ]; Finally, X′ is input into the attention layer; The calculation formula of the attention layer is as follows: Perform linear transformation on X′ to obtain the Query matrix Q, Key matrix K and Value matrix V; Q=X′W Q K=X′W K V=X′W V in is the learnable weight matrix, d k is the dimension of the attention head; Calculate the attention weight matrix Multiply the attention weight matrix by the Value matrix and transform it linearly to get the output autoencoder feature Z=AVW O in is the output layer weight matrix.

6. A method for accurately locating a distributed power transmission line fault according to claim 1, characterized in that: The graph structure learning layer includes 3 hidden layers and one linear layer. The calculation formula of the lth hidden layer is as follows: in is the attention weight of node u to node v, N(v) represents the set of nodes that have edges with node v, Represents the first weight parameter of the lth layer, which is calculated by the following formula: a is a learnable attention vector, ∥ represents the vector concatenation operation, represents the second weight parameter of the lth layer, and LeakyReLU represents the LeakyReLU function; where x v and x u Represent the feature representation of the vth and uth nodes respectively The calculation formula of the linear layer is as follows: Where Q t Represents global features, It represents the feature representation of node c in the third layer, N(all) represents the set of all nodes, and Concat represents vector concatenation.

7. A method for accurately locating faults in distributed power transmission lines according to claim 1, characterized in that: Tensor quantization of the tensor layer is to expand the matrix row by row into vectors and then concatenate them.

8. A method for accurately locating faults in distributed power transmission lines according to claim 1, characterized in that: The neural network model also includes a second fully connected layer, which is used to input global features and then output a classification result representing the fault type.

9. A distributed transmission line fault accurate positioning system, characterized in that: include: Signal detection terminals installed on the trunk and branches of the transmission line; a communication terminal for receiving data from a signal detection terminal, the communication terminal being used to send the data to a system master station; The system master station is used to execute steps 102-104 of the method for accurately locating a distributed power transmission line fault as described in any one of claims 1-8.

10. A storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are read, a method for accurately locating a distributed power transmission line fault as described in any one of claims 1-8 can be executed.

Citation Information

Patent Citations

  • Primary power distribution network fault identification and positioning method and related equipment

    CN116243110A

  • Overhead transmission line fault diagnosis method and computer readable medium

    CN116644361A

  • Power transmission line fault identification method and system based on environmental characteristics

    CN118209817A

  • Apparatus and process for real-time detection of high-impedance faults in power lines

    US20220373612A1