Power distribution network fault diagnosis and positioning method and system based on deep neural network

Abstract

The application discloses a power distribution network fault diagnosis and positioning method and system based on a deep neural network, which comprises the following steps: constructing a variant of an IEEE33 node system in Simulink, and generating simulation data through different topological structures and distributed power supply configurations. The Attention-LSTM network is used for classification and positioning of fault types and fault line positions, and the design of a double-output head improves the diagnosis efficiency and reduces the calculation consumption. The TCN-BiLSTM network is used for regression of the accurate position of the fault section, and the deficiency of traditional methods in positioning accuracy is made up. Robustness verification is carried out under various topological structures and power supply configurations, and the stability and reliability are proved. The application can accurately and quickly judge the fault type and position in the power distribution network, and provides effective support for fault repair of the power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power systems and their automation technology, and in particular to a two-stage method and system for fault diagnosis and location in distribution networks based on deep neural networks. Background Technology

[0002] The technology of fault diagnosis and location for overhead lines based on voltage and current values ​​is an important part of the development of distribution network technology, especially with the continuous advancement of smart grid technology. With the continuous expansion of distribution network capacity and scale, as well as the access of distributed power sources and diversified loads, the distribution network has changed from "passive" to "active," and the power flow has changed from "unidirectional" to "bidirectional." This has changed the structure and operating characteristics of the distribution network, increasing the complexity of fault diagnosis and location.

[0003] Fault diagnosis and location technologies mainly include three aspects: fault line selection, section location, and fault ranging. Fault line selection technology is relatively mature, but section location and fault ranging technologies are still under development, especially in the automatic location of single-phase grounding faults in low-current grounding distribution networks. Fault diagnosis technology determines whether a fault has occurred and its type and cause by monitoring voltage, current, and other information during system operation; fault location technology uses measured voltage, current, and other parameters to determine the fault area and precise location in a timely and reliable manner.

[0004] In distribution networks with ineffectively grounded neutral points, the weak fault current and indistinct fault characteristics pose challenges to fault diagnosis and location. Research has primarily focused on transient methods and injection-based distance measurement to improve the identification and feature extraction capabilities of weak fault signals. Furthermore, the introduction of modern signal processing techniques, such as wavelet analysis and the Prong algorithm, contributes to enhancing fault signal identification capabilities.

[0005] With the development of technology, automated and intelligent fault location methods are becoming the main direction of future development. These include traveling wave ranging and impedance location methods, which can shorten the fault location time and improve the safety, stability, and economic efficiency of power transmission systems. Summary of the Invention

[0006] This invention addresses the shortcomings of existing technologies by providing a two-stage deep neural network-based method and system for fault diagnosis and location in power distribution networks. It utilizes an Attention-LSTM network to analyze the location and type of faults in the power distribution network under disaster impacts, providing a reference for emergency repair personnel. Repair personnel can prioritize troubleshooting based on the predicted fault location and type. The first stage simultaneously identifies the fault type and fault section, while the second stage quickly determines the precise location of the fault within the section.

[0007] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for fault diagnosis and location in power distribution networks based on deep neural networks includes the following steps:

[0009] S1: Construct a variant of the IEEE 33-node system with distributed power sources in Simulink, set different fault types and locations, obtain three-phase voltage and three-phase current data under simulated fault conditions through simulation, and construct a dataset for network training and testing.

[0010] S2: Construct and train a fault classification and line location network model based on Attention-LSTM. The Attention-LSTM module is used to extract the time-series feature vectors of three-phase voltage and current. The feature vectors output the fault type and the line location of the fault through two output heads, and the cross-entropy loss function is used to train the fault classification and line location network model.

[0011] S3: Construct and train a regression TCN-BiLSTM network model for accurate fault location. The input of the model is the three-phase voltage and current time series data during the line fault period, and the output is the specific location of the line fault. The network model is trained using the Huber loss function to obtain the pre-trained network model for fault location.

[0012] S4: Input the test dataset into the fault classification and line location network model trained in step 2 to predict the fault type and line location in the test dataset, and obtain the fault type and fault location prediction results; if there is no fault, return that the system is normal and end the process; if there is a fault, return the fault time series data, including the fault type and the line location where the fault is located, and execute S5.

[0013] S5: Input the fault time sequence data obtained in step S4 into the fault section location pre-trained network model obtained in step S3 for prediction, and output the accurate predicted value of the fault location, thereby realizing the two-stage diagnosis and location of the distribution network fault.

[0014] Further, step S1 includes:

[0015] Step S11: Construct a variant of the IEEE 33-node system in Simulink and add distributed power sources. Set the load level, load size, main power source and distributed power source types, as well as the active and reactive power of the power sources. Add distributed power sources to each node of the IEEE 33 system to form new rural topology, high-tech zone topology, old city topology and new urban area topology, and set the load level. At the same time, adjust the output of the distributed power sources.

[0016] Step S12: Randomly set different fault types in each branch and use random numbers to determine the precise location of the fault in the line section. During the simulation, faults are randomly set in the thirty-two branches, and the fault type is randomly set in each branch;

[0017] Step S13: For the four topologies of the IEEE 33-node system constructed in Step S11, the fault type and fault location are randomly set using a script, and Simulink is used for simulation to collect the current and voltage data of each line and node when the fault occurs. The collected three-phase voltage and three-phase current time-series waveform data of the thirty-three nodes and thirty-two lines are saved as an Excel file, and the file name records the fault type, the faulty line, and the impedance value of the line during the simulation.

[0018] Step S14: Construct datasets Dataset1 and Dataset2 for network training and testing. Dataset1 is used for network fault classification and line location tasks, and Dataset2 is used for network fault segment precise location tasks.

[0019] Furthermore, the fault types include seventeen types, namely seven open circuit faults, namely A-phase open circuit, B-phase open circuit, C-phase open circuit, AB-phase open circuit, AC-phase open circuit, BC-phase open circuit, and ABC three-phase open circuit; and ten short circuit faults, namely A-phase ground short circuit, B-phase ground short circuit, C-phase ground short circuit, AB-phase short circuit, AC-phase short circuit, BC-phase short circuit, AB-phase ground short circuit, AC-phase ground short circuit, BC-phase ground short circuit, and ABC-phase short circuit.

[0020] Furthermore, the input data of Dataset1 includes eighteen categories, namely, the time-series three-phase voltage and three-phase current of thirty-two branches under seventeen fault and normal states, with the true value being the fault category and fault location; the input data of Dataset2 is the time-series three-phase voltage and three-phase current of the corresponding fault branch, with the true value being the distance of the fault from the load.

[0021] Furthermore, S2 includes:

[0022] S21: Construct a fault classification and fault line location network model for an IEEE 33-node system variant based on Attention-LSTM. The Attention-LSTM module extracts feature vectors from the time-series data of three-phase voltage and current. These feature vectors are output through two headers: one outputs the fault type, and the other outputs the location of the fault on the line. The network model outputs the fault type classification result and the fault line location. Cross-entropy loss is used as the loss function for the classification task.

[0023] S22: Divide the simulation dataset Dataset1 constructed in step S1 into training set, validation set and test set in a ratio of 9:1:10. Use the data in the training set to train the dual-head Attention-LSTM network model. During the training process, use the validation set to judge the convergence of the model and obtain the pre-trained network model M1 for fault classification and line localization.

[0024] Furthermore, the formula for calculating the loss function described in S21 is as follows:

[0025]

[0026] Where M is the number of categories, y o,c It is a binary indicator that returns 1 if class c is the correct classification of sample o, and 0 otherwise; while p o,c It is the probability that the model predicts that sample o belongs to category c.

[0027] Further, S3 includes:

[0028] S31: Construct a network model for precise location in fault sections based on temporal dilated convolution and bidirectional long short-term memory. Use the temporal dilated convolution module to extract feature vectors from the time-series data of three-phase voltage and current. The feature vectors are passed through a bidirectional long short-term memory layer and a fully connected layer, and activated using the Sigmoid activation function.

[0029] S32: Divide the simulation dataset Dataset2 constructed in step S1 into training set, validation set and test set in a ratio of 9:1:10. Use the data in the training set to train the network model of the precise location in the fault section. During the training process, use the validation set to judge the convergence of the model and obtain the pre-trained network model M2 of the precise location in the fault section. The network uses Huber loss as the loss function.

[0030] S33: Input the training dataset Dataset2 obtained in step 1, train the network model, and obtain the pre-trained network model M2 for precise location regression in the fault segment.

[0031] Furthermore, the formula for calculating the loss function described in S32 is as follows:

[0032]

[0033] Where δ is a threshold parameter, and y is the true value. This is a predicted value.

[0034] Further, S51: Input the three-phase current and three-phase voltage of the corresponding fault line in the fault time sequence data obtained in step 4 into the pre-trained network model M2 for accurate location regression in the fault section obtained in step 3, perform forward propagation to calculate the predicted value of accurate location regression in the fault section, and output the accurate predicted value of the fault location on the line.

[0035] This invention also discloses a distribution network fault diagnosis and location system based on a deep neural network, used to execute the above-mentioned distribution network fault diagnosis and location method, the system comprising:

[0036] Simulink Simulation Module: Used to build variants of the IEEE 33-bus system, add distributed power sources, and set different topologies and fault conditions to simulate and generate three-phase voltage and three-phase current time series data for model training and testing;

[0037] The fault diagnosis model building module based on Attention-LSTM network is used to build and train a network model based on Attention-LSTM. This model can extract the time-series features of three-phase voltage and current to classify fault types and locate the line where the fault is located.

[0038] The fault location model building module based on TCN-BiLSTM network is used to build and train a network model based on temporal dilated convolution and bidirectional long short-term memory, which can further accurately locate the specific location in the fault section.

[0039] Model training module: used to train the above fault diagnosis model and fault location model, including dataset partitioning, model training, validation and model optimization processes;

[0040] Model prediction module: It is used to input the test data generated by simulation into the pre-trained model to predict the fault type, fault line location and fault section location, and output the fault diagnosis and location results.

[0041] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned power distribution network fault diagnosis and location method.

[0042] The present invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for fault diagnosis and location in the power distribution network.

[0043] Compared with the prior art, the advantages of the present invention are as follows:

[0044] 1. Improved diagnostic efficiency: The dual-output network model design enables simultaneous prediction of fault type and fault location. This dual-output model reduces computational complexity and improves diagnostic speed, thus enabling a more rapid response to fault conditions in the distribution network.

[0045] 2. Precise Fault Location: A two-stage fault location method is employed. First, an Attention-LSTM network is used for fault type classification and line localization. Then, a TCN-BiLSTM network is used for high-precision regression of the specific fault location. This staged processing significantly improves the accuracy of fault location and overcomes the limitations of traditional methods in precise location.

[0046] 3. Enhanced Robustness: This invention has been validated under different topologies and distributed power source configurations, demonstrating excellent robustness. Through simulation of various real-world operating conditions, the method proves to be able to stably and reliably diagnose and locate faults in multiple environments, adapting to the needs of distribution networks under different operating states.

[0047] 4. Provides strong support: This invention provides strong technical support for the rapid repair of power distribution networks when facing faults or disasters. By accurately determining the type and location of the fault, it can help maintenance personnel quickly locate the fault, reduce repair time, and improve the reliability and recovery capability of the system. Attached Figure Description

[0048] Figure 1 This is a flowchart of the power distribution network fault diagnosis and location method according to an embodiment of the present invention;

[0049] Figure 2 This is a topology diagram of a high-tech industrial park with 33 nodes according to Embodiment 33 of the present invention;

[0050] Figure 3 This is a city center topology map of node 33 in embodiment 3 of the present invention;

[0051] Figure 4 This is the topology diagram of the old city area with node 33 in Embodiment 3 of the present invention;

[0052] Figure 5 This is a new rural topology diagram with 33 nodes according to Embodiment 3 of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0054] The implementation steps of this embodiment are as follows: Figure 1 As shown. This embodiment uses four improved IEEE 33-node distribution network systems as test examples, with the specific structures as follows. Figures 2-5As shown. In the high-tech park scenario, distributed power sources are configured at nodes 11, 18, 21, 24, and 31; in the city center scenario, distributed power sources are configured at nodes 11, 15, 18, 19, 21, 24, 30, and 33; in the old city scenario, distributed power sources are configured at nodes 16, 20, 24, and 27; and in the new rural area scenario, distributed power sources are configured at nodes 5, 10, 14, 18, 20, 22, 24, 27, 30, and 33. Each power source consists of a diesel generator set, thermal power, wind turbine, photovoltaic, and energy storage equipment. A fault occurs randomly at some point in the system. The specific steps of the two-stage deep neural network-based distribution network fault diagnosis and precise location method and system described in this invention are as follows:

[0055] S1: Construct a variant of the IEEE 33-bus system in Simulink, incorporating distributed generation. Set different fault types and locations in the topology, and simulate the three-phase voltage and current data under simulated fault conditions.

[0056] S11: Build a variant of the IEEE 33-node system in Simulink, adding distributed power sources to the nodes of the corresponding scenario. Each power source consists of a wind turbine, a photovoltaic system, and an energy storage device. Configure the load level and load size in the topology.

[0057] S12: Faults are randomly assigned to 32 branches, and the fault type is randomly assigned to each branch. There are 17 possible open-circuit and short-circuit fault types for each line, including: 7 open-circuit faults (A-phase open circuit, B-phase open circuit, C-phase open circuit, AB-phase open circuit, AC-phase open circuit, BC-phase open circuit, ABC three-phase open circuit), and 10 short-circuit faults (A-phase ground short circuit, B-phase ground short circuit, C-phase ground short circuit, AB-phase short circuit, AC-phase short circuit, BC-phase short circuit, AB-phase ground short circuit, AC-phase ground short circuit, BC-phase ground short circuit, ABC-phase short circuit);

[0058] S13: Randomly set the fault type and location using a script, and perform simulation using Simulink to collect current and voltage data for each line and node during the fault. Save the collected three-phase voltage and current time-series waveform data of 33 nodes and 32 lines into an Excel file. The file name should record the fault type, the faulty line, and the line impedance value during the simulation.

[0059] S2: Construct and train a fault classification and fault location network model for an IEEE 33-node system variant based on Attention-LSTM. The Attention-LSTM module extracts feature vectors from the time-series data of three-phase voltage and current. These feature vectors are processed by two output heads: one outputting the fault type, and the other outputting the line location of the fault. The network model outputs the fault type classification and the line on which the fault is located. The dual-head Attention-LSTM network is trained using a cross-entropy loss function.

[0060] S21: Construct a fault classification and fault line location network model for an IEEE 33-node system variant based on Attention-LSTM. The Attention-LSTM module extracts feature vectors from the time-series data of three-phase voltage and current. These feature vectors are processed by two output heads: one outputting the fault type, and the other outputting the line location of the fault. The network model outputs the fault type classification result and the line on which the fault is located. Cross-entropy loss is used as the loss function for the classification task, and the specific calculation formula is as follows:

[0061]

[0062] Where M is the number of categories, y o,c It is a binary indicator (1 if class c is the correct classification of sample o, 0 otherwise), while p o,c It is the probability that the model predicts that sample o belongs to category c.

[0063] S22: The simulation dataset Dataset1 constructed in step 1 is divided into a training set, a validation set, and a test set in a 9:1:10 ratio. The dual-head Attention-LSTM network model is trained using the data from the training set. During the training process, the convergence of the model is judged using the validation set, resulting in a pre-trained network model M1 for fault classification and line localization.

[0064] S3: Construct and train a TCN-BiLSTM network model for regressing the precise location of the fault section. The input to the model is the time-series three-phase voltage and current values ​​during the fault period, and the output is the specific location of the fault. The network uses Huber loss as the loss function, and the training dataset Dataset2 obtained in step 1 is input to train the network model, resulting in a pre-trained network model M2 for regressing the precise location of the fault section.

[0065] S31: Construct a network model for precise location of fault sections based on temporal dilated convolution and bidirectional long short-term memory (TCN-BiLSTM). The TCN module is used to extract feature vectors from the time-series data of three-phase voltage and current. These feature vectors are then passed through a BiLSTM layer and a fully connected layer, activated by a Sigmoid activation function.

[0066] S32: The simulation dataset Dataset2 constructed in step 1 is divided into a training set, a validation set, and a test set in a 9:1:10 ratio. The training set data is used to train the network model for accurately locating the fault sections. During training, the model convergence is assessed using the validation set, resulting in the pre-trained network model M2 for accurately locating the fault sections. The network uses Huber loss as its loss function. The formula for calculating the loss function is as follows:

[0067]

[0068] Where δ is a threshold parameter, and y is the true value. This is the predicted value. Huber loss combines the advantages of MSE and MAE, and is less sensitive to outliers. The optimal value of δ is selected using K-fold cross-validation. First, the dataset is randomly divided into K subsets of equal size; then, training and testing are performed cyclically, with each subset serving as the test set and the remaining K-1 subsets serving as the training set. The model's performance is evaluated on the test set.

[0069] S33: Input the training dataset Dataset2 obtained in step 1, train the network model, and obtain the pre-trained network model M2 for precise location regression in the fault section;

[0070] S4: The pre-trained network model M1 obtained in step 2 for fault classification and line location predicts the fault type and fault location in the test dataset. If there is no fault, it returns that the topology network is running normally and the process ends. If a fault exists, it returns the fault type and the location of the faulty section, i.e., which line it is on. Then proceed to step 5;

[0071] S41: Input the test dataset from Dataset1 into the pre-trained network model M1 for fault classification and line location obtained in step 2 to obtain the fault type and fault location prediction results. If there is no fault, return that the topology network is running normally and the process ends. If there is a fault, return the fault type and the location of the faulty segment, i.e., which line it is. Then proceed to step 5.

[0072] S5: Input the fault time-series data segment obtained in step 4 into the pre-trained network model M2 for precise location regression of the fault section obtained in step 3, and perform precise location regression prediction in the fault section. The model returns a precise predicted value of the fault location. A two-stage method is adopted to model the fault diagnosis and location of the distribution network using a deep learning model, and the model is trained through simulation data.

[0073] S51: Input the three-phase current and three-phase voltage of the corresponding faulty line from the fault time-series data segment obtained in step 4 into the pre-trained network model M2 for precise location regression of the fault section obtained in step 3. Perform forward propagation to calculate the predicted value of precise location regression of the fault section, and output the precise predicted value of the fault location on the line. The location regression adopts the impedance location method, which calculates the fault location based on the impedance value of the line from the two end nodes at the time of the fault.

[0074] This embodiment was tested under various topologies, including different load levels and variations in distributed power output. Through experiments under these diverse conditions, the method demonstrated good generalization performance, consistently identifying fault types and locations.

[0075] In another embodiment of the present invention, a distribution network fault diagnosis and location system based on a deep neural network is provided. This system can be used to implement the above-described distribution network fault diagnosis and location method. The system includes:

[0076] Simulink Simulation Module: Used to build variants of the IEEE 33-bus system, add distributed power sources, and set different topologies and fault conditions to simulate and generate three-phase voltage and three-phase current time series data for model training and testing;

[0077] The fault diagnosis model building module based on Attention-LSTM network is used to build and train a network model based on Attention-LSTM. This model can extract the time-series features of three-phase voltage and current to classify fault types and locate the line where the fault is located.

[0078] The fault location model building module based on TCN-BiLSTM network is used to build and train a network model based on temporal dilated convolution and bidirectional long short-term memory, which can further accurately locate the specific location in the fault section.

[0079] Model training module: used to train the above fault diagnosis model and fault location model, including dataset partitioning, model training, validation and model optimization processes;

[0080] Model prediction module: It is used to input the test data generated by simulation into the pre-trained model to predict the fault type, fault line location and fault section location, and output the fault diagnosis and location results.

[0081] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a power distribution network fault diagnosis and location method.

[0082] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0083] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the power distribution network fault diagnosis and location method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by a processor.

[0084] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the implementation methods of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the present invention.

Claims

1. A method for fault diagnosis and location in power distribution networks based on deep neural networks, characterized in that, Includes the following steps: S1: Construct a variant of the IEEE 33-node system with distributed power sources in Simulink, set different fault types and locations, obtain three-phase voltage and three-phase current data under simulated fault conditions through simulation, and construct a dataset for network training and testing. S2: Construct and train a fault classification and line location network model based on Attention-LSTM. The Attention-LSTM module is used to extract the time-series feature vectors of three-phase voltage and current. The feature vectors output the fault type and the line location of the fault through two output heads, and the cross-entropy loss function is used to train the fault classification and line location network model. S3: Construct and train a regression TCN-BiLSTM network model for accurate fault location. The input of the model is the three-phase voltage and current time series data during the line fault period, and the output is the specific location of the line fault. The network model is trained using the Huber loss function to obtain the pre-trained network model for fault location. S4: Input the test dataset into the fault classification and line location network model trained in step 2 to predict the fault type and line location in the test dataset, and obtain the fault type and fault location prediction results; if there is no fault, return that the system is normal and end the process; if there is a fault, return the fault time series data, including the fault type and the line location where the fault is located, and execute S5. S5: Input the fault time sequence data obtained in step S4 into the fault section location pre-trained network model obtained in step S3 for prediction, and output the accurate predicted value of the fault location, thereby realizing the two-stage diagnosis and location of the distribution network fault.

2. The method for fault diagnosis and location in a power distribution network according to claim 1, characterized in that, Step S1 includes: Step S11: Construct a variant of the IEEE 33-node system in Simulink and add distributed power sources. Set the load level, load size, main power source and distributed power source type, as well as the active and reactive power of the power source. Add distributed power sources to each node of the IEEE 33 system to form a new rural area topology, a high-tech zone topology, an old city topology and a new city topology, and set the load level. At the same time, adjust the output of the distributed power sources. Step S12: Randomly set different fault types in each branch and set the precise location of the fault in the line section by random number; during the simulation, faults are randomly set in the thirty-two branches, and the fault type is randomly set in each branch. Step S13: For the four topologies of the IEEE 33-node system constructed in step S11, the fault type and fault location are randomly set by script, and Simulink is used for simulation to collect the current and voltage data of each line and node when the fault occurs; the collected three-phase voltage and three-phase current time-series waveform data of the thirty-three nodes and thirty-two lines are saved as an Excel file, and the file name records the fault type, the faulty line and the impedance value of the line during the simulation. Step S14: Construct datasets Dataset1 and Dataset2 for network training and testing. Dataset1 is used for network fault classification and line location tasks, and Dataset2 is used for network fault segment precise location tasks.

3. The method for fault diagnosis and location in a power distribution network according to claim 2, characterized in that: The fault types include seventeen types, namely seven open circuit faults, namely A-phase open circuit, B-phase open circuit, C-phase open circuit, AB-phase open circuit, AC-phase open circuit, BC-phase open circuit, and ABC three-phase open circuit; and ten short circuit faults, namely A-phase ground short circuit, B-phase ground short circuit, C-phase ground short circuit, AB-phase short circuit, AC-phase short circuit, BC-phase short circuit, AB-phase ground short circuit, AC-phase ground short circuit, BC-phase ground short circuit, and ABC-phase short circuit.

4. The method for fault diagnosis and location in a power distribution network according to claim 2, characterized in that: The input data for Dataset1 includes eighteen categories, namely, the time-series three-phase voltage and three-phase current of thirty-two branches under seventeen fault and normal conditions, with the true value being the fault category and fault location; the input data for Dataset2 is the time-series three-phase voltage and three-phase current of the corresponding faulty branch, with the true value being the distance of the fault from the load.

5. The method for fault diagnosis and location in a power distribution network according to claim 1, characterized in that, S2 includes: S21: Construct a fault classification and fault line location network model for a variant of the IEEE 33-node system based on Attention-LSTM. The Attention-LSTM module is used to extract feature vectors from the time-series data of three-phase voltage and current. The feature vectors are output by two output heads, one of which outputs the fault type and the other outputs the line location of the fault. The output of the network model is the classification result of the fault type and the location of the fault line. Cross-entropy loss is used as the loss function for the classification task. S22: Divide the simulation dataset Dataset1 constructed in step S1 into training set, validation set and test set in a ratio of 9:1:

10. Use the data in the training set to train the dual-head Attention-LSTM network model. During the training process, use the validation set to judge the convergence of the model and obtain the pre-trained network model M1 for fault classification and line localization.

6. The method for fault diagnosis and location in a power distribution network according to claim 5, characterized in that: The formula for calculating the loss function described in S21 is as follows: Where M is the number of categories, y o,c It is a binary indicator that returns 1 if class c is the correct classification of sample o, and 0 otherwise; while p o,c It is the probability that the model predicts that sample o belongs to category c.

7. The method for fault diagnosis and location in a power distribution network according to claim 1, characterized in that: S3 includes: S31: Construct a network model for the precise location of fault sections based on temporal dilated convolution and bidirectional long short-term memory. Use the temporal dilated convolution module to extract the feature vectors of the time-series data of three-phase voltage and current. The feature vectors are passed through the bidirectional long short-term memory layer and the fully connected layer, and activated by the Sigmoid activation function. S32: Divide the simulation dataset Dataset2 constructed in step S1 into training set, validation set and test set in a ratio of 9:1:

10. Use the data in the training set to train the network model of the precise location in the fault section. During the training process, use the validation set to judge the convergence of the model and obtain the pre-trained network model M2 of the precise location in the fault section. The network uses Huber loss as the loss function. S33: Input the training dataset Dataset2 obtained in step 1, train the network model, and obtain the pre-trained network model M2 for precise location regression in the fault segment.

8. The method for fault diagnosis and location in a power distribution network according to claim 7, characterized in that: The formula for calculating the loss function described in S32 is as follows: Where δ is a threshold parameter; y is the true value. This is a predicted value.

9. The method for fault diagnosis and location in a power distribution network according to claim 1, characterized in that: S5 specifically includes: inputting the three-phase current and three-phase voltage of the corresponding fault line in the fault time sequence data obtained in step 4 into the pre-trained network model M2 for accurate location regression in the fault section obtained in step 3, performing forward propagation to calculate the predicted value of accurate location regression in the fault section, and outputting the accurate predicted value of the fault location on the line.

10. A power distribution network fault diagnosis and location system based on deep neural networks, characterized in that: The system is used to perform the power distribution network fault diagnosis and location method according to any one of claims 1 to 9, the system comprising: Simulink Simulation Module: Used to build variants of the IEEE 33-bus system, add distributed power sources, and set different topologies and fault conditions to simulate and generate three-phase voltage and three-phase current time series data for model training and testing; The fault diagnosis model building module based on Attention-LSTM network is used to build and train a network model based on Attention-LSTM. This model can extract the time-series features of three-phase voltage and current to classify fault types and locate the line where the fault is located. The fault location model building module based on TCN-BiLSTM network is used to build and train a network model based on temporal dilated convolution and bidirectional long short-term memory, which can further accurately locate the specific location in the fault section. Model training module: used to train the above fault diagnosis model and fault location model, including dataset partitioning, model training, validation and model optimization processes; Model prediction module: It is used to input the test data generated by simulation into the pre-trained model to predict the fault type, fault line location and fault section location, and output the fault diagnosis and location results.

Images