Power distribution network single-phase earth fault positioning method and system based on transient traveling waves

Abstract

The invention relates to a power distribution network single-phase earth fault positioning method and system based on transient traveling waves, and the method comprises the steps: 1, collecting a zero-sequence current signal at the moment when a single-phase earth fault occurs, carrying out the decomposition of the collected zero-sequence current signal through a variational mode decomposition algorithm, carrying out the feature extraction through a convolutional neural network, and carrying out the optimization through an alternating direction multiplier method; wherein the zero-sequence current simultaneously comprises a high-frequency traveling wave component and a transient abrupt change characteristic; 2, on the basis of the sampling signals optimized in the step 1, carrying out feature extraction from two dimensions of a transient abrupt change feature and a traveling wave propagation feature, and packaging the features into a fusion vector; 3, identifying a fault line by means of a Grubm angle and field method and the improved DenseNet; and 4, based on the fusion vector in the step 2 and the fault line information determined in the step 3, carrying out iterative operation by adopting an improved sparrow search algorithm, and outputting a finally determined fault section position result. According to the invention, the requirements of intelligent fault sensing and accurate response in a complex power distribution environment are met.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power system fault detection and positioning, and particularly relates to a single-phase grounding fault positioning method and system for distribution networks based on transient traveling waves. BACKGROUND

[0002] With the continuous evolution of new distribution systems, overhead lines, cable lines and distributed power sources are widely connected, making the structure of distribution networks increasingly complex. In medium-voltage distribution networks, especially in small-current grounding systems, most of them belong to single-phase grounding faults. However, due to the weak zero-sequence current, short duration and non-obvious traditional steady-state characteristics, it is difficult for conventional protection devices to accurately identify the faults, resulting in misjudgment and missed judgment. Especially in the multi-branch and multi-feed structure, the traditional algorithm is easily affected by the distribution of electrical parameters, load changes and fault initial phase angle, which can easily lead to line selection errors or positioning deviations. At the same time, although the high-frequency traveling wave signals generated in the initial stage of the fault have excellent distance measuring capability, they depend on high-precision clock synchronization systems and high-speed sampling equipment, and are sensitive to noise, which cannot be widely promoted. On the other hand, machine learning methods based on transient signal characteristics are gradually used for intelligent fault diagnosis of power systems, but the results usually depend on large data model training, which is difficult to meet the dual requirements of real-time and accuracy. Therefore, it is necessary to build a multi-feature perception system that combines transient information and high-frequency traveling wave signals, a system with Beidou time and space synchronization, high-speed data acquisition and intelligent identification capability, which can comprehensively improve the fault line and positioning accuracy of distribution networks, and form a distribution network fault positioning overall solution that takes into account robustness, accuracy and response speed. SUMMARY

[0003] The application proposes a single-phase grounding fault positioning system and method for distribution networks based on transient traveling waves to overcome the shortcomings of the prior art.

[0004] One of the above purposes of the application is achieved by the following technical solution: A single-phase grounding fault positioning method for distribution networks based on transient traveling waves, comprising the following steps: Step 1: Collecting the zero-sequence current signal at the moment of single-phase grounding fault occurrence, decomposing the collected zero-sequence current signal by a variational mode decomposition algorithm VMD, extracting features by a convolutional neural network CNN, and optimizing by an alternating direction multiplier method ADMM; wherein the zero-sequence current contains both high-frequency traveling wave components and transient mutation characteristics; Step 2: Based on the optimized sampling signal of step 1, feature extraction is carried out from two dimensions of transient mutation characteristics and traveling wave propagation characteristics, and is packaged as a fusion vector; Step 3, the fusion vector established in step 2 is converted into a two-dimensional image by the Gram angle and field GASF method, the feature channels are adaptively weighted by using the DenseNet network combined with the SENet attention mechanism, the local and global semantic information in the image is deeply mined, and the DenseNet image feature vector enhanced by the attention mechanism is obtained; these feature vectors not only contain local details and global information, but also highlight the key features of the fault through the adaptive weighting mechanism of SENet, and suppress noise and secondary features; finally, the enhanced DenseNet image feature vector is combined with the fusion vector, and the precise identification of the fault line is realized through the Gini coefficient of the random forest; Step 4, randomly initialize the population of the improved sparrow search algorithm, based on the fusion vector of step 2 and the fault line information determined in step 3, then start multiple rounds of iteration to calculate the optimal fitness position, and update the population position according to the fitness and aggregation state each time. When the maximum number of iterations is reached, the algorithm terminates and outputs the final determined fault section position result, accurately positioning the fault section.

[0005] Moreover, in step 1, the feeder terminal unit (FTU) deployed on the distribution network line collects the original signal at the moment of single-phase ground fault occurrence in real time, realizes 10ns-level clock synchronization between each measuring point based on Beidou satellite timing, and collects signals at a sampling frequency of 10MHz or above.

[0006] Moreover, in step 1, the zero sequence high frequency component related to fault transient is extracted from the current signal, the three-phase instantaneous current ia(t), ib(t), ic(t) is linearly transformed to obtain the zero sequence current i0(t), and the calculation formula is as follows: ; The zero sequence current is de-trended to obtain the processed signal ; the is further normalized: .

[0007] Moreover, in step 1, the normalized zero sequence current signal and the voltage traveling wave signal are preprocessed by applying the variational mode decomposition algorithm; the original signal is decomposed into a group of finite modes with different center frequencies, and the mode components containing transient mutation and high-frequency traveling wave characteristics are extracted: Original signal: x(t); Mode component: ; Each mode satisfies:

[0008] Wherein, is the kth modal component, ωk is its center frequency, and the asterisk is the convolution symbol.

[0009] Moreover, in step 1, the Lagrange multiplier and penalty term are introduced to use the alternating direction multiplier method ADMM to iteratively optimize the frequency domain representation of each mode; in each round of iteration, the mode The update in the frequency domain is as follows:

[0010] The iterative formula for each mode center frequency is:

[0011] in, is the Fourier transform of the original signal in the frequency domain; is the frequency domain representation of the i-th mode; i≠k, which means that when updating the k-th mode, the other modes are considered to be known; is the representation of the Lagrange multiplier in the frequency domain, which is used to constrain the sum of all modes to be equal to the original signal; Penalty factor, which controls the strength of the modal bandwidth constraint; The center frequency of the kth mode; The energy spectral density of the kth mode in the frequency domain.

[0012] Moreover, in step 1, the selected modal components containing key signal components are input into the convolutional neural network, and the network structure used is as follows:

[0013] in: , are the convolution layer parameters, For the The weight parameters of the layer, For the The bias vector of the layer; For the The output features of the layer, For the previous layer ( The output of the layer) is used as the input of the current layer; ∗ is the convolution operation; f() is the nonlinear activation function; The parameter optimization form of the neural network based on ADMM is:

[0014] in, is the set of all parameters of the neural network; is the task loss function; is the regularization term; is the regularization coefficient; , For the matrix and vector in the constraint condition, used for ADMM optimization.

[0015] Moreover, the step 2 comprises: Step 2.1, the current, voltage traveling wave signals obtained by sampling are decomposed by VMD algorithm, and the zero sequence current high frequency mode i0HF, voltage sudden change Δv and energy spectrum change are extracted; Step 2.2, using the nanosecond level time stamp provided by Beidou synchronization, combined with the initial current polarity identification of multiple measuring points, the propagation delay Δt from the fault point to the measuring point is calculated, and the propagation path feature is constructed; Step 2.3, the extracted multi-modal high frequency features are uniformly packaged into fusion vectors, which provide input for the subsequent network model of fault line and positioning, wherein the fusion vector features include time domain mutation, frequency domain energy, spatial propagation delay and initial traveling wave current polarity features.

[0016] Moreover, the step 3 comprises: Step 3.1, the one-dimensional zero sequence current time series signal is converted into a two-dimensional image by using GASF, which comprises: Step 1: scale the time series X={x1,...,xi,...,xd} with length d to the range of [-1,1], and the calculation formula is:

[0017] Step 2: transform the scaled sequence according to the following formula, encode the timestamp as radius, and encode the sequence value as cosine angle, so that the signal is converted to polar coordinate system, and the calculation formula is:

[0018] In the formula: is the polar coordinate angle, is the radius, is the timestamp, and M is the coordinate span adjustment constant; Calculate the cosine function between each point to get the gram angle field, and the calculation method is as follows:

[0019] Step 3.2, deeply dig the features in the image through DenseNet and SENet attention mechanism, and highlight the key information; Wherein, the SENet attention module is composed of squeezing operation and excitation operation; The squeezing operation is:

[0020] In the formula: is the output feature vector, H and W are the length and width of the feature map respectively, is an input matrix; h, c are its rows and columns; the squeezing operation compresses the feature map of WxHxT into a feature vector of 1x1xT, and T is the number of output channels; The excitation operation is as follows:

[0021] In the formula: s is a weight vector, and the dimension is 1x1xT; is an input feature vector; is a ReLU activation function; is a Sigmoid function, and the obtained weight can be normalized; W1 and W2 are two full connection layer operations.

[0022] Moreover, in step 4, the calculation formula of the nonlinear inertia weight factor w used in the iterative calculation process is as follows:

[0023] In the formula, alpha is 0.7, and beta is 0.3; The improved finder position updating formula is as follows:

[0024] The follower updating formula is as follows:

[0025] In the formula, is a global optimal position, is a global worst position; A is a matrix with a value of 1 or -1 ; and The position updating formula of the guard is as follows: .

[0026] The second of the above purposes of the application is realized by the following technical scheme: A power distribution network single-phase ground fault positioning system based on transient traveling wave is used to implement the power distribution network single-phase ground fault positioning method based on transient traveling wave, and comprises a high-precision synchronous sampling unit, a transient-traveling wave fusion feature extraction module, a fault line module and a fault section positioning module. The high-precision synchronous sampling unit constructs a time-space reference with Beidou clock synchronization of <10 ns and high-frequency sampling of >=10 MHz, collects high-frequency traveling wave signals in the initial stage of fault, decomposes the collected zero sequence current signals by a variational mode decomposition algorithm VMD, extracts features by a convolutional neural network CNN, and optimizes by ADMM to form a "synchronization-capture-purification-learning" closed loop, and realize high-fidelity perception of fault signals. The transient-traveling wave fusion feature extraction module is used for fusing time domain transient quantity, frequency domain spectrum and space propagation delay, and constructing a three-dimensional feature vector of time domain mutation, frequency domain energy and space propagation. The fault line module adopts the Gram angle and field method (GASF) to map a one-dimensional zero sequence current sequence into a two-dimensional image, fuses the dense connection structure of DenseNet and the attention mechanism of SENet, deeply mines the local and global semantic information in the image, and simultaneously introduces a random forest to replace a traditional classifier, so that intelligent and accurate identification of the fault line of the distribution network is realized. The fault section positioning module is used for accurately positioning the fault section based on the improved sparrow algorithm and the propagation characteristics of the initial traveling wave current polarity of the fusion vector, so that efficient and accurate identification of the fault of the distribution network is realized.

[0027] The application has the advantages and positive effects that: The single-phase grounding fault positioning system level method of the application constructs an integrated solution of high-precision perception, intelligent identification and rapid positioning driven by the Beidou time scale. The system realizes more than 95% identification accuracy of single-phase grounding faults in a small current grounding system through high-frequency synchronous sampling and multi-modal feature extraction, combined neural network classification and intelligent optimization algorithm; combined with traveling wave initial polarity and propagation time difference modeling, the average response time of fault line and section positioning is less than 1 second, and the positioning error is controlled within ±1.6 meters. The system structure has good topological adaptability and grounding type compatibility, and shows significant robustness in different feeder scenes, supports modular deployment, can be flexibly embedded in intelligent switch stations, distribution rooms and remote master stations, and meets the intelligent fault perception and accurate response requirements in complex distribution environments. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 It is a system block diagram of the single-phase grounding fault positioning system of the application; Figure 2 It is a data acquisition and synchronization flowchart of the application; Figure 3 It is a zero sequence current signal of line L1 and L3 and its GASF image of the embodiment of the application; Figure 4 It is a random forest structure diagram used in the application; Figure 5 It is an improved sparrow algorithm flowchart used in the application. DETAILED DESCRIPTION

[0029] The structure of the application will be further described below in combination with the drawings and through embodiments. It should be noted that the embodiments are descriptive rather than limiting.

[0030] A power distribution network single-phase grounding fault positioning method and system based on transient traveling wave, with Beidou time and space synchronization, high-speed data acquisition and intelligent recognition capability, comprehensively improves the fault line and positioning accuracy of the power distribution network, and forms a power distribution network fault positioning overall solution considering robustness, accuracy and response speed. The power distribution network single-phase grounding fault positioning method of the application first realizes 10ns level clock synchronization between each measuring point based on Beidou satellite timing by a high-precision synchronous sampling unit, captures the high-frequency traveling wave and sudden signal at the initial stage of fault with a sampling frequency of more than 10MHz, and decomposes the collected zero sequence current signal by a variational mode decomposition algorithm VMD, extracts features by a convolutional neural network CNN, and optimizes by an alternating direction multiplier method ADMM; then the transient traveling wave fusion feature extraction module extracts features from the transient sudden variable and the traveling wave propagation characteristic in two dimensions, extracts the transient and traveling wave propagation characteristics based on the sampling signal and encapsulates them as a fusion vector; then the fault line module identifies the fault line with the help of the Gram angle and field method GASF and the improved DenseNet; finally, the fault section positioning module uses the improved sparrow search algorithm to accurately locate the fault section based on the propagation characteristics of the initial traveling wave current polarity, and finally realizes efficient and accurate identification of power distribution network faults.

[0031] The power distribution network single-phase grounding fault positioning system based on transient traveling wave of the application comprises the following parts: 1. High-precision synchronous sampling unit HSTS: The high-precision synchronous sampling module integrates Beidou timing, high-frequency sampling, signal decomposition and feature extraction, etc. Key technologies, and builds a high-fidelity data acquisition and processing front end for power distribution network fault identification, realizes time and space benchmark reinforcement, signal intelligent purification, and feature deep mining fusion innovation. The module realizes <10ns level clock synchronization between measuring points through Beidou satellite, ensures accurate alignment of fault transient signals in time domain, and the sampling frequency reaches more than 10MHz, which can completely capture the high-frequency traveling wave signal at the initial stage of fault. The variational mode decomposition algorithm VMD is introduced to decompose the complex current signal into multiple scale high-frequency modes, which significantly enhances the representation ability of weak sudden variables; combined with the convolutional neural network CNN, the key local waveform features are automatically extracted, and the network convergence and robustness are optimized by the alternating direction multiplier method ADMM, which innovatively constructs a high-precision fault perception unit with synchronous mechanism and deep perception ability.

[0032] 2. Transient-traveling wave fusion feature extraction module TF-FEM: The transient-traveling wave fusion feature extraction module innovatively fuses transient sudden variables and traveling wave propagation information to construct a physical perception feature system for power distribution network faults. The module extracts zero sequence current high frequency mode i0HF, fault phase voltage sudden drop ∆v and frequency domain energy spectrum through VMD to comprehensively capture high frequency disturbance features in the initial stage of the fault; further combined with multi-measurement point initial current polarity identification and Beidou timing support under nanosecond level time difference, the propagation delay ∆t is accurately calculated to depict the propagation path of electromagnetic waves in the network. Finally, the time domain mutation, frequency domain features and spatial propagation behavior are fused and packaged into a unified feature vector to construct a fault behavior chain from disturbance generation to path evolution, realizing deep perception and accurate modeling of fault type and location. Through the three-dimensional fusion of "time domain mutation + frequency energy + spatial propagation", TF-FEM first realizes the leap from signal analysis to behavior chain cognition of power distribution network fault.

[0033] 3. Fault line module GASF-DenseNet: The fault line module innovatively converts the time domain waveform of the zero sequence current of the power distribution network into visual image features to realize accurate identification of the fault line under weak signal conditions. The module uses the GASF to map the one-dimensional zero sequence current sequence into a two-dimensional image, effectively enhancing the separability and visual expression of fault features; it combines the dense connection structure of DenseNet and the attention mechanism of SE-Net to deeply mine the local and global semantic information in the image; at the same time, it introduces random forest to replace the traditional classifier to improve the robustness and classification stability of the model under complex conditions, and constructs a high-precision line selection system from signal imaging to intelligent identification. GASF-DenseNet converts fault signals into image space (GASF) + dense connection feature reuse (DenseNet) to realize intelligent and accurate identification of fault lines in power distribution networks.

[0034] 4. Fault section positioning module ISSA-Locator: The fault section positioning module constructs a high-precision and efficient fault section search system for power distribution networks by introducing an intelligent optimization mechanism. Based on the improved ISSA, the module combines the nonlinear inertia factor weight regulation to dynamically balance the global exploration and local convergence ability, significantly improving the positioning accuracy and algorithm stability; innovatively uses the initial current polarity and high-frequency energy difference as fitness evaluation indexes to accurately represent the physical area where the fault occurs; the algorithm internally cooperates and iterates through "discoverer-follower-sentry" multi-agent to quickly complete the optimal search and identification of the fault section, and constructs an integrated closed-loop mechanism from perception triggering to intelligent positioning. The nonlinear inertia weight factor w is introduced to improve the discoverer position update formula of the sparrow search algorithm, realizing the accurate positioning of the fault section by ISSA.

[0035] The application is based on a power distribution network single-phase grounding fault positioning method based on transient traveling wave, comprising the following steps: Step 1, collect the zero sequence current signal at the moment of single-phase grounding fault, decompose the collected zero sequence current signal by VMD, extract features by CNN, and optimize by ADMM; wherein the zero sequence current contains high-frequency traveling wave components and transient mutation characteristics, and the specific implementation is as follows: Firstly, the system relies on the Beidou satellite timing module to build a high-precision clock synchronization mechanism, realizes 10ns level clock synchronization accuracy between each monitoring terminal, and guarantees the alignment consistency of sampling data in the time domain from the source, providing a time reference for traveling wave propagation time difference analysis and fault location. The basic principle is as follows:

[0036] Wherein, Δx is the position error caused by clock error, is the traveling wave propagation speed, and the positioning error can be controlled within about 1.6 meters.

[0037] In order to completely capture the transient signal and traveling wave signal at the initial stage of fault, the system sampling frequency is set to 10MHz and above, the transient signal mainly occurs within tens of microseconds at the initial stage of fault, and the frequency spectrum covers hundreds of kHz to several MHz; the high-frequency traveling wave frequency band is concentrated in the range of 100kHz~5MHz. Taking zero sequence current as the key index, the zero sequence high-frequency component related to fault transient is extracted from the current signal, the three-phase instantaneous current ia(t), ib(t), ic(t) is linearly transformed to obtain the zero sequence current i0(t), and the calculation formula is as follows:

[0038] Then, the zero sequence current is de-trended to eliminate DC bias and low-frequency drift interference, improve the sensitivity to mutation components, and obtain the processed signal Further normalized processing, the standard form is:

[0039] The normalized result is input to the VMD unit as the signal to be decomposed.

[0040] The timing reference innovation of 10ns level high-precision clock synchronization, the signal integrity guarantee of more than 10MHz, and the feature extraction optimization of zero sequence current processing and VMD modal decomposition are realized.

[0041] ​Then, the normalized zero-sequence current signal and voltage traveling wave signal are preprocessed using the variational mode decomposition (VMD) algorithm. The variational mode decomposition (VMD) algorithm decomposes the original signal into a set of finite modes (IMFs) with different center frequencies, extracting the modal components containing transient mutations and high-frequency traveling wave characteristics: Original signal: x(t) Modal components:

[0042] Each mode satisfies:

[0043] Where uk(t) is the kth modal component, ωk is its center frequency, and the asterisk is the convolution symbol.

[0044] In order to improve the stability of decomposition, Lagrange multipliers and penalty terms are introduced to use the alternating direction multiplier method ADMM to iteratively optimize the frequency domain representation of each mode. The update in the frequency domain is as follows:

[0045] The iterative formula for each mode center frequency is:

[0046] The joint optimization innovation of the variational mode decomposition algorithm VMD-ADMM is implemented to reduce the error in traveling wave mode extraction.

[0047] Then, the selected modal components containing key signal components are input into the convolutional neural network. Leveraging the convolutional neural network's advantages in local structure perception, the network automatically learns transient waveforms and high-frequency change patterns: the convolutional layer perceives local changes and identifies representative waveform structures in zero-sequence current; the activation function preserves nonlinear mutation characteristics; and the pooling layer compresses invalid information to improve representation capabilities. The network structure is as follows:

[0048] in: , are the convolution layer parameters, For the The weight parameters of the layer, For the The bias vector of the layer; For the The output features of the layer, For the previous layer ( The output of the layer) is used as the input of the current layer; ∗ is the convolution operation; f() is the nonlinear activation function; The ADMM-based neural network parameter optimization introduces an alternating direction multiplier method (ADMM) to constraint decomposition and joint solution of the parameter optimization process of a convolutional neural network (CNN) to improve convergence and robustness of model training.

[0049] wherein, is a set of all parameters of the neural network; is a task loss function; is a regularization term; is a regularization coefficient; , is a matrix and a vector in a constraint condition, used for ADMM optimization.

[0050] By introducing auxiliary variables and Lagrange multipliers, the ADMM decomposes the optimization objective into sub-problems to be solved iteratively, thereby improving stability and generalization capability of parameter updating.

[0051] Step 2, the transient mutation variable and the traveling wave propagation characteristic are extracted from the transient mutation variable and the traveling wave propagation characteristic through the transient wave fusion feature extraction module, and the transient and traveling wave propagation features are extracted based on the sampling signal and packaged into a fusion vector, and the specific implementation is as follows: The traditional method is mostly characterized by steady-state amplitude or instantaneous peak value, and the present application improves the physical consistency description of the fault evolution process through joint modeling of modal time-frequency features and traveling wave propagation features. The fault wave propagation trajectory atlas is constructed by using the propagation time difference and the spatial distribution data of the measuring points, and the spatial reasoning and matching positioning of the fault location are realized. Specifically, the following steps are included: Step 2.1, the current and voltage traveling wave signals obtained by sampling are decomposed by a variational modal decomposition algorithm (VMD) to extract zero sequence current high frequency mode i0HF, voltage sudden drop amount Δv, energy spectrum change, etc. Step 2.2, the nanosecond-level time stamp provided by Beidou synchronization is used to calculate the propagation delay Δt from the fault point to the measuring point, and the propagation path feature is constructed in combination with the initial current polarity identification of the multi-measuring point. Step 2.3, the extracted multi-modal high-frequency features are fused and packaged into a unified feature vector for behavior chain reconstruction and subsequent positioning network modeling, wherein the multi-modal high-frequency features include time-domain mutation, frequency-domain energy, and spatial propagation delay.

[0052] Step 3, the fusion vector established in step 2 is converted into a two-dimensional image by the Gram angle and field GASF method, the time sequence characteristics are converted into a two-dimensional image, a DenseNet network combined with an SENet attention mechanism is used to adaptively weight the feature channels, local and global semantic information in the image is deeply mined, and a DenseNet image feature vector enhanced by the attention mechanism is obtained; these feature vectors not only contain local details and global information, but also highlight key features of the fault through the adaptive weighting mechanism of the SENet, and suppress noise and secondary features; finally, the enhanced DenseNet image feature vector is combined with the fusion vector, and the Gini coefficient of the random forest is used to realize accurate identification of the fault line. The specific implementation is as follows: Step 3.1, convert the one-dimensional zero sequence current time sequence signal into a two-dimensional image; Due to the weak fault characteristics of the small current grounding system and the nonlinear characteristics of the zero sequence current signal, various processing methods in one-dimensional time domain cannot achieve good results. Therefore, the one-dimensional time sequence signal can be first converted into a two-dimensional image signal, so as to better exhibit the detailed features of the fault signal. GASF transformation is a method of converting one-dimensional time series into two-dimensional images. This transformation can represent the relationship between the local and global of the sequence. The transformed image not only preserves the time correlation on the diagonal line, but also reflects the nonlinear characteristics of the signal. Using GASF transformation to convert the zero sequence current signal into a two-dimensional image provides a prerequisite for subsequent neural network feature mining.

[0053] The GASF transformation steps are as follows: Step 1: Scale the time series X = {x1,..., xi,..., xd} of length d to the range [-1, 1], and the calculation formula is:

[0054] Step 2: Transform the scaled sequence according to the following formula, encode the time stamp as the radius, and encode the sequence value as the cosine angle, so that the signal is converted to the polar coordinate system. The calculation formula is:

[0055] In the formula: is the polar coordinate angle, is the radius, is the time stamp, and M is the coordinate span adjustment constant. The encoded polar coordinate points correspond to each other with the time series, ensuring the uniqueness of the encoding.

[0056] Calculate the cosine and function between each point to get the Gram angle and field, and the calculation method is as follows:

[0057] The single-phase ground fault of the line L1 is simulated on the simulation software, and the zero sequence current signal waveforms of the fault line L1 and the non-fault line L3 are taken for GASF transformation, and the results are shown in Figure 3

[0058] Through GASF transformation, the zero sequence current signals of the fault line and the non-fault line show obvious differences in the image, such as color distribution and depth, thereby providing a more intuitive and effective data basis for subsequent feature extraction and classification.

[0059] Step 3.2, deeply dig features in the image through DenseNet and SENet attention mechanism, and highlight key information The residual neural network can solve the problems of gradient disappearance and gradient explosion by using residual units, and DenseNet can avoid the problems of network gradient disappearance and gradient explosion by feature reuse and jump connection. DenseNet is composed of dense blocks (DB) and transition layers (TL) in series. The dense block is the core structure of DenseNet. Each layer in the dense block is connected to all subsequent layers in a dense connection, realizing feature reuse.

[0060] In order to improve the feature extraction ability of the line selection model, the DenseNet neural network is improved. The SENet attention module can highlight meaningful data features and suppress the attention of secondary features by the network layer, and at the same time, the complexity of the SENet attention module is small, which can be easily integrated into the existing network.

[0061] The SENet attention module is composed of squeezing operation and excitation operation.

[0062] (1) Squeezing operation is:

[0063] In the formula: is the output feature vector, H and W are the length and width of the feature map, is the input matrix; h, c are its rows and columns. The squeezing operation compresses the feature map of WxHxT into a feature vector of 1x1xT, and T is the output channel number.

[0064] (2) Excitation operation: Similar to the mechanism of the door in the recurrent neural network, each feature channel can be adaptively weighted according to the loss function, and the excitation operation is as follows:

[0065] In the formula: s is the weight vector, with a dimension of 1x1xT; is the input feature vector; ​ReLU activation function; Sigmoid function, the obtained weight can be normalized; W1 and W2 are two full connection layer operations, which play a role in fusing feature values of each channel.

[0066] Step 3.3 adopts a random forest algorithm as a classifier to accurately identify the fault line based on the extracted features.

[0067] In order to shorten the time of the fault line, the random forest structure is integrated into the neural network, so that the neural network can have certain anti-noise ability without the pre-posed data noise reduction link. Random forest (RF) is an ensemble learning algorithm based on decision trees, and its structure is shown in Figure 4 The random forest is composed of many decision trees, which makes the random forest have good anti-noise ability.

[0068] The random forest algorithm well avoids the overfitting problem of a single decision tree and can deal with high-dimensional data and a large number of features. When it is used as a classifier of the neural network, the generalization ability of the neural network can be improved. The decision tree in the random forest selects features based on the Gini coefficient. In the process of constructing the decision tree, the Gini coefficient is used to evaluate the purity of the node to select the optimal split feature. The Gini coefficient ranges from 0 to 1. When all samples under a node belong to the same class, the Gini coefficient is 0, and the node has the highest purity and the best classification effect. On the contrary, when the Gini coefficient tends to 1, it indicates that the node classification effect is worse. Let the number of sample categories be a, and the probability that a sample belongs to category j be pj, then the Gini coefficient of the probability distribution is:

[0069] Step 4, randomly initialize the population of the improved sparrow search algorithm, based on the fusion vector of step 2 and the fault line information determined in step 3, then start multiple rounds of iteration to calculate the optimal fitness position, and update the population position according to the fitness and aggregation state each time. When the maximum number of iterations is reached, the algorithm is terminated and the final determined fault section position result is output, which accurately locates the fault section, and the specific process is as follows: In order to overcome the problem of slow iteration speed of the original algorithm in fault section positioning, an optimized sparrow search algorithm is adopted. The best result of the last iteration is combined into the finder position update formula, and a nonlinear inertia weight factor w is introduced to balance the convergence speed and global search range of the algorithm. With the increase of the inertia weight factor w, it is helpful to expand the range of global search; while with the decrease of the inertia weight factor w, it is more conducive to strengthening the ability of local search. The population is initialized based on the FTU fault current data, and the optimal fitness position is calculated through multiple iterations to realize efficient and accurate positioning of the fault section. The calculation formula of the nonlinear inertia weight factor w is as follows:

[0070] where α is 0.7 and β is 0.3.

[0071] The improved finder position updating formula is as follows:

[0072] When the finder individual locates the suspected fault position, the follower immediately updates its own position and approaches the optimal position, realizing the rapid aggregation of the population in the solution space to the fault area, thereby accelerating the positioning speed and accuracy of the algorithm in the fault section. The position updating formula is as follows:

[0073] where is the global optimal position, is the global worst position; is a matrix with a value of 1 or -1; when , the ith follower does not gather to the fault position, and it needs to approach other suspected fault places; otherwise, it means that the follower is in the best position, and it can search for fault places around.

[0074] 10%-20% of the individuals in the population are aware of the danger. The positions of these individuals in the population are randomly generated, and they are called sentinels. The position updating formula of the sentinels is as follows:

[0075] The population position is updated according to the fitness and aggregation state every iteration. When the maximum number of iterations is reached, the algorithm is terminated and the final determined fault section position result is output, realizing the fault point section positioning under the support of Beidou high-precision time tag. The fault section positioning process of the improved sparrow algorithm is as shown in Figure 5

[0076] ​​In summary, the system is aimed at the problem of single-phase ground fault perception and positioning in small current grounding systems, and an integrated intelligent fault positioning system is constructed, which integrates Beidou clock synchronization, high-frequency sampling, advanced signal processing and intelligent identification technology. The system uses a Beidou satellite timing module to achieve 10 ns level high-precision clock synchronization of the monitoring terminal, providing a unified time reference for traveling wave propagation analysis; through high-frequency sampling above 10 MHz, the fault initial transient and traveling wave signals are accurately captured, and the zero sequence current is taken as the core analysis object, and its high-frequency component is extracted, and the multi-modal features are obtained by VMD decomposition to suppress interference and highlight mutation information. The decomposed modal signals are input into the improved CNN model, and the ADMM optimization algorithm is combined to improve the learning stability and non-linear feature perception ability; at the same time, the GASF transform is introduced to image the one-dimensional signal, which further enhances the expression effect of weak fault features in the spatial domain. In order to improve the expression and robustness of the network, the system combines DenseNet and SE-Net attention mechanism to realize multi-scale feature reuse and channel adaptive weighting, and enhance the attention to key structures. After fault classification, the system uses an intelligent optimization method based on the improved ISSA algorithm to efficiently complete fault line and section positioning with the support of the random forest model. The algorithm improves search accuracy and convergence efficiency through nonlinear inertia weight and iteration behavior strategy, and realizes meter-level spatial positioning error control combined with the Beidou time scale, and fully realizes the accurate perception and rapid positioning of complex fault signals in distribution networks, with high robustness, high sensitivity and engineering deployment adaptability.

[0077] Although the embodiments of the present application and the drawings are disclosed for the purpose of illustration, those skilled in the art can understand that various substitutions, changes and modifications are possible without departing from the spirit of the present application and the appended claims, therefore, the scope of the present application is not limited to the disclosed content of the embodiments and the drawings.

Claims

1. A method for locating single-phase grounding faults in a distribution network based on transient traveling waves, characterized by: The steps include: Step 1: Collect the zero-sequence current signal at the moment of a single-phase grounding fault, decompose it using the variational mode decomposition algorithm (VMD), extract features using the convolutional neural network (CNN), and optimize it using the alternating direction multiplier method (ADMM). The zero-sequence current contains both high-frequency traveling wave components and transient mutation characteristics. Step 2: Based on the sampled signal optimized in step 1, feature extraction is performed from two dimensions: transient mutation characteristics and traveling wave propagation characteristics, and the features are encapsulated into a fusion vector. Step 3: The fusion vector created in Step 2 is converted into a two-dimensional image using the Gram angle and field (GASF) method. A DenseNet network combined with the SENet attention mechanism is used to adaptively weight feature channels, deeply mining local and global semantic information in the image to obtain a DenseNet image feature vector enhanced by the attention mechanism. Finally, the enhanced DenseNet image feature vector is combined with the fusion vector, and the Gini coefficient of the random forest is used to accurately identify the fault line. Step 4: Randomly initialize the population of the improved sparrow search algorithm. Based on the fusion vector from step 2 and the fault line information determined in step 3, multiple rounds of iterations are then initiated to calculate the optimal fitness position. Each iteration updates the population position based on fitness and aggregation status. When the maximum number of iterations is reached, the algorithm terminates and outputs the final determined fault section location result, accurately locating the fault section.

2. The method for locating single-phase grounding faults in distribution networks based on transient traveling waves according to claim 1, characterized in that: In step 1, the feeder terminal unit (FTU) deployed on the distribution network line collects the original signal at the moment of single-phase grounding fault in real time, realizes 10ns clock synchronization between each measuring point based on Beidou satellite timing, and collects signals at a sampling frequency of more than 10MHz.

3. The method for locating single-phase grounding faults in distribution networks based on transient traveling waves according to claim 1, characterized in that: In step 1, the zero-sequence high-frequency component related to the fault transient is extracted from the current signal. The three-phase instantaneous currents ia(t), ib(t), and ic(t) are linearly transformed to obtain the zero-sequence current i0(t). The calculation formula is as follows: ； Detrend the zero-sequence current to obtain the processed signal ;right Further normalization: 。 4. The method for locating single-phase grounding faults in a distribution network based on transient traveling waves according to claim 1, characterized in that: In step 1, the normalized zero-sequence current signal and voltage traveling wave signal are preprocessed using the variational mode decomposition algorithm. The original signal is decomposed into a set of finite modes with different center frequencies, and the modal components containing transient mutations and high-frequency traveling wave characteristics are extracted: Original signal: x(t); Modal components: ; Each mode satisfies: ； in, is the kth modal component, ωk is its center frequency, and the asterisk is the convolution symbol.

5. The method for locating single-phase grounding faults in distribution networks based on transient traveling waves according to claim 1, characterized in that: In step 1, the Lagrange multiplier and penalty term are introduced to use the alternating direction multiplier method ADMM to iteratively optimize the frequency domain representation of each mode; in each round of iteration, the mode The update in the frequency domain is as follows: ； The iterative formula for each mode center frequency is: ； in, is the Fourier transform of the original signal in the frequency domain; is the frequency domain representation of the i-th mode; i≠k, which means that when updating the k-th mode, the other modes are considered to be known; is the representation of the Lagrange multiplier in the frequency domain, which is used to constrain the sum of all modes to be equal to the original signal; Penalty factor, which controls the strength of the modal bandwidth constraint; The center frequency of the kth mode; The energy spectral density of the kth mode in the frequency domain.

6. The method for locating single-phase grounding faults in a distribution network based on transient traveling waves according to claim 1, characterized in that: In step 1, the selected modal components containing key signal components are input into the convolutional neural network. The network structure used is as follows: ； in: , are the convolution layer parameters, For the The weight parameters of the layer, For the The bias vector of the layer; For the The output features of the layer, is the output of the previous layer, which serves as the input of the current layer; ∗ is the convolution operation; f() is the nonlinear activation function; The parameter optimization form of the neural network based on ADMM is: ； in, is the set of all parameters of the neural network; is the task loss function; is the regularization term; is the regularization coefficient; , are the matrices and vectors in the constraints used for ADMM optimization.

7. The method for locating single-phase grounding faults in a distribution network based on transient traveling waves according to claim 1, characterized in that: The step 2 includes: Step 2.1, perform variational mode decomposition (VMD) on the sampled current and voltage traveling wave signals to extract the zero-sequence current high-frequency mode i0HF, the voltage sag Δv, and the energy spectrum change; Step 2.2: Using the nanosecond timestamps provided by BeiDou synchronization and the initial current polarity identification at multiple measuring points, calculate the propagation delay Δt from the fault point to the measuring point and construct the propagation path characteristics. Step 2.3: The extracted multimodal high-frequency features are uniformly encapsulated into a fusion vector to provide input for the subsequent fault line and positioning network model. The fusion vector features include time domain mutation, frequency domain energy, spatial propagation delay, and initial traveling wave current polarity characteristics.

8. The method for locating single-phase grounding faults in a distribution network based on transient traveling waves according to claim 1, characterized in that: The step 3 comprises: Step 3.1: Use the Gram angle sum field method (GASF) to convert the one-dimensional zero-sequence current time series signal into a two-dimensional image, including: Step 1: Scale the time series X = {x1, ..., xi, ..., xd} of length d to the range of [-1, 1], and the calculation formula is: ； Step 2: Transform the scaled sequence according to the following formula: encode the timestamp as the radius and the sequence value as the cosine angle to convert the signal to the polar coordinate system. The calculation formula is: ； Where: is the polar coordinate angle, is the radius, is the timestamp, M is the coordinate span adjustment constant; Computing the cosine sum function between each point gives the Gram angle sum field, which is calculated as follows: ； Step 3.2: Deeply mine the features in the image through the DenseNet and SENet attention mechanisms to highlight key information; Among them, the SENet attention module consists of a squeeze operation and an excitation operation; The extrusion operation is: ； Where: is the output feature vector, H and W are the length and width of the feature map respectively, is the input matrix; h and c are its rows and columns; the squeezing operation compresses the W×H×T feature map into a 1×1×T feature vector, where T is the number of output channels; The incentive operation is as follows: ； Where: s is the weight vector with dimension 1×1×T; is the input feature vector; is the ReLU activation function; is the Sigmoid function; W1 and W2 are two fully connected layer operations.

9. The method for locating single-phase grounding faults in a distribution network based on transient traveling waves according to claim 1, characterized in that: In step 4, the calculation formula of the nonlinear inertia weight factor w used in the iterative calculation process is as follows: ； Where α is 0.7, β is 0.3; The improved discoverer position update formula is as follows: ； The follower update formula is as follows: ； Where, is the global optimal position, is the global worst position; , A is a value of 1 or -1 Matrix of The position update formula of the sentinel is as follows: 。 10. A single-phase ground fault location system for distribution network based on transient traveling waves, characterized by: A method for locating single-phase grounding faults in a distribution network based on transient traveling waves, used to implement any one of claims 1 to 9, comprising a high-precision synchronous sampling unit, a transient-traveling wave fusion feature extraction module, a fault line module, and a fault section locating module; The high-precision synchronous sampling unit uses Beidou's <10ns clock synchronization and ≥10MHz high-frequency sampling to establish a time-space benchmark, collects high-frequency traveling wave signals at the initial stage of a fault, decomposes the collected zero-sequence current signal using the variational mode decomposition algorithm (VMD), extracts features using a convolutional neural network (CNN), and optimizes it using ADMM, forming a synchronization-capture-purification-learning closed loop to achieve high-fidelity perception of fault signals. The transient-traveling wave fusion feature extraction module is used to fuse the time domain transient quantity, frequency domain energy spectrum and spatial propagation delay to construct a three-dimensional feature vector of time domain mutation + frequency domain energy + spatial propagation; The fault line module uses the Gram Angle Sum Field method (GASF) to map the one-dimensional zero-sequence current sequence into a two-dimensional image. It integrates the dense connection structure of DenseNet and the attention mechanism of SENet to deeply mine the local and global semantic information in the image. At the same time, random forest is introduced to replace traditional classifiers to achieve intelligent and accurate identification of fault lines in the distribution network; The fault section locating module is used to determine the fault direction and accurately locate the fault section based on the propagation characteristics of the initial traveling wave current polarity of the fusion vector based on the improved sparrow algorithm, thereby achieving efficient and accurate identification of distribution network faults.

Images