Distribution line fault positioning method and system based on multi-scale time-frequency feature extraction

Abstract

The invention discloses a power distribution line fault positioning method and system based on multi-scale time-frequency feature extraction, and relates to the technical field of line fault monitoring, and the method comprises the steps: obtaining an electrical signal of a key node of a power distribution line, and carrying out the denoising and standardization preprocessing; extracting multi-scale time-frequency characteristics of the preprocessed electrical signals; encoding the topological structure and node features of the power distribution network through a graph neural network to obtain abstract features of each node; taking the extracted multi-scale time-frequency features and the abstract features of each node as input, carrying out training learning based on a long-short-term memory network, and obtaining a trained fault positioning model with an output distribution line fault positioning function; and inputting the multi-scale time-frequency features extracted in real time and the abstract features of each node into the trained fault positioning model, and outputting the fault position of the distribution line. According to the invention, accurate output of the fault position is realized through multi-scale time-frequency feature extraction.

Description

Technical Field

[0001] The present invention relates to the technical field of line fault monitoring, and in particular to a distribution line fault location method and system based on multi-scale time-frequency feature extraction. Background Art

[0002] As the terminal link of the power grid, the accuracy of fault location on distribution lines directly affects power supply reliability and the development of smart grids. As the complexity of distribution networks increases, traditional fault location methods are unable to meet the requirements.

[0003] Traditional fault location methods, such as the impedance method, are susceptible to line parameters and load current. The traveling wave method requires high-quality detection hardware and complex waveform recognition. While time-frequency analysis techniques (such as wavelet transforms) have been used, single-scale analysis fails to fully capture fault characteristics, resulting in insufficient location accuracy. Multi-scale time-frequency feature extraction technology can analyze signals in layers, characterizing fault information from multiple dimensions and overcoming the limitations of traditional methods.

[0004] In view of this, a distribution line fault location method and system based on multi-scale time-frequency feature extraction is needed. Summary of the Invention

[0005] To address the problem in existing technologies where single-scale analysis fails to fully capture fault characteristics, resulting in insufficient positioning accuracy, the present invention provides a distribution line fault positioning method and system based on multi-scale time-frequency feature extraction. This method achieves accurate output of the fault location through multi-scale time-frequency feature extraction. The specific technical solution is as follows:

[0006] The present invention provides a distribution line fault location method based on multi-scale time-frequency feature extraction, comprising:

[0007] Acquire electrical signals from key nodes of distribution lines and perform denoising and standardization preprocessing;

[0008] Extracting multi-scale time-frequency features of preprocessed electrical signals;

[0009] The topological structure and node characteristics of the distribution network are encoded through graph neural networks to obtain the abstract characteristics of each node;

[0010] The extracted multi-scale time-frequency features and the abstract features of each node are used as input, and training is performed based on the long short-term memory network to obtain a fault location model that is trained to locate output distribution line faults.

[0011] The multi-scale time-frequency features extracted in real time and the abstract features of each node are input into the trained fault location model to output the fault location of the distribution line.

[0012] Preferably, the extracting of multi-scale time-frequency features of the preprocessed electrical signal includes:

[0013] The electrical signal is decomposed into multi-scale sub-bands using wavelet transform, and the energy and variance characteristics of each sub-band signal are extracted;

[0014] Obtain the intrinsic mode function of the electrical signal through HHT transformation, and extract the instantaneous amplitude characteristics and instantaneous frequency characteristics;

[0015] The features extracted by wavelet transform and HHT transform are fused to form multi-scale time-frequency features.

[0016] Preferably, decomposing the electrical signal into multi-scale sub-bands by using wavelet transform and extracting the energy characteristics and variance characteristics of each sub-band includes:

[0017] The pre-processed electrical signal is decomposed layer by layer through discrete wavelet transform to obtain the low-frequency approximate coefficient sequence and high-frequency detail coefficient sequence of each layer decomposition result;

[0018] The energy characteristics and variance characteristics of each subband signal are calculated based on the low-frequency approximation coefficient sequence and the high-frequency detail coefficient sequence.

[0019] Preferably, the obtaining of the intrinsic mode function of the electrical signal by HHT transformation and extracting the instantaneous amplitude feature and the instantaneous frequency feature includes:

[0020] Decomposing the preprocessed electrical signal into intrinsic mode functions by empirical mode decomposition;

[0021] The decomposed intrinsic mode function is subjected to Hilbert transform to obtain the instantaneous amplitude characteristics and instantaneous frequency characteristics.

[0022] Preferably, the topological structure and node features of the distribution network are encoded by the graph neural network to obtain the abstract features of each node, including:

[0023] The distribution network is abstracted into graph data, where nodes and edges represent busbars and feeder branches respectively;

[0024] The structural information of the graph and the attribute characteristics of the nodes are respectively input into the VGAE model built based on GCN. The node characteristics are mapped to independent Gaussian distributions through the model to obtain the abstract characteristics of each node in the graph.

[0025] Preferably, the structural information of the graph and the attribute features of the nodes are respectively input into the VGAE model built based on GCN, and the node features are respectively mapped to independent Gaussian distributions through the model to obtain the abstract features of each node in the graph, including:

[0026] According to the structural information of the distribution network graph data, the adjacency matrix is ​​obtained;

[0027] Obtain the amplitude and phase angle of the three-phase voltage, the amplitude and phase angle of the three-phase current, the active power and the reactive power of each node in the distribution network. Based on graph theory, calculate the degree centrality, eigenvector centrality, betweenness centrality and closeness centrality of each node to obtain the eigenvector matrix of each node.

[0028] The adjacency matrix and eigenvector matrix are respectively input into the VGAE model built on GCN. The encoder part of VGAE is used to learn the mean and variance of the Gaussian distribution mapped to the node features based on a two-layer graph convolutional network.

[0029] According to the mean and variance of the node features, the sampling results obtained from the corresponding Gaussian distribution are used as the abstract features of the node.

[0030] Preferably, the extracted multi-scale time-frequency features and the abstract features of each node are used as input, and training and learning are performed based on a long short-term memory network to obtain a trained fault location model capable of output distribution line fault location, including:

[0031] Align and concatenate the multi-scale time-frequency features and node abstract features according to time steps to form fusion features;

[0032] Input the fused features into the constructed LSTM network, add an attention layer to the LSTM output, weight the features of the key time steps, input the output of the last time step of the LSTM into the fully connected layer, and output the hidden state of each node;

[0033] The node fault probability distribution is generated through the fully connected layer and the Softmax function, and the model parameters are optimized with the cross entropy loss function. After continuous training, a fault location model capable of locating output distribution line faults is obtained.

[0034] Preferably, inputting the real-time extracted feature vector into a trained fault location model to output the distribution line fault location comprises:

[0035] The fault probability distribution output by the fault location model is judged by threshold. If the maximum probability exceeds the preset fault threshold, it is determined to be the fault location. If it does not exceed the preset fault threshold, a secondary check is performed in combination with the probability distribution of adjacent nodes.

[0036] The present invention also provides a distribution line fault location system based on multi-scale time-frequency feature extraction, which applies the above-mentioned distribution line fault location method based on multi-scale time-frequency feature extraction, including:

[0037] Signal acquisition and processing unit, used to acquire electrical signals from key nodes of the distribution line and perform denoising and standardization preprocessing;

[0038] A time-frequency feature extraction unit, used to extract multi-scale time-frequency features of the preprocessed electrical signal;

[0039] The node feature extraction unit is used to encode the topological structure and node features of the distribution network through a graph neural network to obtain the abstract features of each node;

[0040] The model training unit is used to take the extracted multi-scale time-frequency features and the abstract features of each node as input, and perform training and learning based on the long short-term memory network to obtain a trained fault location model capable of output distribution line fault location;

[0041] The fault location generation unit is used to input the multi-scale time-frequency features extracted in real time and the abstract features of each node into the trained fault location model and output the fault location of the distribution line.

[0042] The present invention also provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the aforementioned distribution line fault location method based on multi-scale time-frequency feature extraction.

[0043] The present invention also provides a processor, which is used to run a program, wherein when the program is running, the aforementioned distribution line fault location method based on multi-scale time-frequency feature extraction is executed.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The distribution line fault location method based on multi-scale time-frequency feature extraction of the present invention can comprehensively capture the key information of electrical signals at different time and frequency scales by extracting multi-scale time-frequency features, and uses graph neural networks to encode the distribution network topology structure and node features. It takes into account the complex topological connection relationship of the distribution network, and deeply integrates the electrical characteristics of the nodes with the topological structure. The obtained abstract features of each node can more accurately reflect the state and mutual influence of the node under fault conditions. The multi-scale time-frequency features are combined with the node abstract features, and are trained and learned using long-short-term memory networks to capture the dynamic changes and historical information during the fault occurrence process, so that the fault location model can not only utilize spatial features, but also grasp the dynamic evolution of the fault, thereby improving the positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0047] Figure 1This is a flow chart of the distribution line fault location method based on multi-scale time-frequency feature extraction of the present invention.

[0048] Figure 2 This is a schematic diagram of the distribution line fault location system based on multi-scale time-frequency feature extraction of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] It should be understood that when used in this specification, the terms "include" and "comprising" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0051] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0052] It should be further understood that the term “and / or” used in the description of the present invention refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0053] Please refer to the following examples Figure 1 and Figure 2 .

[0054] The present application provides a method for locating distribution line faults based on multi-scale time-frequency feature extraction, including:

[0055] Step S1: Acquire electrical signals of key nodes of the distribution line and perform denoising and standardization preprocessing;

[0056] By deploying high-precision sensors at key nodes along distribution lines, we monitor and capture electrical signals in real time. These signals contain information reflecting the operating status of the distribution lines, such as voltage and current. The selection of key nodes is based on an analysis of the distribution line topology and operating characteristics, ensuring comprehensive coverage and effective capture of fault information.

[0057] Raw electrical signals are susceptible to external interference and contain noise. A filtering algorithm removes high-frequency and impulse noise from these signals, preserving fault signature information. Standardizing the denoised electrical signals eliminates the impact of differences in dimension and data size on subsequent feature extraction and analysis, improving the stability and efficiency of model training.

[0058] Step S2, extracting multi-scale time-frequency features of the preprocessed electrical signal;

[0059] The preprocessed electrical signal is decomposed into multi-scale subbands using a wavelet transform. Appropriate wavelet basis functions (such as Daubechies or Symlets) and a decomposition scale (typically 3-8 layers) are selected. Low-scale subbands capture high-frequency components of the signal, reflecting transient faults; high-scale subbands reflect low-frequency components, reflecting steady-state changes. By calculating the energy and variance of each subband, the energy distribution and fluctuation characteristics of the signal at different frequency bands are determined.

[0060] The HHT transform is used to decompose the signal into multiple intrinsic mode functions (IMFs). Valid IMF components are screened to remove noise and unreasonable components. The retained IMF components are then Hilbert transformed to obtain instantaneous amplitude and frequency characteristics, reflecting the signal amplitude changes and frequency dynamics, respectively.

[0061] The features extracted by wavelet transform and HHT transform are fused to form a feature vector containing multi-scale time-frequency information of the electrical signal.

[0062] Step S3: Encode the topological structure and node features of the distribution network through a graph neural network to obtain the abstract features of each node;

[0063] By building a graph neural network model, the input includes the adjacency matrix and node feature vector matrix of the distribution network graph. The adjacency matrix describes the node connection relationship, and the node feature vector matrix contains information such as node electrical characteristics.

[0064] The distribution network topology and node features are encoded using a graph neural network encoder, such as one based on GCN. The encoder propagates and aggregates node features, enabling each node to integrate its own and its neighboring node features. Through multiple layers of transformation and nonlinear activation functions, abstract features are obtained for each node, reflecting the node's state and topological structure.

[0065] Step S4: using the extracted multi-scale time-frequency features and the abstract features of each node as input, training and learning are performed based on the long short-term memory network to obtain a fault location model that is trained to locate output distribution line faults;

[0066] The extracted multi-scale time-frequency features are integrated with the abstract features of each node as input data for a long short-term memory (LSTM) network. The corresponding distribution line fault location labels are collected as training targets. The data is divided into training, validation, and test sets.

[0067] Build an LSTM network model, including an LSTM layer and a fully connected layer. The LSTM layer can process time series data and capture dynamic changes and historical information during the fault process; the fully connected layer maps the LSTM layer output to the fault location output space.

[0068] The LSTM network is trained using the training set, minimizing the loss function through an optimization algorithm and updating the network parameters. When the model's performance on both the training and validation sets meets the preset expectations, training is complete, resulting in a fault location model capable of locating output distribution line faults.

[0069] Step S5: Input the multi-scale time-frequency features extracted in real time and the abstract features of each node into the trained fault location model to output the distribution line fault location.

[0070] In actual operation, electrical signals from key distribution line nodes are acquired in real time. Preprocessing and feature extraction operations in steps S1 and S2 are then performed to obtain real-time multi-scale time-frequency features. Simultaneously, the graph neural network model in step S3 is used to encode the real-time distribution network topology and node features, generating real-time abstract features for each node.

[0071] The multi-scale time-frequency features and node abstract features extracted in real time are input into the trained fault location model. Based on the mapping relationship between the learned features and the fault location, the model quickly outputs the fault location of the distribution line, providing repair personnel with accurate positioning information and timely repair of the fault, thus ensuring the stable operation of the distribution network.

[0072] Specifically, the step of inputting the real-time extracted feature vector into a trained fault location model and outputting the distribution line fault location includes:

[0073] The fault probability distribution output by the fault location model is judged by threshold. If the maximum probability exceeds the preset fault threshold, it is determined to be the fault location. If it does not exceed the preset fault threshold, a secondary check is performed in combination with the probability distribution of adjacent nodes.

[0074] The distribution line fault location method based on multi-scale time-frequency feature extraction of the present invention can comprehensively capture the key information of electrical signals at different time and frequency scales by extracting multi-scale time-frequency features, and uses graph neural networks to encode the distribution network topology structure and node features. It takes into account the complex topological connection relationship of the distribution network, and deeply integrates the electrical characteristics of the nodes with the topological structure. The obtained abstract features of each node can more accurately reflect the state and mutual influence of the node under fault conditions. The multi-scale time-frequency features are combined with the node abstract features, and are trained and learned using long-short-term memory networks to capture the dynamic changes and historical information during the fault occurrence process, so that the fault location model can not only utilize spatial features, but also grasp the dynamic evolution of the fault, thereby improving the positioning accuracy.

[0075] Specifically, in a preferred embodiment of the present application, extracting multi-scale time-frequency features of the preprocessed electrical signal includes:

[0076] The electrical signal is decomposed into multi-scale sub-bands using wavelet transform, and the energy and variance characteristics of each sub-band signal are extracted;

[0077] By selecting a basis function suitable for power signal analysis and comparing the performance of different wavelet bases in fault signal reconstruction error or feature discrimination, the optimal basis function is determined.

[0078] According to the Nyquist sampling theorem and the signal frequency range, the decomposition scale is set to 8 layers (8-layer decomposition can divide the signal into low-frequency approximate components and high-frequency detail components. The decomposition formula is as follows:

[0079] The preprocessed signal x(t) is decomposed layer by layer through discrete wavelet transform (DWT):

[0080]

[0081] Where A j [n] and D j [n] represents the low-frequency approximate coefficient sequence and high-frequency detail coefficient sequence obtained by decomposition at the jth layer, n is the discrete time index; h[k] and g[k] represent the low-pass filter coefficient and high-pass filter coefficient respectively; A j-1 [2n-k] is the low-frequency approximation coefficient sequence of the j-1th layer, which is convolved with the low-pass filter coefficient h[k] through the 2n-k index method to obtain the low-frequency approximation coefficient A of the jth layer. j [n], or convolution operation with the low-pass filter coefficient g[k] to obtain the high-frequency approximation coefficient D of the jth layer j [n].

[0082] Calculate the energy E of each subband signal j , expressed as:

[0083] (j=1,2,...,J;for A J (Similarly)

[0084] Where, E j represents the energy of the j-th subband; L j is the length of the subband signal.

[0085] Calculate the fluctuation intensity of the subband signal Expressed as:

[0086]

[0087] Where, represents the variance of the j-th subband signal; μ j is the subband signal D j The mean of [n].

[0088] Obtain the intrinsic mode function of the electrical signal through HHT transformation, and extract the instantaneous amplitude characteristics and instantaneous frequency characteristics;

[0089] The preprocessed electrical signal x(t) is decomposed into several eigenmode functions by EMD And the residual component r(t) satisfies:

[0090]

[0091] For each IMF component c i (t) Perform Hilbert transform to obtain the analytical signal in:

[0092] The instantaneous amplitude is expressed as:

[0093]

[0094] The instantaneous frequency is expressed as:

[0095]

[0096] in, Indicates c i (t) The result after Hilbert transform is used to construct the analytical signal; |z i (t)| is the modulus of the analytical signal.

[0097] The features extracted by wavelet transform and HHT transform are fused to form multi-scale time-frequency features.

[0098] The preferred embodiment of the present application extracts the multi-scale time-frequency features of electrical signals by combining wavelet transform and HHT transform, making full use of the advantages of the two transforms in signal analysis to comprehensively capture the key information of electrical signals at different time and frequency scales. The wavelet transform decomposes the signal into multi-scale sub-bands and calculates the energy and variance characteristics of each sub-band, which can effectively reflect the local time-frequency characteristics of the signal; the HHT transform decomposes the signal into intrinsic mode functions, further extracts the instantaneous amplitude and frequency characteristics, and better captures the dynamic changes of the signal. The features extracted by the two are combined to form multi-scale time-frequency features, which improves the accuracy of fault location and can also better adapt to different types of fault signal analysis.

[0099] Specifically, in a preferred embodiment of the present application, the topological structure and node features of the distribution network are encoded by the graph neural network to obtain the abstract features of each node, including:

[0100] The distribution network is abstracted into graph data, where nodes and edges represent busbars and feeder branches respectively;

[0101] The distribution network system is abstracted as graph data G = (V, E), where the node V in the graph represents the bus. Suppose the graph G has |V| = N nodes, then the dimension of its adjacency matrix A is: N×N.

[0102] The structural information of the graph and the attribute characteristics of the nodes are respectively input into the VGAE model built based on GCN. The node characteristics are mapped to independent Gaussian distributions through the model to obtain the abstract characteristics of each node in the graph.

[0103] When building a graph neural network model, its input consists of information on the distribution network topology and state characteristic information of each node in the distribution network. The information on the distribution network topology is represented in the form of an adjacency matrix.

[0104] Specifically, the structural information of the graph and the attribute features of the nodes are respectively input into the VGAE model built based on GCN. The node features are mapped to independent Gaussian distributions through the model, and the abstract features of each node in the graph are obtained, including:

[0105] According to the structural information of the distribution network graph data, the adjacency matrix is ​​obtained;

[0106] The nodes represent the busbars in the distribution network, the edges represent the feeder branches between the busbars, and the adjacency matrix A∈R N*N :

[0107]

[0108] Obtain the amplitude and phase angle of the three-phase voltage, the amplitude and phase angle of the three-phase current, the active power and the reactive power of each node in the distribution network. Based on graph theory, calculate the degree centrality, eigenvector centrality, betweenness centrality and closeness centrality of each node to obtain the eigenvector matrix of each node.

[0109] Degree centrality is expressed as:

[0110]

[0111] Eigenvector centrality is expressed as:

[0112] E i =v i

[0113] Where, v i Represents the i-th element in the eigenvector v corresponding to the largest eigenvalue of the adjacency matrix A.

[0114] Betweenness centrality is expressed as:

[0115]

[0116] Where, σ st represents the total number of shortest paths from node s to node t; σ st (i) represents the number of shortest paths from node s to node t that passes through node i.

[0117] Closeness centrality is expressed as:

[0118]

[0119] Where, d ij represents the shortest distance from node i to node j.

[0120] The node feature matrix is ​​expressed as:

[0121]

[0122] Where V a 、V b and V c is the amplitude of the three-phase voltage; and is the phase angle of the three-phase voltage; I a , I b and I c is the amplitude of the three-phase current; and is the phase angle of the three-phase current; P and Q are active power and reactive power respectively.

[0123] The adjacency matrix and eigenvector matrix are respectively input into the VGAE model built on GCN. The encoder part of VGAE is used to learn the mean and variance of the Gaussian distribution mapped to the node features based on a two-layer graph convolutional network.

[0124] According to the mean and variance of the node features, the sampling results obtained from the corresponding Gaussian distribution are used as the abstract features of the node.

[0125] A two-layer graph convolutional network (GCN) is used to generate the mean μ and variance σ of the Gaussian distribution respectively:

[0126] First layer GCN (shared parameters):

[0127]

[0128] in, for The degree matrix of

[0129] Second layer GCN (independent parameters):

[0130] Mean branch:

[0131]

[0132] Variance branch:

[0133]

[0134] Generating abstract features of nodes through reparameterization techniques

[0135]

[0136] The preferred embodiment of this application encodes the topological structure and node characteristics of the distribution network through a VGAE model built on GCN. Leveraging the powerful representation capabilities of graph neural networks, the electrical characteristics of the nodes are deeply integrated with the topological structure to obtain abstract features of each node. By mapping the node features to independent Gaussian distributions and using reparameterization techniques to generate abstract features of the nodes, not only the electrical characteristics of the nodes are effectively captured, but also the topological positions and relationships of the nodes in the network are reflected.

[0137] Specifically, in a preferred embodiment of the present application, the extracted multi-scale time-frequency features and the abstract features of each node are used as input, and training and learning are performed based on a long short-term memory network to obtain a trained fault location model capable of output distribution line fault location, including:

[0138] Align and concatenate the multi-scale time-frequency features and node abstract features according to time steps to form fusion features;

[0139] The extracted multi-scale time-frequency features and the abstract features of each node are aligned according to the time step. After alignment, the two features are concatenated along the feature dimension to form a fused feature. This concatenation ensures that the features at each time step contain both the time-frequency information of the electrical signal and the topology and state information of the node.

[0140] Input the fused features into the constructed LSTM network, add an attention layer to the LSTM output, weight the features of the key time steps, input the output of the last time step of the LSTM into the fully connected layer, and output the hidden state of each node;

[0141] The node fault probability distribution is generated through the fully connected layer and the Softmax function, and the model parameters are optimized with the cross entropy loss function. After continuous training, a fault location model capable of locating output distribution line faults is obtained.

[0142] The output of the last time step of the LSTM is input into the fully connected layer. The number of neurons in the fully connected layer is determined based on actual needs. In this embodiment, it is set equal to the number of nodes in the distribution network. The fully connected layer maps the features of the LSTM output to the fault probability space of each node through linear transformation and nonlinear activation function. The output of the fully connected layer is then converted into a node failure probability distribution using the Softmax function. The Softmax function maps the output value to the (0, 1) interval and ensures that the sum of the probabilities of all nodes is 1, thereby obtaining an estimated probability of failure for each node.

[0143] The fully connected layer maps the weighted features to node failure probability, expressed as:

[0144] P i =Softmax(O attn W f +b f )∈R N×1 (W f ∈R 2H×1 ,b f ∈R)

[0145] Where, P i represents the failure probability of node i, that is, the probability of node failure predicted by the model; O attn represents the output of the attention mechanism, which is used to capture important information in the input features; W f Represents the weight matrix of the fully connected layer, whose dimension is R 2H ×1 ; b f represents the bias term of the fully connected layer.

[0146] In this embodiment, the cross-entropy loss function is used as the model optimization objective. The cross-entropy loss function measures the difference between the model's predicted node failure probability distribution and the actual fault location. By minimizing the cross-entropy loss, the model can more accurately predict faulty nodes. The Adam optimization algorithm is used to update model parameters. The Adam optimization algorithm combines the advantages of momentum and adaptive learning rate, enabling rapid convergence during training and achieving good optimization results.

[0147] The fused features and corresponding fault labels form a training dataset. During training, the dataset is divided into batches, each containing a certain number of samples. The samples from each batch are input into the constructed LSTM network, and the output is calculated using forward propagation and the cross-entropy loss function. Gradients are calculated using the backpropagation algorithm, and the Adam optimization algorithm is used to update model parameters, including the weights of the LSTM layer, the parameters of the attention layer, and the weights and biases of the fully connected layer. The training process is iterated until the model achieves optimal performance on the validation set. This results in a trained fault location model capable of locating output distribution line faults.

[0148] The present invention also provides a distribution line fault location system based on multi-scale time-frequency feature extraction, which applies the above-mentioned distribution line fault location method based on multi-scale time-frequency feature extraction, including:

[0149] Signal acquisition and processing unit, used to acquire electrical signals from key nodes of the distribution line and perform denoising and standardization preprocessing;

[0150] A time-frequency feature extraction unit, used to extract multi-scale time-frequency features of the preprocessed electrical signal;

[0151] The node feature extraction unit is used to encode the topological structure and node features of the distribution network through a graph neural network to obtain the abstract features of each node;

[0152] The model training unit is used to take the extracted multi-scale time-frequency features and the abstract features of each node as input, and perform training and learning based on the long short-term memory network to obtain a trained fault location model capable of output distribution line fault location;

[0153] The fault location generation unit is used to input the multi-scale time-frequency features extracted in real time and the abstract features of each node into the trained fault location model and output the fault location of the distribution line.

[0154] The functional explanation of each unit in this embodiment is the same as that of the distribution line fault location method based on multi-scale time-frequency feature extraction, and the technical effect is the same, so it will not be repeated here.

[0155] The present invention also provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the aforementioned distribution line fault location method based on multi-scale time-frequency feature extraction.

[0156] The technical effect of this embodiment is the same as the technical effect of the distribution line fault location method system based on multi-scale time-frequency feature extraction in the embodiment, and will not be repeated here.

[0157] The present invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above.

[0158] The present invention also provides a processor, which is used to run a program, wherein when the program is running, the aforementioned distribution line fault location method based on multi-scale time-frequency feature extraction is executed.

[0159] The technical effect of this embodiment is the same as the technical effect of the distribution line fault location method system based on multi-scale time-frequency feature extraction in the embodiment, and will not be repeated here.

[0160] The processor in this embodiment may be a central processing unit (CPU), a controller, a microcontroller, or other data processing chips.

[0161] Those skilled in the art will appreciate that the units of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0162] In the embodiments provided by the present invention, it should be understood that the division of units is merely a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0163] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0164] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the specification of the present invention.

Claims

1. A distribution line fault location method based on multi-scale time-frequency feature extraction, characterized in that: include: Acquire electrical signals from key nodes of distribution lines and perform denoising and standardization preprocessing; Extracting multi-scale time-frequency features of preprocessed electrical signals; The topological structure and node characteristics of the distribution network are encoded through graph neural networks to obtain the abstract characteristics of each node; The extracted multi-scale time-frequency features and the abstract features of each node are used as input, and training is performed based on the long short-term memory network to obtain a fault location model that is trained to locate output distribution line faults. The multi-scale time-frequency features extracted in real time and the abstract features of each node are input into the trained fault location model to output the fault location of the distribution line.

2. The distribution line fault location method based on multi-scale time-frequency feature extraction according to claim 1 is characterized in that: The method of extracting the multi-scale time-frequency features of the pre-processed electrical signal includes: The electrical signal is decomposed into multi-scale sub-bands using wavelet transform, and the energy and variance characteristics of each sub-band signal are extracted; Obtain the intrinsic mode function of the electrical signal through HHT transformation, and extract the instantaneous amplitude characteristics and instantaneous frequency characteristics; The features extracted by wavelet transform and HHT transform are fused to form multi-scale time-frequency features.

3. The distribution line fault location method based on multi-scale time-frequency feature extraction according to claim 2 is characterized in that: Decomposing the electrical signal into multi-scale sub-bands by using wavelet transform and extracting the energy characteristics and variance characteristics of each sub-band includes: The pre-processed electrical signal is decomposed layer by layer through discrete wavelet transform to obtain the low-frequency approximate coefficient sequence and high-frequency detail coefficient sequence of each layer decomposition result; The energy characteristics and variance characteristics of each subband signal are calculated based on the low-frequency approximation coefficient sequence and the high-frequency detail coefficient sequence.

4. The distribution line fault location method based on multi-scale time-frequency feature extraction according to claim 2 is characterized in that: The method of obtaining the intrinsic mode function of the electrical signal through HHT transformation and extracting the instantaneous amplitude feature and the instantaneous frequency feature includes: Decomposing the preprocessed electrical signal into intrinsic mode functions by empirical mode decomposition; The decomposed intrinsic mode function is subjected to Hilbert transform to obtain the instantaneous amplitude characteristics and instantaneous frequency characteristics.

5. The distribution line fault location method based on multi-scale time-frequency feature extraction according to claim 1, characterized in that: The topological structure and node features of the distribution network are encoded by the graph neural network, and the abstract features of each node are obtained, including: The distribution network is abstracted into graph data, where nodes and edges represent busbars and feeder branches respectively; The structural information of the graph and the attribute characteristics of the nodes are respectively input into the VGAE model built based on GCN. The node characteristics are mapped to independent Gaussian distributions through the model to obtain the abstract characteristics of each node in the graph.

6. The method for locating distribution line faults based on multi-scale time-frequency feature extraction according to claim 5, characterized in that: The structural information of the graph and the attribute features of the nodes are respectively input into the VGAE model built based on GCN. The node features are mapped to independent Gaussian distributions through the model, and the abstract features of each node in the graph are obtained, including: According to the structural information of the distribution network graph data, the adjacency matrix is ​​obtained; Obtain the amplitude and phase angle of the three-phase voltage, the amplitude and phase angle of the three-phase current, the active power and the reactive power of each node in the distribution network. Based on graph theory, calculate the degree centrality, eigenvector centrality, betweenness centrality and closeness centrality of each node to obtain the eigenvector matrix of each node. The adjacency matrix and eigenvector matrix are respectively input into the VGAE model built on GCN. The encoder part of VGAE is used to learn the mean and variance of the Gaussian distribution mapped to the node features based on a two-layer graph convolutional network. According to the mean and variance of the node features, the sampling results obtained from the corresponding Gaussian distribution are used as the abstract features of the node.

7. The method for locating distribution line faults based on multi-scale time-frequency feature extraction according to claim 1, characterized in that: The extracted multi-scale time-frequency features and the abstract features of each node are used as input, and training and learning are performed based on the long short-term memory network to obtain a trained fault location model capable of output distribution line fault location, including: Align and concatenate the multi-scale time-frequency features and node abstract features according to time steps to form fusion features; Input the fused features into the constructed LSTM network, add an attention layer to the LSTM output, weight the features of the key time steps, input the output of the last time step of the LSTM into the fully connected layer, and output the hidden state of each node; The node fault probability distribution is generated through the fully connected layer and the Softmax function, and the model parameters are optimized with the cross entropy loss function. After continuous training, a fault location model capable of locating output distribution line faults is obtained.

8. A distribution line fault location system based on multi-scale time-frequency feature extraction, characterized in that: The method for locating a distribution line fault based on multi-scale time-frequency feature extraction according to any one of claims 1 to 7 is applied, comprising: Signal acquisition and processing unit, used to acquire electrical signals from key nodes of the distribution line and perform denoising and standardization preprocessing; A time-frequency feature extraction unit, used to extract multi-scale time-frequency features of the preprocessed electrical signal; The node feature extraction unit is used to encode the topological structure and node features of the distribution network through a graph neural network to obtain the abstract features of each node; The model training unit is used to take the extracted multi-scale time-frequency features and the abstract features of each node as input, and perform training and learning based on the long short-term memory network to obtain a trained fault location model capable of output distribution line fault location; The fault location generation unit is used to input the multi-scale time-frequency features extracted in real time and the abstract features of each node into the trained fault location model and output the fault location of the distribution line.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the distribution line fault location method based on multi-scale time-frequency feature extraction according to any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the distribution line fault location method based on multi-scale time-frequency feature extraction according to any one of claims 1 to 7.

Images