Intelligent fault diagnosis method and system for power distribution network

Abstract

The invention discloses an intelligent fault diagnosis method and system for a power distribution network. The method comprises the following steps: acquiring a real-time potential signal of a node; inputting the real-time potential signal into a pre-trained fault type classifier to obtain a fault type of the node; and performing three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visual potential image, determining the position of a fault point according to the three-dimensional visual potential image and the spatial topological structure of the power distribution network; and displaying the position of the fault point and the fault type on the three-dimensional visual potential image. According to the method, the fault type can be accurately identified, the fault point can be positioned and can be visually displayed, and the efficiency and accuracy of fault diagnosis are improved.

Description

Technical Field

[0001] This application relates to the technical field of distribution network fault location, and particularly to an intelligent fault diagnosis method and system for a distribution network. Background Art

[0002] The stable operation of the power system is crucial to modern society. As the core of energy transmission, the fault diagnosis and location technology of the power grid is directly related to power supply reliability and economic benefits. In recent years, with the expansion of the power grid scale and the increase in complexity, quickly and accurately identifying and locating faults has become a research hotspot. Traditional fault diagnosis methods often have problems such as slow response speed and insufficient accuracy, especially in the case of multi-point faults or non-typical fault scenarios, where the misjudgment rate is relatively high. Summary of the Invention

[0003] This application provides an intelligent fault diagnosis method and system for a distribution network to accurately identify the fault type and locate the fault point.

[0004] To solve the above technical problems, this application provides an intelligent fault diagnosis method for a distribution network, including: Obtain the real-time potential signal of the node; Input the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node. Among them, the fault type classifier is obtained by feature classification training with a potential difference matrix and a pre-established spatial topology feature library, and the fault type classifier contains the corresponding relationship between the potential difference of the node and the fault type, as well as the potential distribution of the fault type. The potential difference matrix is obtained from the potential signals of the nodes; Perform three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visual potential image; Determine the location of the fault point according to the three-dimensional visual potential image and the spatial topology structure of the distribution network; Display the location of the fault point and the fault type on the three-dimensional visual potential image.

[0005] In some embodiments, inputting the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node includes: Obtain the propagation path characteristics of the real-time potential signal according to the real-time potential signal and the spatial topology structure; Match the propagation path characteristics of the real-time potential signal with the fault path characteristics in the fault type classifier to obtain the fault type.

[0006] In some embodiments, determining the location of the fault point according to the three-dimensional visual potential image and the spatial topology structure of the power grid includes: Obtain the coverage matrix of the nodes based on the three-dimensional visualized potential image and the spatial topology structure; Perform regional division on the three-dimensional visualized potential image according to the coverage matrix; If the potential difference of the divided region matches the fault type, mark the fault region through a boundary extraction algorithm to obtain a fault region image; Perform digital processing on the fault region image to obtain a digital potential matrix; Iteratively optimize the spatial correspondence between the pixel values of the digital potential matrix and the fault type through a gradient descent algorithm to determine the position of the fault point.

[0007] In some embodiments, it further includes the step of pre-establishing a spatial topology feature library, and the step of pre-establishing a spatial topology feature library includes: Collect the potential signals of the nodes and obtain a potential change sequence according to the potential signals; Obtain a target sequence according to the potential change sequence and the energy distribution characteristics of the fault signal; Obtain the time-frequency localization feature of the target sequence, and obtain the potential feature vector of the nodes according to the geographical location coordinates and electrical connection types of the nodes; Obtain the propagation path characteristics of the time-frequency localization feature in the spatial topology structure according to the potential feature vector of the nodes to obtain the spatial time-frequency distribution feature; Obtain a spatial topology feature library according to the spatial time-frequency distribution feature.

[0008] In some embodiments, the step of obtaining a spatial topology feature library according to the spatial time-frequency distribution feature includes: Use a density clustering algorithm to calculate the clustering density and separation degree of the feature points in the spatial time-frequency distribution feature to obtain the initial clustering center sequence; Use the initial clustering center sequence and adopt a kernel density estimation algorithm to calculate the spatial distribution characteristics of the clustering center to obtain the spatial distribution feature sequence; If the separation degree of the spatial distribution feature sequence is greater than a preset threshold, use a minimum spanning tree algorithm to construct the topological structure of the distribution network corresponding to the clustering center to obtain the topological connection matrix; According to the topological connection matrix, use a graph traversal algorithm to extract the fault path characteristics of the fault type to obtain the spatial topology feature library.

[0009] In some embodiments, the step of obtaining the potential difference matrix of the nodes from the potential signals includes: Perform time-frequency decomposition on the potential signals by using a wavelet transform algorithm to obtain a time-frequency distribution feature sequence; According to the time-frequency distribution feature sequence, the Euclidean distance algorithm is used to calculate the potential difference between nodes, and a potential difference matrix of the nodes is obtained.

[0010] In some embodiments, the real-time potential signal is three-dimensionally reconstructed to obtain a three-dimensional visualized potential image, including: The real-time potential signal is subjected to time-frequency decomposition to obtain a time-frequency feature sequence; Three-dimensional stereomicroscopy technology is used to extract spatial distribution data from the time-frequency feature sequence to obtain a three-dimensional visualized potential image.

[0011] In some embodiments, the present application also proposes an intelligent fault diagnosis system for a distribution network, including: A data acquisition module for acquiring the real-time potential signal of nodes; A fault type acquisition module for inputting the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the nodes, wherein the fault type classifier is obtained by feature classification training of the potential difference matrix and a pre-established spatial topology feature library, and the fault type classifier includes the correspondence between the potential difference of the nodes and the fault type, as well as the potential distribution of the fault type, and the potential difference matrix is obtained from the potential signals of the nodes; A three-dimensional visualization module for three-dimensionally reconstructing the real-time potential signal to obtain a three-dimensional visualized potential image; A fault point determination module for determining the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the distribution network; The three-dimensional visualization module is further configured to display the location of the fault point and the fault type on the three-dimensional visualized potential image.

[0012] The intelligent fault diagnosis method and system for the distribution network in the above embodiments obtain the real-time potential signal of nodes; input the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the nodes; three-dimensionally reconstruct the real-time potential signal to obtain a three-dimensional visualized potential image; determine the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the distribution network; and display the location of the fault point and the fault type on the three-dimensional visualized potential image. The above method and system can accurately identify the fault type and locate the fault point, and can display it intuitively, improving the efficiency and accuracy of fault troubleshooting. Description of the Drawings

[0013] Figure 1 is a schematic flowchart of an intelligent fault diagnosis method for a distribution network provided by the present application; Figure 2It is a schematic structural diagram of an intelligent fault diagnosis system for a distribution network provided by this application. Specific implementation manners

[0014] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0015] To solve the above problems, referring to Figure 1 , the intelligent fault diagnosis method for a distribution network proposed by this application includes the following steps: Step 101, obtain the real-time potential signal of the node.

[0016] The power grid consists of distribution equipment such as overhead lines, poles, cables, distribution transformers, switchgear, reactive compensation capacitors and ancillary facilities. Its main role in the power grid is to distribute electric energy. From the perspective of the nature of the distribution network, the distribution equipment also includes the distribution device of the substation. The so-called node in the power grid is a set of several equipotential physical points. In this embodiment, a high-precision sensor is used to collect the real-time potential signal from the node, and the synchronous timestamp technology is used to perform time-domain marking on the collected data. Exemplarily, in the scenario of collecting the potential signal of the node, the high-precision sensor usually adopts the capacitive or resistive principle and can capture weak potential changes. For example, the high-precision sensor deployed at a certain node can detect a potential fluctuation of 0.01 volts, with high sensitivity and strong anti-interference ability. The synchronous timestamp technology marks each group of potential signal data with an accurate time tag, such as in units of 1 microsecond, ensuring the consistency of the subsequent analysis time sequence. In this way, transient changes in the power grid, such as voltage dips or surges, can be effectively tracked, thereby providing a basis for fault location.

[0017] Step 102, input the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node. Among them, the fault type classifier is obtained by feature classification training using a potential difference matrix and a pre-established spatial topology feature library, and the fault type classifier includes the corresponding relationship between the potential difference of the node and the fault type, as well as the potential distribution template of the fault type. The potential difference matrix is obtained from the potential signal of the node.

[0018] In this embodiment, a one-dimensional convolutional neural network is used to perform feature classification training on the potential difference matrix of the nodes and a pre-established spatial topology feature library, and a fault type classifier is obtained. Among them, the one-dimensional convolutional neural network extracts local patterns in the matrix through a sliding convolutional kernel and learns the mapping relationship between the potential difference and the fault type. The fault types include but are not limited to single-point short circuit and multi-point grounding. The fault type classifier also includes potential distribution templates of various fault types. For example, a single-point short circuit fault corresponds to a specific frequency distribution. For example, the potential difference matrix of a certain power grid is input into the one-dimensional convolutional neural network. The convolutional neural network includes 3 convolutional layers and 2 fully connected layers, and a fault type classifier is obtained through training. During the training process, the convolutional neural network extracts the features of known fault patterns from the spatial topology feature library. For example, the potential difference pattern of a single-point short circuit is concentrated on a certain node pair. This fault type classifier can efficiently identify complex fault patterns.

[0019] Step 103: Perform three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visualized potential image.

[0020] In some embodiments, a wavelet transform technique or a fast Fourier transform algorithm is used to extract a time-frequency feature sequence from the real-time potential signal. A database of the spatial topology structure of the distribution network is queried for the time-frequency feature sequence to obtain a preliminary feature set of the potential signal. If the deviation between the preliminary feature set and the potential distribution of a certain fault type in the pre-established fault type classifier is less than a preset threshold, for example, the deviation is less than 5%, then a three-dimensional stereomicroscopy technique is used to perform spatial reconstruction on the preliminary feature set to obtain a three-dimensional visualized potential image and construct a three-dimensional model of the potential signal. This three-dimensional image more intuitively shows the spatial changes of the potential signal and provides a visual basis for fault area identification.

[0021] In this embodiment, the database of the spatial topology structure stores the geographical locations and connection relationships of the nodes. For example, a certain database contains the coordinates and adjacency information of 10 nodes. If the time-frequency feature sequence matches the node attributes in the database, a preliminary feature set including the node positions and potential features is generated. For example, the time-frequency feature sequence of a certain node shows an abnormal specific frequency component. By querying the database, it is confirmed that this node is located near the substation. In the above manner, it helps to initially locate the abnormal area.

[0022] Step 104: Determine the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the distribution network.

[0023] In some embodiments, the spatial topology structure of the distribution network includes a spatial topology feature library, which contains node coordinates and the connection relationships between nodes and their neighboring nodes. If the potential difference of the divided area matches the fault type, the boundary extraction algorithm is used to label the fault area and generate a fault area image. If the potential difference of the divided area has a deviation less than a preset threshold from the potential distribution of a certain fault type in the spatial topology feature library, the gradient descent optimization algorithm is used to determine the location of the fault point based on the analog-to-digital conversion accuracy and time-frequency localization characteristics.

[0024] Step 105, display the location of the fault point and the fault type on the three-dimensional visualized potential image.

[0025] In some embodiments, displaying the location of the fault point and the fault type on the three-dimensional visualized potential image not only intuitively shows the fault point but also facilitates the maintenance personnel to accurately locate and monitor the fault point.

[0026] The intelligent fault diagnosis method for the distribution network in the above embodiments includes obtaining the real-time potential signal of the node; inputting the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node; performing three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visualized potential image; determining the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the distribution network; and displaying the location of the fault point and the fault type on the three-dimensional visualized potential image. The above method can accurately identify the fault type and locate the fault point, and can be visually displayed, improving the efficiency and accuracy of fault troubleshooting.

[0027] Further, in some embodiments, inputting the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node includes: Step 201, obtain the propagation path characteristics of the real-time potential signal according to the real-time potential signal and the spatial topology structure.

[0028] In this embodiment, the propagation path characteristics of the real-time potential signal are generated by analyzing the propagation trajectory of the real-time potential signal in the spatial topology structure.

[0029] Step 202, match the propagation path characteristics of the real-time potential signal with the fault path characteristics in the fault type classifier to obtain the fault type.

[0030] In this embodiment, if the propagation path characteristics of the real-time potential signal match the path characteristics in the spatial topology feature library, the fault type classifier directly outputs the fault type. For example, for a certain fault signal propagating from node A to node C with a path length of 10 kilometers, it matches the single-point short-circuit path in the fault type classifier. The fault type classifier outputs the fault type according to the matching result, such as "single-point short circuit". This method uses path characteristics to quickly locate the fault type and improves the detection efficiency.

[0031] In some embodiments, determining the fault point location according to the three-dimensional visualized potential image and the spatial topology of the power grid includes: Step 301, obtaining the coverage matrix of the nodes according to the three-dimensional visualized potential image and the spatial topology.

[0032] For the three-dimensional visualized potential image, in combination with the pre-established database of the spatial topology of the distribution network, obtain the node distribution information, calculate the spatial coverage range between nodes through the Euclidean distance algorithm, and obtain the node coverage matrix. For example, a certain power grid has 8 nodes, and each node has three-dimensional coordinates. Calculate the geometric distance between nodes through the Euclidean distance algorithm to generate an 8×8 node coverage matrix. The matrix elements represent the spatial coverage range between nodes. For example, the coverage range of a certain node pair is 2.5 kilometers. This matrix quantifies the spatial relationship between nodes and provides a basis for regional division.

[0033] Step 302, dividing the three-dimensional visualized potential image according to the coverage matrix.

[0034] In this embodiment, according to the node coverage matrix, use an image segmentation algorithm to divide the three-dimensional visualized potential image. The image segmentation algorithm divides the image into several regions by clustering pixel points with similar potential values. For example, a certain three-dimensional visualized potential image is divided into 5 regions after segmentation, and the average potential value of a certain region is abnormally high. By comparing with the potential value range of the preset fault type, for example, the potential value range corresponding to the multi-point grounding fault is 0.8 to 1.2, it is confirmed that this region matches the fault type. Through the above method, the fault-related region can be accurately separated.

[0035] Step 303, if the potential difference of the divided region matches the fault type, mark the fault region through a boundary extraction algorithm to obtain a fault region image.

[0036] In this embodiment, if the potential value of the divided region matches the preset fault type, mark the fault region through a boundary extraction algorithm to obtain a fault region image. Otherwise, re-execute step 301. For example, the potential gradient of a certain fault region is higher than 0.5, and a marked image containing this region is generated. The fault region in the image is highlighted in red, which can visually present the fault region and facilitate subsequent analysis.

[0037] Step 304: Digitally process the fault area image to obtain a digital potential matrix.

[0038] In this embodiment, an analog-to-digital conversion algorithm is used to digitally process the pixel values of the three-dimensional potential distribution image to obtain a digital potential matrix.

[0039] Step 305: Iteratively optimize the spatial correspondence between the pixel values of the digital potential matrix and the fault type through a gradient descent algorithm to determine the location of the fault point.

[0040] In this embodiment, if the potential distribution deviation between the three-dimensional visualization image and a certain fault type in the spatial topology feature library is less than a preset threshold, then through a gradient descent optimization algorithm, based on the analog-to-digital conversion accuracy and time-frequency localization characteristics, the location of the fault point is determined.

[0041] In some embodiments, the intelligent fault diagnosis method for a distribution network proposed in this application further includes the step of pre-establishing a spatial topology feature library. This step includes: Step 401: Obtain the potential signal acquisition data of the node, and obtain a potential change sequence according to the potential signal acquisition data.

[0042] In some implementation manners, a high-precision sensor is used to collect potential signals from the node, and a synchronous timestamp technology is used to perform a time-domain marking on the acquisition data to obtain an original potential signal sequence with timestamps. If the amplitude of the original potential signal sequence is lower than a preset threshold, then a signal amplifier is used to perform a gain process on the original potential signal to adjust the original potential signal sequence to an adapted range to obtain an amplified potential signal sequence. According to the amplified potential signal sequence, a high-precision analog-to-digital converter is used for digital processing, and the signal is discretized by combining a high sampling frequency technology to generate a digital potential signal sequence. Through Fourier transform, a frequency-domain analysis is performed on the digital potential signal sequence. If it is detected that the noise frequency is higher than a preset threshold, then a low-pass filter is applied for noise reduction processing to obtain an initial potential change sequence.

[0043] In a possible implementation manner, the amplitude of the original potential signal sequence may be low due to power grid load fluctuations. For example, the signal amplitude of a certain node is only 0.05 volts, which is lower than the preset threshold of 0.1 volts. At this time, the signal amplifier amplifies the signal to 0.2 volts through an adjustable gain module to adapt to the input range of the analog-to-digital converter. It should be noted that additional noise should be avoided during the amplification process, and a low-noise amplifier is preferably used to ensure signal fidelity. The step of signal amplification significantly improves the accuracy of subsequent digital processing.

[0044] In a possible implementation, a high-precision analog-to-digital converter combines with high sampling frequency technology to discretize the amplified potential signal sequence. For example, an analog-to-digital converter with 24-bit precision and a sampling frequency set at 10 kHz can convert a continuous signal into a high-resolution digital sequence. This high sampling frequency can capture rapidly changing potential signals, such as transient fluctuations caused by switch operations in the power grid, thereby providing data support for accurate fault location.

[0045] In a possible implementation, Fourier transform is used to perform frequency-domain analysis on the digitized potential signal sequence. For example, it is found through analysis that there is high-frequency noise of 500 Hz in the potential signal, which exceeds the preset threshold of 300 Hz. A low-pass filter is applied with a cut-off frequency set at 400 Hz to effectively filter out the high-frequency noise and retain the useful signal components. The initial potential change sequence after filtering more clearly reflects the actual operating state of the power grid, such as potential fluctuations caused by load changes or short circuits.

[0046] For example, a certain substation successfully detected a weak potential anomaly caused by line aging through this solution, and early warning avoided a power outage accident. The overall solution forms a complete link from data acquisition to feature extraction through high-precision acquisition, synchronous marking, signal amplification, digital processing, and noise reduction analysis. Each link supports each other to ensure the high fidelity of the signal and the reliability of the analysis.

[0047] Further, in step 401, if the noise component in the initial potential change sequence exceeds the preset threshold, wavelet transform technology is used. Based on wavelet basis functions and multi-scale decomposition, noise spectrum separation is performed through adaptive thresholds and threshold shrinkage rules to obtain the denoised potential change sequence.

[0048] Obtain the high-frequency components from the initial potential change sequence. Using wavelet transform technology, perform multi-scale decomposition on the sequence based on a preselected wavelet basis function to obtain a wavelet coefficient sequence. For the wavelet coefficient sequence, calculate the amplitudes of the coefficients at each scale. If the amplitude exceeds the preset threshold, adjust it through an adaptive threshold algorithm to obtain an adjusted coefficient sequence. According to the adjusted coefficient sequence, apply the threshold shrinkage rule to suppress the high-frequency noise components, and use wavelet reconstruction technology to restore the signal to obtain a preliminary denoised sequence. Perform frequency-domain analysis on the preliminary denoised sequence through Fourier transform. If residual noise spectra are detected, apply low-pass filter processing to obtain the final denoised potential change sequence.

[0049] Exemplarily, in node potential signal processing, wavelet transform technology is used to extract high-frequency components. By decomposing the signal into frequency components of different scales, wavelet transform can effectively separate high-frequency noise from useful signals. The wavelet basis functions used in wavelet transform, such as Daubechies wavelets, are suitable for capturing transient changes in power grid signals due to their orthogonality and compact support characteristics. In a possible implementation, if the potential change sequence of a certain node collected contains high-frequency components of voltage mutation, the db4 wavelet basis function is selected for 4-layer decomposition to obtain a coefficient sequence from low frequency to high frequency. This decomposition method can clearly distinguish the steady part and transient disturbance in the signal data.

[0050] Furthermore, Fourier transform analysis is used to analyze the frequency domain characteristics of the preliminary denoised sequence. For example, it is found that there is still residual noise of 400 Hz in the sequence, exceeding the expected normal range of 300 Hz. Preferably, a low-pass filter is applied with a cut-off frequency set to 350 Hz to filter out the high-frequency residual noise, obtaining the final denoised potential change sequence. This multi-step processing method of decomposition, adjustment, suppression, and filtering forms a complete signal optimization link, with each link supporting each other to ensure the high fidelity of the potential signal data.

[0051] Step 402, according to the potential change sequence and the energy distribution characteristics of the fault signal, obtain the target sequence.

[0052] In some embodiments, fast Fourier transform is used to perform frequency domain decomposition on the (denoised) potential change sequence to obtain the frequency domain component sequence of the potential change sequence. For this frequency domain component sequence, a band-pass filter is used to screen out the frequency band components related to the fault. If the sequence part of the components in the frequency domain component sequence whose amplitude exceeds a preset threshold is used as the fault-related frequency band sequence. Hilbert transform is used to extract the energy distribution characteristics of each frequency band in the fault-related frequency band sequence, calculate the energy values of each frequency band, and obtain the energy distribution sequence. The principal component analysis algorithm is used to perform dimensionality reduction processing on the energy values of the energy distribution sequence to obtain the target sequence.

[0053] Exemplarily, the fast Fourier transform technique is used to transform the (denoised) sequence of potential changes from the time domain to the frequency domain, and decompose it into a sequence of frequency domain components. Preferably, the sequence of potential changes collected at a certain node contains periodic voltage fluctuations. After applying the fast Fourier transform, a sequence of frequency domain components is obtained, showing a 50 Hz fundamental frequency and several high-frequency harmonics. This decomposition method facilitates subsequent analysis of fault-related frequency characteristics. For the sequence of frequency domain components, a band-pass filter is used to set the upper and lower cut-off frequencies, allowing only a specific frequency band to pass through, screening out the frequency band components related to the fault, and removing the irrelevant components. For example, power grid faults are often accompanied by high-frequency components in the range of 200 to 500 Hz. The band-pass filter is set with a frequency range of 200 to 500 Hz to screen out the relevant components. If the amplitude of the component exceeds a preset threshold, such as 0.05, then the component is retained to obtain a sequence of frequency bands related to the fault. Another example is that a 300 Hz component is detected in the signal of a certain node, with an amplitude of 0.07, exceeding the threshold, indicating that there may be fault characteristics, and the component is retained for subsequent analysis.

[0054] Exemplarily, the Hilbert transform is used to extract the energy distribution characteristics of the sequence of frequency bands related to the fault, calculate the energy values of the sequence of frequency bands related to the fault, and generate an energy distribution sequence. For example, the energy value of the signal of a certain node increases from 0.02 to 0.08 at the fault moment, clearly reflecting the dynamic characteristics of the fault. The principal component analysis algorithm is used to perform dimensionality reduction on the energy values of the energy distribution sequence to obtain the target sequence. Principal component analysis reduces the data dimension by extracting the main change directions of the energy values and retains the key information. For example, an energy distribution sequence contains energy values of multiple frequency bands. Principal component analysis identifies the first two principal components, which explain 90% of the variance, and generates a simplified target sequence.

[0055] The target sequence clearly reflects the trend of potential changes caused by the fault, facilitating subsequent fault location and diagnosis. This multi-step processing efficiently extracts fault characteristics through frequency domain decomposition, screening, energy extraction, and dimensionality reduction.

[0056] Step 403, obtain the time-frequency localization characteristics of the target sequence, and according to the geographical location coordinates and electrical connection type of the node, obtain the potential feature vector of the node.

[0057] In this embodiment, a time-frequency analysis method is used to extract the time-frequency localization characteristics of the target sequence, and combined with the geographical location coordinates and electrical connection type in the node distribution coverage information, a node potential feature vector is generated.

[0058] Exemplarily, the wavelet transform technology is adopted to perform time-frequency decomposition on the target sequence, obtain the distribution characteristics of the target sequence in terms of time and frequency, and obtain the time-frequency localization feature sequence. According to the geographical location coordinates in the node distribution coverage information, a spatial interpolation algorithm is used to calculate the spatial weights of each node, and a weighted time-frequency feature sequence of the time-frequency localization feature sequence is obtained. If the eigenvalue of any node in the weighted time-frequency feature sequence exceeds the preset threshold, then according to the electrical connection type, the graph convolutional neural network algorithm is used to fuse the electrical topological structure information of the nodes to obtain the preliminary node potential feature vector. The principal component analysis algorithm is used to perform dimensionality reduction processing on the preliminary node potential feature vector, extract the spatial correlation characteristics related to the fault signal, and obtain the potential feature vector of the target node. In this embodiment, the electrical connection type and the electrical topological structure information adopt the general types and structures such as power grids or substations, such as external electrical connections and internal electrical connections, etc., and the electrical topological structure information is such as modern power grid control systems, etc., which are not specifically limited here.

[0059] In a possible implementation manner, the potential change sequence collected by a certain node contains a transient fault signal. After applying the discrete wavelet transform, a time-frequency localization feature sequence is obtained, showing a mutation of the 300 Hz high-frequency component within 0.1 second when the fault occurs. This decomposition method is convenient for capturing the instantaneous characteristics of the fault.

[0060] Combined with the geographical location coordinates in the node distribution coverage information, the spatial interpolation algorithm is used to calculate the spatial weights of each node. For example, the geographical coordinates of 10 nodes in a certain power grid are known. The inverse distance weighted interpolation algorithm is used to calculate that the weights of the nodes closer to the fault node are 0.8 and those farther away are 0.3, generating a weighted time-frequency feature sequence. This method effectively fuses spatial information and enhances the regional representativeness of the features. When the eigenvalue of a certain node in the weighted time-frequency feature sequence exceeds the preset threshold, such as 0.06, it indicates that there may be a fault signal and further analysis is required. The graph convolutional neural network aggregates the features of neighbor nodes through the electrical connection relationship between nodes to generate the preliminary node potential feature vector. For example, a certain node is connected to 3 neighbor nodes through high-voltage lines. The graph convolutional neural network analyzes its 300 Hz eigenvalue and combines the spatial topological structure to generate a feature vector containing spatial and electrical information. This way makes full use of the topological characteristics of the power grid and enhances the expression ability of the features. Through the preliminary node potential feature vector, the principal component analysis algorithm performs dimensionality reduction processing to extract the spatial correlation characteristics related to the fault signal.

[0061] Step 404, according to the potential feature vector of the node, obtain the propagation path characteristics of the time-frequency localization feature in the spatial topological structure, and obtain the spatio-temporal frequency distribution feature.

[0062] Based on the geographical location information in the potential feature vector of the nodes, the Kriging interpolation algorithm is used to calculate the geographical space weights of each node, and the weighted geographical location sequence is obtained. Through the weighted geographical location sequence, the minimum spanning tree algorithm is used to construct the spatial topology structure of the distribution network, and the topology connection matrix is obtained. From the topology connection matrix, the graph traversal algorithm is used to extract the propagation path of the time-frequency features, and the path characteristic sequence is obtained. For the path characteristic sequence, the kernel density estimation algorithm is used to calculate the distribution law of the features in space, and the spatial time-frequency distribution characteristics are obtained.

[0063] Exemplarily, in power grid fault analysis, based on the geographical location information in the potential feature vector of the nodes, the Kriging interpolation algorithm is used to calculate the geographical space weights. Kriging interpolation is an interpolation method based on spatial autocorrelation. Considering the distance between nodes and the statistical characteristics of data distribution, it generates a smooth weight distribution. For example, a power grid contains 8 nodes with known geographical coordinates. According to the distance between nodes and the variance of potential changes, the Kriging interpolation algorithm calculates that the weight of the nodes close to the fault area is 0.75, and that of the nodes far away from the area is 0.25, generating a weighted geographical location sequence. This method makes full use of spatial statistical characteristics to ensure that the weights reflect the geographical correlation of fault signals.

[0064] In a possible implementation, through the weighted geographical location sequence, the minimum spanning tree algorithm constructs the spatial topology structure of the distribution network. The minimum spanning tree algorithm uses the geographical distance between nodes as the edge weight, connects all nodes to form an acyclic graph, and generates a topology connection matrix. For example, in a power grid, 10 nodes use the minimum spanning tree algorithm to generate a topology connection matrix based on the weighted distance. The matrix elements represent the connection strength between nodes. For example, the connection weight of a pair of nodes is 0.9. This topological graph clearly reflects the spatial structure of the distribution network and is convenient for subsequent analysis.

[0065] The graph traversal algorithm is used to extract the propagation path of the time-frequency features from the topology connection matrix, and the path characteristic sequence is obtained. The graph traversal algorithm traces the propagation trajectory of the fault signal in the topological graph by means of depth-first or breadth-first traversal. For example, when a fault occurs at a certain node, its 300 Hz time-frequency feature generates a characteristic sequence containing 3 propagation paths through breadth-first traversal, and the path lengths are 5, 7, and 10 kilometers respectively. This path characteristic sequence intuitively shows the spatial propagation law of the signal.

[0066] In some embodiments, for the path feature sequence, the kernel density estimation algorithm is used to obtain the distribution law of features in space, and the spatio-temporal distribution features in space are generated. Exemplarily, the path feature sequence of a certain power grid shows that the 300-Hz feature is concentrated in a certain area, and the kernel density estimation generates a density map, showing that the peak of the feature density in this area is 0.85. This distribution feature clearly reveals the spatial aggregation characteristics of fault signals and provides a basis for fault location. For example, when a certain node fails, its weighted geographical location sequence shows a weight of 0.7, the topological connection matrix indicates its strong connection with neighbor nodes, the path feature sequence reveals that the signal propagates along the main path, and the kernel density estimation further confirms that the feature is concentrated in a certain area. The above method makes full use of geographical and topological information and enhances the comprehensiveness of feature extraction.

[0067] Step 405, obtain a spatial topological feature library according to the spatio-temporal distribution features in space.

[0068] In this embodiment, the density clustering algorithm is used to calculate the clustering density and separation degree of feature points in the spatio-temporal distribution features in space to obtain the initial clustering center sequence. Through the initial clustering center sequence, the kernel density estimation algorithm is used to calculate the spatial distribution characteristics of the clustering centers to obtain the spatial distribution feature sequence. If the separation degree of the spatial distribution feature sequence is greater than a preset threshold, the minimum spanning tree algorithm is used to construct the power grid topological structure corresponding to the clustering centers to obtain the topological connection matrix. According to the topological connection matrix, the graph traversal algorithm is used to extract the fault path characteristics of the fault type to obtain the spatial topological feature library.

[0069] Exemplarily, in power grid fault analysis, the density clustering algorithm is used to calculate the clustering density and separation degree of spatio-temporal distribution feature points to generate an initial clustering center sequence. The density clustering algorithm identifies high-density areas as clustering centers by analyzing the density of feature points in space. For example, the spatio-temporal distribution features of a certain power grid contain 1000 feature points, and the density clustering algorithm identifies 5 high-density areas and generates 5 initial clustering centers, and the separation degree between the center points is 0.8. This method effectively captures the spatial aggregation characteristics of feature points and provides a reliable starting point for subsequent analysis.

[0070] In a possible implementation, based on the initial cluster center sequence, the kernel density estimation algorithm further calculates the spatial distribution characteristics of the cluster centers to generate a spatial distribution feature sequence. For example, the spatial distribution feature sequence is generated by kernel density estimation for 5 cluster centers of a certain power grid, showing that the density peak of the center points in a certain area is 0.9. This sequence clearly reflects the spatial distribution law of the cluster centers in the power grid, facilitating subsequent topological analysis. If the separation degree of the spatial distribution feature sequence is greater than a preset threshold, such as 0.7, the minimum spanning tree algorithm is used to construct the power grid topological structure corresponding to the cluster centers to generate a topological connection matrix. The minimum spanning tree algorithm uses the geographical distance between the cluster centers as the edge weight to connect all the centers to form an acyclic graph. For example, the topological connection matrix is generated based on the weighted distance for 5 cluster centers of a certain power grid by the minimum spanning tree algorithm, and the matrix elements represent the connection strength between the centers. For example, the connection weight between a pair of centers is 0.85. This topological structure intuitively shows the spatial relationship between the cluster centers.

[0071] For the topological connection matrix, the graph traversal algorithm extracts the fault path characteristics of the fault type to generate a spatial topology feature library. The graph traversal algorithm tracks the corresponding propagation path characteristics of the fault type in the spatial topology structure of the distribution network in a breadth-first manner.

[0072] Furthermore, in some embodiments, the potential difference matrix is obtained from the potential signals of the nodes, including: Step 501, obtain the time-frequency distribution characteristics from the potential signals, and perform time-frequency decomposition on the real-time potential signals by using the wavelet transform algorithm to obtain a time-frequency distribution feature sequence; Step 502, according to the time-frequency distribution feature sequence, calculate the potential difference between nodes by using the Euclidean distance algorithm to obtain a node potential difference matrix.

[0073] For example, a certain power grid contains 10 nodes, and each node's time-frequency feature sequence contains 100 feature points. The Euclidean distance algorithm calculates the distance between each pair of nodes to generate a 10×10 potential difference matrix. The matrix elements represent the potential difference between nodes. For example, the difference between a pair of nodes is 0.75. This matrix intuitively reflects the potential relationship between nodes and provides a key basis for fault location.

[0074] Refer to Figure 2 , some other embodiments of the present application also provide an intelligent fault diagnosis system for a distribution network, including: A data acquisition module 601, configured to acquire the real-time potential signals of the nodes; A fault type acquisition module 602 is configured to input the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node. The fault type classifier is obtained by performing feature classification training on the potential difference matrix and a pre-established spatial topology feature library, and the fault type classifier includes the correspondence between the potential difference of the node and the fault type, as well as the potential distribution of the fault type. The potential difference matrix is obtained from the potential signals of the nodes. A three-dimensional visualization module 603 is configured to perform three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visualized potential image. A fault point determination module 604 is configured to determine the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the distribution network. The three-dimensional visualization module 603 is further configured to display the location of the fault point and the fault type on the three-dimensional visualized potential image.

[0075] It should be noted that the intelligent fault diagnosis system for a distribution network provided in the embodiments of the present application is used to execute all the process steps of the intelligent fault diagnosis method for a distribution network in the above embodiments. The working principles and beneficial effects of the two correspond one by one, and thus will not be elaborated herein.

[0076] The embodiments of the present application further provide an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition program. When the processor executes the computer program, the steps in the embodiments of the above intelligent fault diagnosis method for a distribution network are implemented, such as Figure 1 the steps shown. Alternatively, when the processor executes the computer program, the functions of each module / unit in the above device embodiments are implemented, such as the data acquisition module.

[0077] Exemplarily, the computer program can be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0078] The electronic device may be a computing device such as a desktop computer, notebook, palm computer, and smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0079] The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device, connecting various parts of the entire electronic device through various interfaces and lines.

[0080] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices.

[0081] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of this application, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0082] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided in this application, the connection relationship between the modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0083] The specific embodiments described above have further elaborated on the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only the specific embodiments of this application and is not used to limit the protection scope of this application. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application should be included in the protection scope of this application.

Claims

1. An intelligent fault diagnosis method for a distribution network, characterized in that Including: Obtain the real-time potential signal of the node; Input the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node. Among them, the fault type classifier is obtained by feature classification training with a potential difference matrix and a pre-established spatial topology feature library, and the fault type classifier includes the corresponding relationship between the potential difference of the node and the fault type, as well as the potential distribution of the fault type. The potential difference matrix is obtained from the potential signals of the node; Perform three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visualized potential image; Determine the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the distribution network; Display the location of the fault point and the fault type on the three-dimensional visualized potential image.

2. The method according to claim 1, wherein Input the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node, including: Obtain the propagation path characteristics of the real-time potential signal according to the real-time potential signal and the spatial topology structure; Match the propagation path characteristics of the real-time potential signal with the fault path characteristics in the fault type classifier to obtain the fault type.

3. The method according to claim 1, characterized in that Determine the location of the fault point according to the three-dimensional visualized potential image and the spatial topology structure of the power grid, including: Obtain the coverage matrix of the node according to the three-dimensional visualized potential image and the spatial topology structure; Divide the three-dimensional visualized potential image according to the coverage matrix; If the potential difference of the divided area matches the fault type, label the fault area through a boundary extraction algorithm to obtain a fault area image; Perform digital processing on the fault area image to obtain a digital potential matrix; Iteratively optimize the spatial correspondence between the pixel values of the digital potential matrix and the fault type through a gradient descent algorithm to determine the location of the fault point.

4. The method according to claim 1, wherein It also includes the step of pre-establishing a spatial topology feature library. The step of pre-establishing a spatial topology feature library includes: Collect the potential signals of the node and obtain a potential change sequence according to the potential signals; Obtain a target sequence according to the potential change sequence and the energy distribution characteristics of the fault signal; Obtain the time-frequency localization characteristics of the target sequence, and obtain the potential feature vector of the node according to the geographical location coordinates and electrical connection type of the node; Obtain the propagation path characteristics of the time-frequency localization characteristics in the spatial topology structure according to the potential feature vector of the node to obtain the spatial time-frequency distribution characteristics; Obtain a spatial topology feature library according to the spatial time-frequency distribution characteristics.

5. The method according to claim 4, wherein The step of obtaining a spatial topology feature library according to the spatial time-frequency distribution characteristics includes: Use a density clustering algorithm to calculate the clustering density and separation degree of the feature points in the spatial time-frequency distribution characteristics to obtain an initial clustering center sequence; Through the initial clustering center sequence, use a kernel density estimation algorithm to calculate the spatial distribution characteristics of the clustering center to obtain a spatial distribution feature sequence; If the separation degree of the spatial distribution feature sequence is greater than a preset threshold, the minimum spanning tree algorithm is used to construct the topological structure of the distribution network corresponding to the clustering center, and a topological connection matrix is obtained; According to the topological connection matrix, a graph traversal algorithm is used to extract the fault path characteristics of the fault type, and the spatial topological feature library is obtained.

6. The method according to claim 1, characterized in that, The obtaining process of the potential difference matrix includes: The wavelet transform algorithm is used to perform time-frequency decomposition on the potential signal to obtain a time-frequency distribution feature sequence; According to the time-frequency distribution feature sequence, the Euclidean distance algorithm is used to calculate the potential difference between nodes, and the potential difference matrix of the nodes is obtained.

7. The method according to claim 1, wherein The three-dimensional reconstruction of the real-time potential signal to obtain a three-dimensional visualized potential image includes: The real-time potential signal is subjected to time-frequency decomposition to obtain a time-frequency feature sequence; The three-dimensional stereomicroscopy technology is used to extract the spatial distribution data from the time-frequency feature sequence to obtain a three-dimensional visualized potential image.

8. An intelligent fault diagnosis system for a distribution network, characterized in that, It includes: A data acquisition module for acquiring the real-time potential signal of the node; A fault type acquisition module for inputting the real-time potential signal into a pre-trained fault type classifier to obtain the fault type of the node. The fault type classifier is obtained by feature classification training with the potential difference matrix and a pre-established spatial topological feature library, and the fault type classifier includes the corresponding relationship between the potential difference of the node and the fault type, as well as the potential distribution of the fault type. The potential difference matrix is obtained from the potential signal of the node; A three-dimensional visualization module for performing three-dimensional reconstruction on the real-time potential signal to obtain a three-dimensional visualized potential image; A fault point determination module for determining the location of the fault point according to the three-dimensional visualized potential image and the spatial topological structure of the distribution network; The three-dimensional visualization module is further configured to display the location of the fault point and the fault type on the three-dimensional visualized potential image.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the intelligent fault diagnosis method for the distribution network according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the intelligent fault diagnosis method for the distribution network according to any one of claims 1 to 7.

Images