Distribution network low-voltage equipment monitoring method, system, equipment and medium under multi-source data

Abstract

The invention discloses a distribution network low-voltage equipment monitoring method, system, equipment and medium under multi-source data, and belongs to the technical field of distribution network equipment monitoring, and the method comprises the steps: obtaining a monitoring data set of distribution network low-voltage equipment; performing preprocessing based on the monitoring data set, extracting electrical data features and image features, and establishing graph nodes according to the electrical data features and the image features; determining a time evolution edge, a spatial topology edge span and a cross-modal edge based on the graph node, and establishing a bimodal abnormal heterogeneous evolution graph in combination with the graph node, the time evolution edge, the spatial topology edge and the cross-modal edge; and performing abnormal diffusion tracking based on the bimodal abnormal heterogeneous evolution graph to complete fault source node identification. Omnibearing sensing and data fusion of the running state of the distribution network low-voltage equipment are realized, the reliability and timeliness of monitoring of the distribution network low-voltage equipment are improved, and the effect of improving the fault diagnosis efficiency and accuracy of the distribution network low-voltage equipment is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distribution network equipment monitoring, in particular to a distribution network low-voltage equipment monitoring method, system, device and medium under multi-source data. BACKGROUND

[0002] As the last link between the power system and the user, the safe and stable operation of the distribution network low-voltage equipment plays a key role in ensuring the quality of power supply and improving the user's power experience. However, the traditional distribution network low-voltage equipment monitoring only monitors the equipment operating state through electrical parameters or only uses visual information to check the equipment appearance, which is difficult to fully capture the complex characteristics and potential fault hidden dangers in the equipment operation process. On the one hand, relying only on electrical modal data such as voltage, current, power and other parameters cannot intuitively present the physical appearance changes of the equipment, such as equipment component aging, damage, foreign matter intrusion, etc. On the other hand, relying only on visual modal data cannot obtain the internal electrical performance changes of the equipment, and cannot timely detect potential fault risks caused by abnormal electrical parameters. At the same time, with the widespread application of Internet of Things and sensors, a large amount of heterogeneous data is collected. How to effectively integrate and utilize multi-source data has become the key to improving the accuracy and reliability of distribution network low-voltage equipment monitoring.

[0003] Therefore, in the related art at present, there are problems of one-sidedness of single-modal data monitoring information, difficulty in fully capturing equipment operation characteristics, poor efficiency and accuracy of distribution network low-voltage equipment monitoring, inability to timely discover potential fault hidden dangers and accurately locate fault sources. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the technical problem solved by the present application is: how to provide a distribution network low-voltage equipment monitoring method, system, device and medium under multi-source data to solve the problems of one-sidedness of single-modal data monitoring information, difficulty in fully capturing equipment operation characteristics, poor efficiency and accuracy of distribution network low-voltage equipment monitoring, inability to timely discover potential fault hidden dangers and accurately locate fault sources in the prior art, realize power supply and demand balance, and achieve the effect of improving the efficiency and accuracy of distribution network low-voltage equipment fault diagnosis.

[0006] To solve the above technical problems, the present application provides the following technical solutions: a distribution network low-voltage equipment monitoring method under multi-source data, comprising,

[0007] Acquire monitoring datasets of low-voltage distribution network equipment; preprocess the monitoring datasets to extract electrical data features and image features, and establish graph nodes based on the electrical data features and image features; determine time evolution edges, spatial topology edges, and cross-modal edges based on the graph nodes, and establish a bimodal anomaly heterogeneous evolution graph by combining the graph nodes, the time evolution edges, the spatial topology edges, and the cross-modal edges; perform anomaly propagation tracking based on the bimodal anomaly heterogeneous evolution graph to complete the identification of fault source nodes.

[0008] As a preferred embodiment of the multi-source data-based monitoring method for low-voltage distribution equipment described in this invention, the method involves: tracking anomaly propagation based on the dual-modal anomaly heterogeneous evolution graph to identify fault source nodes, including: identifying anomalies in the graph nodes and establishing anomaly node identifiers; performing anomaly propagation analysis based on the dual-modal anomaly heterogeneous evolution graph to establish propagation chains for anomaly nodes; and using cross-modal edges to perform cross-modal anomaly coupling of graph nodes, combining the cross-modal anomaly coupling results with the propagation chains to perform joint anomaly propagation tracking, thereby completing fault source node identification. The beneficial effect of this preferred embodiment is that by utilizing cross-modal edges for cross-modal anomaly coupling of graph nodes and combining the cross-modal anomaly coupling results with the propagation chains for joint anomaly propagation tracking, it can fully integrate anomaly information from electrical and visual modes, improving the accuracy and reliability of fault source node identification.

[0009] As a preferred embodiment of the multi-source data-based monitoring method for low-voltage distribution equipment described in this invention, the method involves: using the cross-modal edge to perform cross-modal anomaly coupling of graph nodes, including: obtaining the required monitoring accuracy of the low-voltage distribution equipment; configuring a calibration traceability window based on the required monitoring accuracy; selecting any modal data as the traceability starting point; obtaining the anomaly degree of the traceability starting point under the current modality; adjusting the traceability window based on the anomaly degree; and performing association capture of another modality data; and completing cross-modal anomaly coupling based on the association capture result. The beneficial effect of this preferred embodiment is that by adjusting the traceability window based on the anomaly degree and performing association capture of another modality data, the time range of cross-modal association analysis can be optimized, improving the accuracy and efficiency of anomaly association identification between different modal data.

[0010] As a preferred embodiment of the multi-source data-based monitoring method for low-voltage distribution equipment described in this invention, the method includes: tracking anomaly propagation based on the dual-modal anomaly heterogeneous evolution graph to identify fault source nodes; converting the electrical data features and image features of the graph nodes into combined feature vectors; using any graph node as a starting node, performing node association analysis using time evolution edges, spatial topology edges, and cross-modal edges to establish a set of neighboring nodes; reconstructing the graph structure based on the starting node and the set of neighboring nodes; performing feature aggregation of the combined feature vectors from the farthest neighbor node in the reconstructed graph structure and passing it to the next neighbor node; after all the feature aggregation results of the neighbor nodes have been passed to the starting node, using the updated starting node to identify anomaly nodes, and completing anomaly propagation tracking based on the anomaly node identification results and the feature correlation of the reconstructed graph structure. The beneficial effect of this preferred technical solution is that by performing feature aggregation of the combined feature vectors from the farthest neighbor node and passing it upwards layer by layer, the deep relationships between nodes in the graph structure can be fully explored, enhancing the comprehensiveness and accuracy of anomaly propagation tracking.

[0011] As a preferred embodiment of the multi-source data-based low-voltage equipment monitoring method for distribution networks described in this invention, the step of performing feature aggregation of combined feature vectors from the farthest neighbor node in the reconstructed graph structure and passing it to the next neighbor node includes: aggregating the combined feature vectors within the farthest neighbor node to generate a first feature aggregation result; sending the first feature aggregation result to the next neighbor node, where the next neighbor node receives the first feature aggregation result and performs aggregation; and repeating the layer-by-layer upward aggregation until the feature aggregation results of all neighbor nodes are passed to the starting node.

[0012] As a preferred embodiment of the multi-source data-based low-voltage equipment monitoring method for distribution networks described in this invention, the method includes: after completing the identification of fault source nodes, the method further includes: fitting the development of potential hazards based on the fault source nodes and the propagation path, and establishing a fitted anomaly level; determining the true anomaly level based on the fault source nodes, and reconstructing and reporting the anomaly based on the true anomaly level and the fitted anomaly level.

[0013] As a preferred embodiment of the multi-source data-based low-voltage equipment monitoring method for distribution networks described in this invention, the method further includes: determining the actual anomaly level based on the fault source node, reconstructing and reporting the anomaly according to the actual anomaly level and the fitted anomaly level; creating an anomaly record table based on the fault source node; performing correlation analysis on the anomaly record table after adding a fault source node at any time to establish a correlated anomaly; and reporting the correlated anomaly as an anomaly alert.

[0014] Another objective of this invention is to provide a monitoring system for low-voltage equipment in a distribution network based on multi-source data.

[0015] To address the aforementioned technical problems, this invention provides the following technical solution: a monitoring system for low-voltage distribution equipment under multi-source data, comprising: a monitoring dataset establishment module for acquiring monitoring datasets of low-voltage distribution equipment; a feature extraction module for preprocessing the monitoring datasets to extract electrical data features and image features, and establishing graph nodes based on the electrical data features and image features; an edge establishment module for determining time-evolution edges, spatial topology edges, and cross-modal edges based on the graph nodes, and establishing a bimodal anomaly heterogeneous evolution graph by combining the graph nodes, time-evolution edges, spatial topology edges, and cross-modal edges; and a fault source node identification module for tracking anomaly propagation based on the bimodal anomaly heterogeneous evolution graph to complete fault source node identification.

[0016] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the method for monitoring low-voltage equipment in a distribution network under multi-source data.

[0017] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method for monitoring low-voltage equipment in a distribution network under multi-source data.

[0018] The beneficial effects of this invention are as follows: By acquiring a multi-source monitoring dataset containing electrical and visual modalities and establishing graph nodes, this invention achieves comprehensive perception and data fusion of the operating status of low-voltage equipment in the distribution network; by constructing a bimodal anomaly heterogeneous evolution graph containing temporal evolution edges, spatial topology edges, and cross-modal edges, it realizes a unified modeling expression of multidimensional relationships between equipment, which can comprehensively describe the operating status and mutual influence mechanisms of equipment from three dimensions: temporal evolution, spatial topology, and modal correlation, effectively solving the problems of information fragmentation and insufficient correlation analysis in traditional methods; through anomaly propagation tracking and fault source node identification, it achieves accurate tracking of anomaly propagation paths and precise location of initial fault sources, improving the efficiency and accuracy of fault diagnosis; through cross-modal anomaly coupling, feature aggregation and transmission, and anomaly record table correlation analysis, it achieves automation of anomaly detection and proactive fault prevention, effectively reducing the false alarm and missed alarm rates, improving the reliability and timeliness of monitoring low-voltage equipment in the distribution network, and ultimately improving the efficiency and accuracy of fault diagnosis for low-voltage equipment in the distribution network. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0020] Figure 1 This is an overall flowchart of a method for monitoring low-voltage equipment in a distribution network using multi-source data, provided in one embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Example 1, referring to Figure 1 As one embodiment of the present invention, this embodiment provides a method for monitoring low-voltage equipment in a distribution network under multi-source data, including:

[0026] S100: Obtain the monitoring dataset of low-voltage equipment in the distribution network.

[0027] S200: Based on the monitoring dataset, preprocessing is performed to extract electrical data features and image features, and graph nodes are established based on the electrical data features and image features.

[0028] S300: Based on graph nodes, determine temporal evolution edges, spatial topological edges, and cross-modal edges, and combine graph nodes, temporal evolution edges, spatial topological edges, and cross-modal edges to establish a bimodal anomalous heterogeneous evolution graph.

[0029] S400: Based on the bimodal anomaly heterogeneous evolution graph, anomaly propagation tracking is performed to complete the identification of fault source nodes.

[0030] It should be noted that, as the last link between the power system and users, the electrical parameters and physical state of low-voltage distribution network equipment change significantly during operation. This results in multimodal, time-varying, and spatially correlated characteristics of equipment operation. Affected by environmental factors, load fluctuations, and equipment aging, the equipment status also changes continuously. Furthermore, the operation of low-voltage distribution network equipment in the power grid environment makes the fusion and analysis of multi-source heterogeneous data difficult, and anomaly detection and fault location are often delayed. At the same time, since traditional monitoring methods rely on only single-modal data, it is difficult to comprehensively capture the characteristics of equipment operation, thus compromising the accuracy of equipment health monitoring. Single-modal monitoring has significant limitations in cross-modal correlation identification and may also lead to inaccurate fault source location due to incomplete information.

[0031] Therefore, to address the aforementioned issues of single-modal data monitoring information being incomplete and unable to fully capture equipment operating characteristics, resulting in poor monitoring efficiency and accuracy of low-voltage equipment in distribution networks and the inability to promptly detect potential faults and accurately locate fault sources, the following steps (S100-S400) are used to construct a dual-modal anomaly heterogeneous evolution graph that fuses electrical and visual modal data. This yields temporal evolution relationships, spatial topological relationships, and cross-modal relationships under the multi-dimensional feature associations of equipment, achieving effective integration of multi-source data and accurate extraction of anomaly features. Through comprehensive monitoring of the operating status of low-voltage equipment in the distribution network, the propagation path of anomalies in the graph structure is tracked, enabling precise early warning of potential faults. Simultaneously, based on graph neural network anomaly recognition and cross-modal anomaly coupling, accurate identification and location of fault source nodes are achieved.

[0032] Example 2, refer to Figure 1 As an embodiment of the present invention, a method for monitoring low-voltage equipment in a distribution network based on multi-source data is provided based on the previous embodiment, including:

[0033] In this embodiment of the invention, obtaining the monitoring dataset of the low-voltage equipment in the distribution network in step S100 refers to using the data acquisition layer to perform equipment monitoring of the low-voltage equipment in the distribution network and obtain the monitoring dataset, which includes electrical modal dataset and visual modal dataset. The data acquisition layer includes a smart acquisition terminal (DTU), an online monitoring unit, and a visual acquisition unit (high-definition camera or inspection drone and thermal imager).

[0034] Specifically, in step S100, obtaining the electrical modal dataset can be performed as follows:

[0035] By using intelligent data acquisition terminals (DTUs) or online monitoring units distributed in low-voltage lines, distribution boxes, transformers and other equipment, various electrical parameters during equipment operation are collected in real time, including but not limited to voltage, current, active power, frequency fluctuation, power factor, THD harmonic distortion rate, etc., to understand the electrical operating status of the equipment and determine whether there are abnormal conditions such as overload, undervoltage, and overcurrent. These electrical parameter monitoring data constitute an electrical modal dataset.

[0036] Specifically, in step S100, the visual modality dataset is obtained, and the specific operations can be as follows:

[0037] The equipment is photographed by cameras or inspection drones and thermal imagers installed near the low-voltage equipment in the distribution network. The cameras can be fixed in positions such as poles and distribution boxes to take real-time pictures of the equipment's appearance. The drone inspection can monitor a large area of ​​low-voltage equipment more flexibly, taking pictures of the overall appearance of the equipment and the surrounding environment, thereby obtaining visible light images of the equipment and intuitively presenting the appearance of the equipment. The image recognition can detect whether there are bulges, charring, cracks, loose parts and other appearance changes on the equipment.

[0038] By detecting the thermal radiation on the surface of equipment using a thermal imager, a temperature field distribution image is formed. When the equipment is operating abnormally, such as poor contact or overload, it may cause local heating. Thermal imaging can capture temperature anomalies, detect potential faults in advance, and achieve non-contact, visual equipment status monitoring.

[0039] The visible light image and temperature field distribution image of the device are combined to obtain the visual modality dataset.

[0040] In one optional implementation, the acquisition of the monitoring dataset of low-voltage equipment in the distribution network in step S100 can also be achieved by deploying edge computing nodes at the distribution network site through a distributed data acquisition architecture based on edge computing, enabling local processing and preliminary screening of data, reducing data transmission volume and latency. At the same time, multi-sensor fusion technology is adopted to integrate various sensing devices such as current transformers, voltage transformers, smart meters, and environmental sensors to build a three-dimensional monitoring network. Time synchronization protocols are used to ensure the consistency of time bases of different data sources, and a data quality assessment mechanism is established to perform real-time verification and labeling of the collected data to ensure the reliability of the data foundation for subsequent analysis.

[0041] In another optional implementation, the monitoring dataset of low-voltage distribution network equipment is obtained in step S100. The equipment status perception network can also be constructed through Internet of Things (IoT) technology. Low-power wide area network (LPWAN) technologies such as narrowband IoT can be used to achieve the connection and data backhaul of a large number of equipment. The data collection frequency can be adjusted according to the equipment operating status and importance based on the sampling strategy. The sampling density can be increased for critical equipment or equipment in abnormal state. At the same time, blockchain technology is introduced to ensure the immutability and traceability of data, establish a trusted storage and sharing mechanism for equipment monitoring data, and optimize network bandwidth utilization through data compression and incremental transmission technologies.

[0042] It should be noted that this invention establishes a comprehensive monitoring dataset containing both electrical and visual modes by activating the data acquisition layer, thereby achieving all-round perception and recording of the operating status of low-voltage equipment in the distribution network. Compared with the existing technology that relies solely on single-mode data monitoring, this invention solves the problems of traditional methods in monitoring information being one-sided and unable to fully capture the operating characteristics of equipment. In particular, by monitoring electrical parameters, it can reflect changes in the internal electrical performance of the equipment, and by visual monitoring, it can intuitively present the appearance and thermal radiation of the equipment. The combination of the two modes of data enables a comprehensive understanding of the equipment's operating status from both internal electrical characteristics and external physical conditions, laying a solid data foundation for subsequent multimodal data fusion analysis.

[0043] In this embodiment of the invention, step S200 involves preprocessing the monitoring dataset to extract electrical data features and image features, and then establishing graph nodes based on these features, including the following steps A1-A3:

[0044] A1: Perform data preprocessing on the monitoring dataset.

[0045] A2: Extract electrical data features and image features from the preprocessed monitoring dataset.

[0046] A3: Establish graph nodes by combining electrical data features and image features.

[0047] Specifically, in step A1, data preprocessing of the monitoring dataset refers to the fact that the collected electrical modal dataset and visual modal dataset often have problems such as noise and missing values. Preprocessing of the monitoring dataset includes data cleaning and data transformation.

[0048] For example, in step A1, data preprocessing of the monitoring dataset can be performed, specifically through the following steps:

[0049] Data cleaning removes outliers from electrical data, such as obviously erroneous voltage and current values ​​caused by sensor malfunctions; blurry or damaged images are removed from visual images; missing values ​​in the data are also addressed by using interpolation methods (such as linear interpolation and Lagrange interpolation) to fill in a small number of missing electrical parameters; for visual images, if some areas are missing, they can be repaired based on information from surrounding images or the image can be directly removed.

[0050] By performing data conversion, including normalizing electrical data, parameters such as voltage and current of different magnitudes are converted to the same range, which facilitates subsequent analysis and comparison; for visual images, grayscale conversion and normalization are performed to unify image format and color space, thereby reducing computational complexity.

[0051] Specifically, in step A2, electrical data features and image features are extracted from the preprocessed monitoring dataset. The specific operations may include:

[0052] By employing mathematical and signal processing methods, key features can be extracted from preprocessed electrical data. For example, the voltage and current signals in the time domain can be converted to the frequency domain using Fast Fourier Transform (FFT) to extract harmonic features. Statistical features such as the mean, variance, maximum, and minimum values ​​of electrical parameters over a period of time can be calculated to characterize the stability and fluctuations of the equipment's electrical operation. Furthermore, machine learning algorithms (such as Principal Component Analysis (PCA)) can be used to extract feature vectors that represent the main information of the electrical data, reducing the data dimensionality and ultimately obtaining the electrical data features.

[0053] For the preprocessed visual images, computer vision techniques are used to extract features, namely edge features (such as using the Canny operator to detect the edges of the device image and highlight the device outline) and corner features (such as Harris corner detection to locate the key corners of objects in the image). Based on convolutional neural networks (CNN), such as ResNet and VGG models, deep and abstract features are automatically extracted from the images, which can more accurately describe the details and overall structure of the device's appearance, and finally obtain image data features.

[0054] Furthermore, in step A3, the creation of graph nodes by combining electrical data features and image features refers to combining the electrical data features and image features of the same device to form graph nodes.

[0055] It should be noted that, since graph nodes are the basic units in graph structure data, each graph node represents a low-voltage distribution network device. The electrical data characteristics and image characteristics of the device are integrated as attributes of the graph node. For example, for transformers, the extracted electrical characteristics such as voltage and current, as well as thermal imaging and visible light image characteristics, are packaged into node information. Each graph node integrates the multimodal information of the device, thus comprehensively presenting the operating status of the device from both electrical performance and physical state aspects, ensuring the reliability of monitoring low-voltage distribution network devices.

[0056] In one optional implementation, step S200 involves preprocessing the monitoring dataset to extract electrical data features and image features, and establishing graph nodes based on these features. Alternatively, a deep learning-based multimodal feature fusion framework can be used to automatically learn the importance weights of different modal features using an attention mechanism, constructing an end-to-end feature extraction network. Simultaneously, generative adversarial networks (GANs) are used for data augmentation to generate diverse training samples, thereby improving the generalization ability of feature extraction. Furthermore, transfer learning techniques are employed to adapt models pre-trained in other domains to the power distribution equipment monitoring scenario, accelerating model convergence and enhancing feature representation capabilities. Finally, a self-supervised learning method is used to mine the inherent correlations within the multimodal data, achieving high-quality feature extraction without requiring a large amount of labeled data.

[0057] In another optional implementation, step S200 involves preprocessing the monitoring dataset to extract electrical data features and image features, and establishing graph nodes based on these features. Alternatively, a graph embedding-based feature representation learning method can be used to map the multimodal features of the equipment to a low-dimensional vector space, maintaining the topological relationships and semantic similarities between features. A variational autoencoder (VAE) is used to learn the probability distribution of features and handle data uncertainty. At the same time, a multi-scale feature fusion strategy is introduced to extract feature information from different granularities and levels. A hierarchical clustering method is used to identify equipment types and operating modes, establishing a feature-driven equipment classification system. Causal reasoning techniques are used to analyze the causal relationships between features, constructing a more robust and interpretable graph node representation.

[0058] It should be noted that this invention, through systematic preprocessing and feature extraction of multi-source monitoring datasets, transforms the original heterogeneous data into a structured graph node representation, achieving effective fusion of electrical and visual modal information. Compared with the existing technology of simply splicing or independently processing different modal data, this invention solves the problems of insufficient feature expression and isolated information between modalities in traditional methods. In particular, it eliminates format differences and quality issues between different data sources through a unified data preprocessing process, fully explores the frequency domain features of electrical data and the spatial features of image data through targeted feature extraction methods, and deeply fuses multimodal features through the construction of graph nodes. This not only improves the expressive power and discriminative power of features, but also lays a solid foundation for subsequent graph structure analysis, enabling a more comprehensive and accurate description of the operating status of low-voltage distribution network equipment.

[0059] In this embodiment of the invention, step S300 involves determining temporal evolution edges, spatial topological edges, and cross-modal edges based on graph nodes, and establishing a bimodal anomalous heterogeneous evolution graph by combining graph nodes, temporal evolution edges, spatial topological edges, and cross-modal edges, including the following steps B1-B4:

[0060] B1: Perform feature temporal change analysis on graph nodes and establish time evolution edges.

[0061] B2: Perform spatial topological relationship analysis on the graph nodes and establish spatial topological edges.

[0062] B3: Perform cross-modal association identification on graph nodes and establish cross-modal edges.

[0063] B4: Construct a bimodal anomalous heterogeneous evolution graph based on graph nodes, temporal evolution edges, spatial topological edges, and cross-modal edges.

[0064] Specifically, in step B1, performing feature time-series change analysis on graph nodes and establishing time evolution edges refers to acquiring the electrical data characteristics and image characteristics of each graph node (i.e., each low-voltage distribution network device) at different time points, and then performing trend analysis and periodic analysis. The specific operations can be as follows:

[0065] By utilizing trend analysis, including the application of statistical analysis and signal processing methods, we can analyze the changing trends of various features over time. For electrical data features, such as observing the fluctuations of voltage and current over time, we can calculate the changes in their mean and variance over time to determine whether there is a gradual upward or downward trend. For example, a slow and prolonged decrease in voltage may indicate a potential problem in the line.

[0066] For image features, analyze the evolution of features related to changes in the appearance of the equipment (such as thermal imaging temperature, degree of damage to the appearance of components) over time, such as whether the temperature of a certain part of the equipment continues to rise, or whether cracks gradually expand.

[0067] It should be noted that the operating parameters of many low-voltage distribution network devices are periodic. By analyzing the periodic components in electrical and image features using Fourier transform, the period length and the occurrence time of peaks and valleys can be determined to distinguish between normal periodic fluctuations and abnormal changes. Then, based on trend analysis and periodic analysis, reasonable thresholds are set to detect anomalies. For example, if electrical parameters exceed the normal fluctuation range, or if image features show unexpected changes (such as a component that was originally without cracks suddenly developing obvious cracks), it is determined to be an abnormal situation, and the time point and specific characteristics of the anomaly are recorded. Finally, based on the results of feature time-series change analysis, when it is determined that a device has feature changes at different time points, time evolution edges are added between the graph nodes at the corresponding time points. The time evolution edge is the edge connecting the same graph node (device) at different time points, representing the evolution relationship of the device state over time, reflecting the continuity and change process of the device's operating state in the time dimension.

[0068] Specifically, in step B2, spatial topological relationship analysis is performed on the graph nodes to establish spatial topological edges. The specific operations can be as follows:

[0069] By clarifying the relative position of low-voltage distribution network equipment in the power grid topology, such as the specific node location on a particular low-voltage line, and its connection relationship with other surrounding equipment, and then by reviewing line wiring diagrams and equipment ledgers, the physical connection methods between low-voltage distribution network equipment can be determined. For example, transformers and distribution boxes, and distribution boxes and power terminals are directly connected by wires. At the same time, the indirect connection relationships between equipment are also clarified, such as connections achieved through busbars, switches, and other equipment. Based on the connection relationships of the equipment, the adjacent equipment of the equipment is then identified, that is, equipment that is directly connected to the target equipment in the spatial topology or is close to it and closely related.

[0070] Based on the spatial topology analysis results, spatial topology edges are added between graph nodes corresponding to devices that have connections or adjacencies. These spatial topology edges are used to connect graph nodes (low-voltage distribution network devices) that have spatial topology relationships, representing the spatial association of devices, reflecting the power transmission path and the interaction between devices. Through spatial topology edges, voltage drop, power loss, etc. during power transmission can be analyzed. When a device fails, the scope of the fault and possible fault propagation paths can be quickly located based on the spatial topology edges.

[0071] Specifically, in step B3, cross-modal association identification is performed on graph nodes, and cross-modal edges are established. The specific operations can be as follows:

[0072] Cross-modal association recognition is performed on graph nodes, that is, the relationship between electrical modal and visual modal data is identified. For example, when there is a current overload in the electrical data, it may be manifested as the equipment heating up, discoloration or sparking in the visual image. By analyzing a large amount of electrical data and corresponding image data, the correspondence between different modalities can be found.

[0073] After identifying cross-modal associations, cross-modal edges are established between the corresponding graph nodes. For example, if there is an association between the electrical data characteristics of a device at a certain moment (such as excessive current) and the image characteristics of the device at the same moment (such as color changes caused by local overheating), then cross-modal edges are established between the graph node representing the electrical data characteristics of the device at that moment and the graph node representing the image characteristics. This represents the direct connection between the two different modal characteristics, enabling the graph to more comprehensively reflect the operating status of the device.

[0074] Furthermore, in step B4, a bimodal anomaly heterogeneous evolution graph is established based on graph nodes, temporal evolution edges, spatial topological edges, and cross-modal edges. The specific operations can be as follows:

[0075] By connecting and combining all graph nodes according to temporal evolution edges, spatial topology edges, and cross-modal edges, a bimodal anomaly heterogeneous evolution graph is generated. This graph comprehensively describes the operating status of low-voltage distribution network equipment from multiple dimensions, including the equipment's own multimodal characteristics, temporal evolution, spatial topological relationships, and correlations between different modes. This allows for more accurate monitoring of equipment status, detection of potential anomalies, fault location, and prediction of future equipment development trends, thereby improving the reliability and stability of low-voltage distribution network equipment and ensuring the safe operation of the power system.

[0076] In one optional implementation, step S300 establishes a bimodal anomalous heterogeneous evolution graph. Alternatively, a time-varying graph neural network can be constructed using graph learning methods to capture the evolution pattern of the device state over time. An attention mechanism is used to adjust the weights of different time steps to identify key time nodes and state transition patterns. Simultaneously, a graph convolutional neural network is introduced to process spatial topological relationships. Through multi-layer graph convolution operations, the mutual influence between neighboring nodes is learned, and a contrastive learning method is used to enhance the correlation of cross-modal features. The intrinsic connections between modalities are learned through the construction of positive and negative sample pairs. Finally, graph regularization techniques are used to maintain the stability and continuity of the graph structure.

[0077] In another optional implementation, the bimodal anomaly heterogeneous evolution graph established in step S300 can also be constructed using a multi-scale analysis method based on graph signal processing. This method can build a hierarchical graph structure at different time and spatial scales, and use wavelet transform and graph Fourier transform to analyze the frequency domain characteristics of the graph signal to identify the periodicity and propagation mode of anomaly diffusion. At the same time, random walk theory is introduced to model the diffusion process of anomalies on the graph, and Markov chain analysis is used to predict the propagation path and arrival probability of anomalies. Furthermore, graph embedding technology is used to map the high-dimensional graph structure to a low-dimensional space, maintaining the similarity and connection mode between nodes, thereby improving the computational efficiency and interpretability of graph analysis.

[0078] It should be noted that this invention achieves unified modeling and expression of multidimensional correlations among low-voltage distribution network equipment by constructing a bimodal anomaly heterogeneous evolution graph containing temporal evolution edges, spatial topology edges, and cross-modal edges. Compared with the existing technology that processes temporal information, spatial relationships, and modal correlations separately, this invention solves the problems of information fragmentation and insufficient correlation analysis in traditional methods. In particular, the temporal evolution edges can capture the temporal evolution law of equipment status, the spatial topology edges can reflect the physical connection and mutual influence between equipment, and the cross-modal edges can establish the correlation mapping between electrical features and visual features. The organic combination of the three types of edges constitutes a complete heterogeneous graph network, which not only preserves the multidimensional characteristics and inherent correlations of the original data, but also provides a unified analytical framework for subsequent anomaly detection and fault tracing. It can comprehensively analyze the generation, propagation, and evolution process of equipment anomalies from three dimensions: time, space, and modality, thereby improving the comprehensiveness and accuracy of monitoring low-voltage distribution network equipment.

[0079] In this embodiment of the invention, step S400 involves tracking the anomaly propagation based on a bimodal anomaly heterogeneous evolution graph to identify the fault source node, including the following steps C1-C3:

[0080] C1: Perform anomaly identification on graph nodes and establish anomaly node identifiers.

[0081] C2: Perform anomaly propagation analysis based on the bimodal anomaly heterogeneous evolution graph and establish the propagation chain of anomaly nodes.

[0082] C3: Utilize cross-modal edges to perform cross-modal anomaly coupling of graph nodes, and combine the cross-modal anomaly coupling results and propagation chains to perform joint anomaly propagation tracking, thereby completing the identification of fault source nodes.

[0083] Specifically, in step C1, anomaly identification is performed on graph nodes, and anomaly node identifiers are established. The specific operations can be as follows:

[0084] Anomaly identification is performed on each graph node using a graph neural network. A graph neural network is a deep learning model used to process graph structure data. By learning a large amount of graph node data under normal and abnormal states, it can automatically extract features and identify which graph nodes represent abnormal equipment states. The graph nodes of the bimodal anomaly heterogeneous evolution graph are used as input. These graph nodes contain multimodal information such as electrical data features and image features of low-voltage distribution network equipment. Then, the graph nodes with abnormal equipment states are identified and anomaly labels are set for them. These labels can be a Boolean value (e.g., True for abnormal, False for normal) or an anomaly category label (e.g., overcurrent anomaly or overheating anomaly).

[0085] Specifically, in step C2, the abnormal propagation analysis based on the bimodal abnormal heterogeneous evolution graph and the establishment of the propagation chain of the abnormal node refers to the abnormal propagation analysis along the time speech edge and the spatial topology edge according to the graph structure propagation rules of the bimodal abnormal heterogeneous evolution graph, and the establishment of the propagation chain of the abnormal node. The propagation chain is set with a propagation probability label.

[0086] It should be noted that the graph structure propagation rules of the bimodal anomaly heterogeneous evolution graph are based on the characteristics of temporal evolution edges and spatial topological edges to define how anomalies propagate in the graph. Then, anomaly propagation analysis is performed along the temporal evolution edges, that is, analyzing the graph node states at the time points before and after the time evolution edges are connected, as well as the changing trends of their characteristics, to infer the propagation of anomalies in the time dimension. For example, if a graph node is marked as an anomaly, graph nodes at adjacent time points may also be marked as anomalies, thus forming a temporal anomaly propagation chain. Then, anomaly propagation analysis is performed along the spatial topological edges, that is, analyzing the propagation path of anomalies between devices based on the physical connections of the devices and spatial influences. For example, an interruption or anomaly in power transmission may cause anomalies in the input voltage or current of the distribution box, thus including the graph node of the distribution box in the anomaly propagation chain. Furthermore, by analyzing the propagation of similar anomalies in historical data, a propagation probability label is set for each edge of the propagation chain to more accurately describe the possibility of anomaly propagation. For example, if two devices are electrically connected and one device has historically experienced an anomaly, the other device is more likely to experience an anomaly, so a higher propagation probability is set. For time evolution edges, if the device state changes relatively smoothly between adjacent time points, the anomaly propagation probability is relatively low.

[0087] It should be noted that in step C3, cross-modal edges are used to perform cross-modal anomaly coupling of graph nodes, that is, to analyze the anomaly correlation of different modal features. For example, when an electrical data display device experiences an overcurrent anomaly, the corresponding image features are checked through cross-modal edges to see if there are any anomalies such as device overheating or arcing. If so, it indicates the existence of cross-modal anomaly coupling, which can more comprehensively confirm the abnormal state of the device. Then, the cross-modal anomaly coupling results are combined with the anomaly propagation chain for joint analysis to more accurately track the diffusion path of the anomaly throughout the graph. For example, when multiple devices in a certain area are found to have anomalies in both the electrical and image modes, and exhibit certain propagation patterns in time and space, the diffusion range and path of the anomaly can be determined more accurately.

[0088] Specifically, step C3 utilizes cross-modal edges to perform cross-modal anomalous coupling of graph nodes, including the following steps C31-C33:

[0089] C31: Obtain the required monitoring accuracy of low-voltage equipment in the distribution network, and configure the calibration traceability window according to the required monitoring accuracy.

[0090] C32: Select any modal data as the starting point for tracing and obtain the anomaly degree of the starting point for tracing under the current modality.

[0091] C33: Adjust the traceability window based on the anomaly degree and perform correlation capture of another modality's data, and complete cross-modal anomaly coupling based on the correlation capture results.

[0092] Specifically, in step C31, obtaining the demand monitoring accuracy of low-voltage distribution network equipment refers to determining the accuracy of monitoring equipment operating parameters based on the importance of the equipment, the operating environment, and the user's requirements for power supply quality. Then, the calibration traceability window is configured according to the demand monitoring accuracy. The higher the monitoring accuracy, the more refined the traceability window is set, so as to more accurately capture minute changes in the equipment status; while when the monitoring accuracy is low, the traceability window can be relatively large.

[0093] Specifically, in step C32, selecting any modal data as the tracing starting point and obtaining the anomaly degree of the tracing starting point under the current modality means taking any modal data of the low-voltage equipment in the distribution network as the tracing starting point, obtaining the anomaly degree of the tracing starting point under the current modality, comparing the modal data of the current tracing starting point with a pre-set normal data threshold, and calculating the degree of anomaly. The greater the difference between the modal data of the tracing starting point and the normal data threshold, the higher the corresponding anomaly degree.

[0094] Specifically, in step C33, the tracing window is adjusted according to the anomaly degree, and the correlation capture of another modality's data is performed. Based on the correlation capture results, cross-modal anomaly coupling is completed. The specific operation can be as follows:

[0095] Adjust the size of the traceability window based on the anomaly level of the traceability starting point. If the anomaly level is high, it indicates that the equipment may have a serious anomaly. Expand the traceability window to more comprehensively analyze the equipment's status changes over a longer period and find the root cause of the anomaly. If the anomaly level is low, conduct further analysis within a smaller traceability window.

[0096] After compensating the tracing window, based on the information within the current tracing window, relevant information from another modality of data is captured. For example, using the current anomaly in the electrical data as the starting point for tracing, the tracing window is adjusted, and image data within the same time range is captured in association to check for image features related to current anomalies, such as equipment overheating or arcing. By capturing the associated data from different modalities, the abnormal information under different modalities is coupled to comprehensively analyze the abnormal condition of the equipment, so as to more accurately determine the type and severity of the equipment failure.

[0097] It should be noted that by using each modality as a starting point for analysis, abnormal information about the device can be discovered from different perspectives, and the results obtained from different starting points can complement and corroborate each other, thereby gaining a more comprehensive understanding of the device's abnormal situation.

[0098] Furthermore, in step C3, the joint anomaly propagation tracing, which combines the cross-modal anomaly coupling results and the propagation chain, is used to identify the fault source node. This involves analyzing the starting point, propagation path, and changes in different modal characteristics of the anomaly propagation using the joint anomaly propagation tracing to identify the fault source node, i.e., the graph node corresponding to the device that initially caused the anomaly. For example, if the anomaly of a certain device occurs earliest, and the anomalies of other devices propagate from that device through time evolution edges and spatial topology edges, and the initial anomaly characteristics of that device in the electrical and image modalities can also be found in the cross-modal analysis, then the graph node corresponding to that device is identified as the fault source node. This helps to quickly locate and resolve faults in low-voltage distribution network equipment, thereby improving fault monitoring efficiency.

[0099] In this embodiment of the invention, step S400, which involves tracking the anomaly propagation based on a bimodal anomaly heterogeneous evolution graph to identify the fault source node, also includes the following steps D1-D4:

[0100] D1: Transform the electrical data features and image features of the graph nodes into a combined feature vector.

[0101] D2: Using any graph node as the starting node, perform node association analysis using time-evolution edges, spatial topology edges, and cross-modal edges to establish a set of neighboring nodes.

[0102] D3: Reconstruct the graph structure based on the starting node and the set of neighboring nodes. Perform feature aggregation on the feature vectors of the farthest neighboring node in the reconstructed graph structure and pass it to the next neighboring node.

[0103] D4: After the feature aggregation results of all neighboring nodes are transmitted to the starting node, the updated starting node is used to identify abnormal nodes, and the abnormal spread tracking is completed based on the abnormal node identification results and the feature correlation of the reconstructed graph structure.

[0104] Specifically, in step D1, converting the electrical data features and image features of the graph nodes into a combined feature vector means mapping the electrical data features and image features of each graph node into a vector space to obtain a combined feature vector, which is used to comprehensively describe the equipment status represented by the graph node.

[0105] Furthermore, in step D2, any graph node is used as the starting node, and node association analysis is performed using time-evolution edges, spatial topological edges, and cross-modal edges to establish a set of neighboring nodes. Specifically, this can be done as follows:

[0106] Randomly select a node from all graph nodes as the starting point of the analysis, and use time evolution edges, spatial topology edges, and cross-modal edges to perform node association analysis. In particular, by considering the changes in device status over time, other nodes related to the starting node in the time series are determined by time evolution edges.

[0107] Based on the spatial location and connection relationships between devices, nodes that are spatially adjacent to or related to the starting node are determined by combining spatial topology edges.

[0108] Since there are different modalities such as electrical data and image data, cross-modal edges are used to establish connections between data of different modalities.

[0109] Starting from the starting node, first determine the first-level neighbor nodes directly connected to the starting node, and then continue to search for their neighbor nodes from these first-level neighbor nodes. This process is repeated to gradually expand the set of neighbor nodes until certain search conditions are met or a preset search depth is reached. Then, nodes that have a direct or indirect relationship with the starting node are combined into a set of neighbor nodes.

[0110] It should be noted that in step D3, when reconstructing the graph structure based on the starting node and the set of neighboring nodes, the focus is on the starting node and its related neighboring nodes, highlighting their connection relationships and feature associations. Then, starting from the farthest neighboring node in the reconstructed graph structure, feature aggregation is performed on its combined feature vectors. For example, multiple feature vectors are averaged or weighted and then fused. The feature information of the farthest neighboring node is then integrated to form a feature vector, which contains the comprehensive feature information of the neighboring node and its surrounding related nodes. The aggregated feature vector is then passed to its adjacent upper-level neighboring nodes.

[0111] Specifically, in step D3, feature aggregation of the combined feature vectors is performed from the farthest neighbor node of the reconstructed graph structure, and then passed to the next neighbor node, including the following steps D31-D33:

[0112] D31: Aggregate the combined feature vectors within the farthest neighbor node to generate the first feature aggregation result.

[0113] D32: Send the first feature aggregation result to the previous neighbor node. The previous neighbor node receives the first feature aggregation result and then performs aggregation.

[0114] D33: Repeatedly execute the aggregation by passing it up layer by layer until the feature aggregation results of all neighboring nodes are passed to the starting node.

[0115] Specifically, in step D31, the combined feature vectors within the farthest neighbor node are aggregated. The specific operation can be as follows:

[0116] By using an arithmetic average method, which sums the corresponding elements of the combined feature vectors of all the farthest neighbor nodes and then divides by the number of nodes, or by using a weighted average method, which assigns different weights to the feature vectors of each node according to their importance, and then sums them up.

[0117] Set an outlier threshold and determine which nodes are outliers. For example, calculate the distance between each combined feature vector and other vectors (such as Euclidean distance). If the distance exceeds the set outlier threshold, then the nodes are determined to be non-outliers. After removing the outliers, the remaining nodes are aggregated to obtain the first feature aggregation result, which more accurately reflects the feature situation of the farthest neighbor node set.

[0118] Specifically, in step D32, the first feature aggregation result is sent to the previous neighbor node. After receiving the first feature aggregation result, the previous neighbor node performs aggregation, which means that the generated first feature aggregation result is sent from the farthest neighbor node to the previous neighbor node, so that the previous neighbor node obtains the comprehensive feature information of the farthest neighbor node set and aggregates it with its own feature information (also in the form of a combined feature vector).

[0119] Specifically, in step D33, the repeated execution of layer-by-layer upward aggregation until the feature aggregation results of all neighboring nodes are transmitted to the starting node means that after completing the aggregation of the previous neighboring node, the updated feature aggregation results are sent to the next higher-level neighboring node, and so on. According to the hierarchical relationship of neighboring nodes in the reconstructed graph structure, the aggregation results are transmitted layer by layer upward. After receiving the feature aggregation results transmitted from the next level, each neighboring node performs aggregation until the feature aggregation results of all neighboring nodes are transmitted to the starting node. The transmission stops when the starting node receives the feature aggregation results of all neighboring nodes and completes the final aggregation.

[0120] Furthermore, in step D4, after the feature aggregation results of all neighboring nodes are transmitted to the starting node, the updated starting node is used to identify abnormal nodes, and the anomaly propagation tracking is completed based on the abnormal node identification results and the feature correlation of the reconstructed graph structure. Specifically, this can be achieved through:

[0121] Once the feature aggregation results of all neighboring nodes have been passed to the starting node, the feature vector of the starting node is updated, containing relevant information about all its neighboring nodes.

[0122] The updated starting point feature vector is compared with the feature vector under normal conditions to determine whether the starting node is an abnormal node.

[0123] Based on the results of abnormal node identification and the feature correlation between nodes in the reconstructed graph structure, the anomaly is tracked to see how it spreads in the graph structure. If a neighboring node has a high feature correlation with the starting node of the anomaly, and the neighboring node is also identified as an anomaly or has a high probability of being an anomaly, it is inferred that the anomaly may have spread from the starting node to the neighboring node, thus completing the anomaly spread tracking.

[0124] In this embodiment of the invention, after the fault source node identification is completed in step S400, the following steps E1-E2 are included:

[0125] E1: Fit the development of potential hazards based on the fault source node and the propagation path, and establish the fitting anomaly level.

[0126] E2: Determine the true anomaly level based on the fault source node, reconstruct the anomaly based on the true anomaly level and the fitted anomaly level, and report it.

[0127] Specifically, in step E1, the development of potential hazards is fitted based on the fault source node and the propagation path. The specific operation can be as follows:

[0128] By analyzing the source node of the fault and the propagation of the anomaly along the diffusion path in the network, the development of hidden dangers is fitted. That is, the initial abnormal characteristics of the source node of the fault and the state changes of each node on the diffusion path are simulated, the degree of development of hidden dangers is quantified, and the fitted anomaly level is obtained, which represents the severity of hidden dangers inferred based on the current source node of the fault and the diffusion path.

[0129] Furthermore, in step E2, the true anomaly level is determined based on the fault source node, and the anomaly is reconstructed and reported based on the true anomaly level and the fitted anomaly level. Specifically, this process can be as follows:

[0130] The actual anomaly level is determined based on the actual operating parameters of the equipment at the fault source node, the fault type, and the severity.

[0131] Based on the actual anomaly level and the fitted anomaly level, the entire anomaly situation is reconstructed and adjusted, and the reconstructed anomaly situation is reported and displayed so that the true anomaly status of the low-voltage equipment in the distribution network can be understood in a timely manner.

[0132] Furthermore, step E2, which determines the true anomaly level based on the fault source node, reconstructs and reports the anomaly according to the true anomaly level and the fitted anomaly level, also includes the following steps E21-E23:

[0133] E21: Create an anomaly record table based on the fault source node.

[0134] E22: After adding a fault source node at any time, call the anomaly record table to perform correlation analysis and establish associated anomalies.

[0135] E23: Report the associated exception as an exception message.

[0136] For example, in step E21, an anomaly record table is created based on the fault source node. The specific operation could be:

[0137] An anomaly log table is created for each fault source node to record various information related to that fault source node in detail, including the name of the equipment represented by the fault source node, the equipment number, the specific location of the equipment; the time and date of the fault; the operating parameters of the equipment at the time of the fault, including electrical data characteristics (such as voltage, current, power, etc.) and image characteristics (such as the appearance of the equipment, whether there is obvious damage or abnormal heating, etc.); the type of fault (such as short circuit fault, overload fault, etc.) and possible fault cause analysis.

[0138] Furthermore, in step E22, after a fault source node is added at any time, the anomaly record table is called for correlation analysis. Establishing a correlation anomaly means that during the operation of low-voltage equipment in the distribution network, due to equipment aging, changes in the external environment, improper operation, etc., a new fault source node may appear at any time. Therefore, the anomaly record table is called for correlation analysis. By comparing and analyzing the characteristics of the new fault source node (such as equipment type, fault characteristics, occurrence time, etc.) with the existing fault source node information in the anomaly record table, the correlation relationship is determined. Then, a correlation anomaly is established based on the correlation analysis results, indicating that the new fault may be related to some historical faults, may be caused by the same reason, or have a certain causal relationship, or may have occurred in a similar operating environment or under similar conditions.

[0139] Furthermore, step E23, reporting the associated anomaly as an anomaly prompt, means reporting the associated anomaly as an anomaly prompt. For example, a detailed description of the associated anomaly can be displayed on the monitoring interface, so that the equipment can be inspected and maintained in a more targeted manner, equipment faults can be detected in a timely manner, and the stability and reliability of the low-voltage equipment in the distribution network can be ensured.

[0140] It should be noted that this invention, by constructing a multi-layered anomaly propagation tracking mechanism, achieves fully automated analysis from anomaly detection to precise fault source localization. Compared with existing fault location methods that rely on manual experience or simple rule matching, this invention solves the problems of low localization accuracy, slow response speed, and high false alarm / missed alarm rates in traditional methods. In particular, it can automatically learn anomaly patterns through graph neural network anomaly recognition, improve detection reliability by integrating multi-source information through cross-modal anomaly coupling, accurately track the propagation path and mechanism of anomalies through feature aggregation and propagation analysis, and accumulate fault knowledge and prevent the recurrence of similar faults through anomaly record tables and correlation analysis. This not only improves the accuracy and timeliness of fault source identification but also establishes a continuously improving fault monitoring system, providing a scientific basis for preventive maintenance and precise operation and maintenance of low-voltage distribution network equipment, and effectively reducing economic losses and power outage risks caused by equipment failures.

[0141] In summary, this invention achieves comprehensive perception and data fusion of the operating status of low-voltage distribution network equipment by acquiring multi-source monitoring datasets containing electrical and visual modalities and establishing graph nodes. By constructing a bimodal anomaly heterogeneous evolution graph containing temporal evolution edges, spatial topology edges, and cross-modal edges, it realizes a unified modeling expression of multidimensional equipment relationships, comprehensively describing the equipment operating status and mutual influence mechanisms from three dimensions: temporal evolution, spatial topology, and modal correlation. This effectively solves the problems of information fragmentation and insufficient correlation analysis in traditional methods. Through anomaly propagation tracking and fault source node identification, it achieves accurate tracking of anomaly propagation paths and precise location of initial fault sources, improving the efficiency and accuracy of fault diagnosis. Through cross-modal anomaly coupling, feature aggregation and transmission, and anomaly record table correlation analysis, it achieves automation of anomaly detection and proactive fault prevention, effectively reducing false alarm and missed alarm rates, improving the reliability and timeliness of low-voltage distribution network equipment monitoring, and ultimately enhancing the efficiency and accuracy of fault diagnosis for low-voltage distribution network equipment.

[0142] Example 3 is an embodiment of the present invention, which provides a monitoring system for low-voltage distribution equipment under multi-source data, including: a monitoring dataset establishment module for acquiring a monitoring dataset of low-voltage distribution equipment; a feature extraction module for preprocessing the monitoring dataset to extract electrical data features and image features, and establishing graph nodes based on the electrical data features and image features; an edge establishment module for determining time evolution edges, spatial topology edges, and cross-modal edges based on the graph nodes, and establishing a bimodal anomaly heterogeneous evolution graph by combining the graph nodes, time evolution edges, spatial topology edges, and cross-modal edges; and a fault source node identification module for tracking anomaly propagation based on the bimodal anomaly heterogeneous evolution graph to complete the identification of fault source nodes.

[0143] Example 4 is an embodiment of the present invention, which differs from the previous three embodiments in that: Figure 2 As shown, if the aforementioned function is implemented as 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 a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0145] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0146] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring low-voltage equipment in a distribution network using multi-source data, characterized in that: include, Obtain the monitoring dataset of low-voltage equipment in the distribution network; Preprocessing is performed on the monitoring dataset to extract electrical data features and image features, and graph nodes are established based on the electrical data features and image features; Based on the graph nodes, determine the temporal evolution edges, spatial topology edges, and cross-modal edges. Combine the graph nodes, the temporal evolution edges, the spatial topology edges, and the cross-modal edges to establish a bimodal anomalous heterogeneous evolution graph. Anomaly propagation tracking is performed based on the bimodal anomaly heterogeneous evolution graph to complete the identification of fault source nodes.

2. The method for monitoring low-voltage equipment in a distribution network under multi-source data as described in claim 1, characterized in that: Based on the aforementioned bimodal anomaly heterogeneous evolution graph, anomaly propagation tracing is performed to complete the identification of fault source nodes, including: Anomaly identification is performed on the graph nodes, and anomaly node identifiers are established; Anomaly propagation analysis is performed based on the bimodal anomaly heterogeneous evolution diagram to establish the propagation chain of the anomaly nodes; Cross-modal anomaly coupling of graph nodes is performed using the cross-modal edges. The cross-modal anomaly coupling results and the propagation chain are combined to perform joint anomaly propagation tracking, thereby completing the identification of fault source nodes.

3. The method for monitoring low-voltage equipment in a distribution network under multi-source data as described in claim 2, characterized in that: Using the cross-modal edges to perform cross-modal anomaly coupling of graph nodes includes: Obtain the required monitoring accuracy for low-voltage equipment in the distribution network, and configure a calibration traceability window based on the required monitoring accuracy; Select any modality data as the starting point for tracing and obtain the anomaly degree of the starting point for tracing under the current modality; The tracing window is adjusted according to the anomaly degree, and the association capture of another modality data is performed. Cross-modal anomaly coupling is completed based on the association capture result.

4. The method for monitoring low-voltage equipment in a distribution network under multi-source data as described in claim 3, characterized in that: Based on the aforementioned bimodal anomaly heterogeneous evolution graph, anomaly propagation tracing is performed to complete fault source node identification, which also includes: The electrical data features and image features of the graph nodes are converted into combined feature vectors; Starting with any graph node, perform node association analysis using time-evolution edges, spatial topology edges, and cross-modal edges to establish a set of neighboring nodes; The graph structure is reconstructed based on the starting node and the set of neighboring nodes. Feature aggregation of combined feature vectors is performed from the farthest neighboring node of the reconstructed graph structure, and then passed to the next neighboring node. After the feature aggregation results of all neighboring nodes are transmitted to the starting node, the updated starting node is used to identify abnormal nodes, and the abnormal spread tracking is completed based on the abnormal node identification results and the feature correlation of the reconstructed graph structure.

5. The method for monitoring low-voltage equipment in a distribution network under multi-source data as described in claim 4, characterized in that: The feature aggregation of combined feature vectors from the farthest neighbor node of the reconstructed graph structure, and its propagation to the next neighbor node, includes: Aggregate the combined feature vectors within the farthest neighbor node to generate the first feature aggregation result; The first feature aggregation result is sent to the previous neighbor node, and the previous neighbor node performs aggregation after receiving the first feature aggregation result; Repeatedly pass the aggregation upwards layer by layer until the feature aggregation results of all neighboring nodes are passed to the starting node.

6. The method for monitoring low-voltage equipment in a distribution network under multi-source data as described in claim 5, characterized in that: After the fault source node identification is completed, the following steps are included: Based on the fault source node and the propagation path, the development of hidden dangers is fitted, and the fitting anomaly level is established; The true anomaly level is determined based on the fault source node, and the anomaly is reconstructed and reported according to the true anomaly level and the fitted anomaly level.

7. The method for monitoring low-voltage equipment in a distribution network under multi-source data as described in claim 6, characterized in that: The method further includes determining the actual anomaly level based on the fault source node, reconstructing the anomaly based on the actual anomaly level and the fitted anomaly level, and reporting it. An anomaly record table is created based on the fault source node; After a fault source node is added at any time, the anomaly record table is called to perform correlation analysis and establish associated anomalies; The associated anomaly will be reported as an anomaly message.

8. A monitoring system for low-voltage distribution equipment under multi-source data, employing the monitoring method for low-voltage distribution equipment under multi-source data as described in any one of claims 1 to 7, characterized in that, include: The monitoring dataset creation module is used to acquire monitoring datasets for low-voltage equipment in the distribution network. The feature extraction module is used to preprocess the monitoring dataset, extract electrical data features and image features, and build graph nodes based on the electrical data features and image features; The edge building module is used to determine temporal evolution edges, spatial topology edges, and cross-modal edges based on graph nodes, and to build a bimodal anomalous heterogeneous evolution graph by combining graph nodes, temporal evolution edges, spatial topology edges, and cross-modal edges. The fault source node identification module is used to track the spread of anomalies based on a dual-modal anomaly heterogeneous evolution graph and complete the identification of fault source nodes.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for monitoring low-voltage equipment in a distribution network under multi-source data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for monitoring low-voltage equipment in a distribution network under multi-source data as described in any one of claims 1 to 7.

Images