A power system operation and maintenance method and system for a smart distribution cabinet for low-voltage control.

Abstract

This invention relates to the field of power operation and maintenance technology, and particularly to a power system operation and maintenance method and system for intelligent distribution cabinets used for low-voltage control. The method includes the following steps: acquiring power operation data and sensor data from the distribution cabinet, and performing power system topology analysis to obtain the dynamic topology of the power system; performing potential load anomaly analysis on the distribution cabinet sensor data to obtain power system load anomaly node data; classifying the load anomaly impact structure based on the power system load anomaly node data to obtain load anomaly impact nodes; classifying the load anomaly impact nodes according to the distribution cabinet sensor data to obtain three-phase load offset nodes and three-phase improper wiring nodes; and reconstructing the power distribution for the three-phase load offset nodes and three-phase improper wiring nodes to obtain abnormal node power distribution data. This invention helps improve the efficiency of low-voltage control and power distribution energy efficiency.

Description

Technical Field

[0001] This invention relates to the field of power operation and maintenance technology, and in particular to a power system operation and maintenance method and system for a smart distribution cabinet for low-voltage control. Background Technology

[0002] Intelligent distribution cabinets not only distribute and manage electrical loads but also perform functions such as real-time monitoring, fault early warning, and energy consumption optimization to ensure stable system operation. However, with the rapid development of intelligent and digital technologies, existing low-voltage control distribution cabinets still face many challenges in power system operation and maintenance. The operation and maintenance of low-voltage control distribution cabinets mainly relies on traditional manual inspection methods. Maintenance personnel periodically check the operating status of electrical equipment, measure key parameters such as current, voltage, and temperature, and perform fault diagnosis and maintenance based on experience. Manual inspections are cyclical and limited by inspection intervals, which may prevent the timely detection of sudden faults. For example, some electrical faults, such as short circuits, cable aging, and load overloads, may occur during inspection intervals, leading to system abnormalities or even safety accidents. Traditional distribution cabinet operation and maintenance methods typically only focus on equipment operating status, neglecting energy consumption optimization management. Because intelligent analysis of the power consumption and operating modes of each load is not performed, the system struggles to dynamically adjust power distribution strategies, resulting in energy waste and affecting overall energy efficiency. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a power system operation and maintenance method and system for a smart distribution cabinet for weak current control, so as to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a power system operation and maintenance method for a smart distribution cabinet for low-voltage control includes the following steps:

[0005] Step S1: Obtain the power operation data and sensor data of the distribution cabinet, and perform power system topology analysis based on the power operation data of the distribution cabinet to obtain the dynamic topology of the power system;

[0006] Step S2: Identify the power system energy efficiency pattern by analyzing the power distribution cabinet sensor data through the power system dynamic topology to obtain power system energy efficiency pattern data; perform potential load anomaly analysis on the power system energy efficiency pattern data to obtain power system load anomaly node data.

[0007] Step S3: Analyze the load anomaly dominant factors based on the load anomaly node data of the power system, and divide the load anomaly impact structure of the dynamic topology of the power system according to the load anomaly dominant factors to obtain the load anomaly affected nodes.

[0008] Step S4: Based on the sensor data of the distribution cabinet, evaluate the three-phase current balance of the distribution cabinet for nodes affected by abnormal loads, and classify abnormal nodes according to the current balance of the distribution cabinet to obtain the three-phase load offset nodes and the improper three-phase wiring nodes.

[0009] Step S5: Reconstruct the power distribution for the three-phase load offset nodes and improperly connected three-phase nodes to obtain the power distribution data for the abnormal nodes, and transmit it to the distribution cabinet management platform to execute the power distribution task.

[0010] The intelligent distribution cabinet anomaly detection and power reconfiguration method provided by this invention effectively solves the problems of traditional low-voltage control distribution cabinet operation and maintenance, such as reliance on manual inspection, delayed response, insufficient anomaly perception, and weak energy efficiency management capabilities, compared with existing technologies. It possesses significant technical advantages and engineering application value. By acquiring power operation data and sensor data from the distribution cabinet and combining this with in-depth modeling of the dynamic topology of the power system, it can restore the electrical connection relationships at the device level while ensuring data comprehensiveness, providing a structural foundation for subsequent energy efficiency identification and anomaly analysis. The introduced Z-score standardization and time offset correction threshold (±1 second) can eliminate errors caused by asynchronous sampling of devices while ensuring data accuracy, improving the stability and robustness of anomaly analysis. Energy efficiency pattern recognition based on the topology structure can not only achieve node-level energy consumption behavior modeling but also uncover high-risk load areas associated with topological locations, enabling precise locking of potential load anomaly nodes, effectively compensating for the shortcomings of coarse anomaly detection granularity and low sensitivity in traditional strategies. By extracting the dominant factors of load anomalies and dividing the topological influence structure, a multi-dimensional analysis of the causes of load anomalies is achieved. This not only identifies "where the problem occurs" but also clarifies "why it occurs," facilitating targeted optimization and scheduling of the system. During the three-phase current balance assessment stage, setting balance interval thresholds (e.g., 0.85, 0.65) allows for hierarchical classification and management of abnormal nodes, improving the accuracy of anomaly identification. The introduction of the improper wiring identifier variable fi∈{0,1} constructs a joint structural and behavioral discrimination framework, effectively enhancing the interpretability and reliability of anomaly classification. Finally, through power distribution reconstruction, differentiated distribution strategies are adjusted for the classified abnormal nodes. Combined with the task execution capabilities of the power management platform, closed-loop anomaly processing is achieved, ensuring that the distribution system still possesses dynamic recovery and energy efficiency optimization capabilities under load or wiring anomalies. Setting the path backtracking depth to [3,5] layers controls the balance between the overhead of backtracking calculations and the coverage of anomaly information propagation. The factor combination extraction strategy with a cumulative contribution rate greater than 0.85 ensures the sufficiency of key factor retention and the simplicity after dimensionality reduction.

[0011] Optionally, step S1 specifically includes:

[0012] Step S11: Obtain the power operation data and sensor data of the power distribution cabinet, and perform data preprocessing on the power operation data and sensor data of the power distribution cabinet to obtain the power operation data and sensor data of the power distribution cabinet to be analyzed.

[0013] Step S12: Extract the voltage distribution characteristics and current flow characteristics of the distribution cabinet based on the power operation data to be analyzed;

[0014] Step S13: Perform power system backbone topology analysis based on the voltage distribution characteristics of the distribution cabinet to obtain the power system backbone topology structure;

[0015] Step S14: Use the current flow characteristics of the distribution cabinet to perform global topology completion on the main topology of the power system, thereby obtaining the power system topology;

[0016] Step S15: Perform power flow calculation on the power system topology and construct the dynamic topology of the power system based on the power flow calculation results.

[0017] This invention improves data consistency and integrity by unifying the preprocessing of power operation data from distribution cabinets and sensor data, effectively reducing the impact of data noise and missing values, and ensuring the reliability of power flow analysis and topology reconstruction. Utilizing the voltage distribution characteristics of distribution cabinets, it accurately identifies node voltage levels and trends, helping to identify the main network of the power system, especially assisting in identifying vulnerable nodes with large voltage drops. It completes the topology structure by using current flow characteristics, compensating for the shortcomings of traditional topology identification in branch connection directionality, improving the accuracy of local branch identification, and automatically correcting incorrect wiring by setting a current vector angle change threshold (e.g., 15°), enhancing system fault tolerance. In power flow calculation, it reasonably sets parameters such as node load upper and lower limits (e.g., ±20%) to adapt to different load fluctuations, improving the convergence and real-time performance of power flow solutions. Mapping the power flow results to the topology structure forms a dynamic topology model, which not only tracks the operating status in real time but also provides accurate references for weak current control and regulation strategies.

[0018] Optionally, step S13 specifically includes:

[0019] Step S131: Classify the voltage levels according to the voltage distribution characteristics of the distribution cabinet, set the voltage threshold range to [400V, 220V] to partition the bus nodes, and construct the bus layer structure to obtain the bus level data.

[0020] Step S132: Calculate the potential difference between different bus levels based on the bus level data, select lines with potential difference constraints less than or equal to [5V, 20V] as main branches, perform topology consistency verification, and generate initial main branch data;

[0021] Step S133: Construct a preliminary backbone topology graph using the bus nodes in the bus hierarchy data as vertices and the initial backbone branch data as edges, and perform connectivity analysis to remove isolated nodes and generate a preliminary backbone topology structure.

[0022] Step S134: Based on the voltage distribution characteristics and current flow characteristics of the distribution cabinet, perform power flow direction analysis, and adjust the branch weights of the preliminary trunk topology according to the power flow direction to obtain the optimized trunk topology.

[0023] Step S135: Perform topology mapping on the optimized backbone topology to obtain the power system backbone topology.

[0024] This invention sets [400V, 220V] as the voltage threshold range in voltage level classification, which helps distinguish between main buses and terminal branches, improving the accuracy of bus layer structure identification. Setting the potential difference constraint to [5V, 20V] in main branch screening effectively eliminates non-physical connections, improves the reliability of topology boundary determination, and adapts to situations where voltage drops between buses in low-voltage areas are small but logical connections still exist. Topology consistency verification and isolated node removal prevent redundant connections from interfering with the overall structure. Power flow direction analysis combined with voltage and current flow characteristics enables dynamic correction of branch directions, improving the physical consistency of the topology diagram. The final main topology mapping result provides a clear network skeleton for low-voltage control, supporting node regulation and anomaly diagnosis.

[0025] Optionally, step S14 specifically includes:

[0026] Step S141: Perform weighted directed graph modeling on the current flow characteristics of the distribution cabinet, set the current directionality threshold to [0.3A, 3.0A], and thus construct the current flow correlation graph;

[0027] Step S142: Perform node alignment and edge mapping analysis on the current flow correlation diagram and the power system backbone topology to identify the missing connection parts in the power system backbone topology and obtain the node edge data to be completed in the backbone structure.

[0028] Step S143: Perform graph structure reasoning on the edge data of the nodes to be completed in the backbone structure, perform missing connection completion, and obtain global topology prediction data;

[0029] Step S144: Perform edge weight fusion and unified node encoding on the global topology prediction data and the power system backbone topology to obtain the preliminary power system topology.

[0030] Step S145: Perform consistency verification and closed-loop verification on the preliminary power system topology to obtain the power system topology.

[0031] This invention achieves accurate expansion of the backbone topology to the global topology by introducing weighted directed graph modeling and graph structure reasoning techniques. Setting the current directionality threshold to [0.3A, 3.0A] effectively filters background current and invalid current fluctuations, retaining only lines with physical flow significance, which is beneficial for identifying branches with weak signals but topological significance in low-voltage areas. Node alignment and edge mapping analysis can accurately identify topological voids caused by missing sensors or modeling errors, improving topology integrity. Graph structure reasoning combined with edge prediction mechanisms can fill in missing connections based on the known structure, strengthening topology recovery capabilities. Edge weight fusion and unified encoding processing enhance the consistent expression of topology data from different sources, facilitating rapid node location and scheduling paths in subsequent low-voltage control. Finally, consistency verification and closed-loop verification ensure the logical closed-loop nature and operational rationality of the topology.

[0032] Optionally, step S2 specifically includes:

[0033] Step S21: Perform spatial mapping and time window segmentation on the sensor data of the distribution cabinet to be analyzed, and bind the time window sensor-topology node according to the node-branch relationship in the dynamic topology of the power system to obtain the spatiotemporal feature map of the power system node.

[0034] Step S22: Extract weighted aggregated energy efficiency indicators based on the spatiotemporal feature map of power system nodes to obtain node energy efficiency indicator data;

[0035] Step S23: Construct an energy efficiency attribute map using node energy efficiency index data, and extract node energy efficiency behavior features based on the energy efficiency attribute map;

[0036] Step S24: Perform energy efficiency behavior clustering based on node energy efficiency behavior characteristics to obtain power system energy efficiency pattern data;

[0037] Step S25: Perform anomaly factor detection and model deviation analysis on the power system energy efficiency model data, extract potential load anomaly candidate nodes, and score the confidence of the potential load anomaly candidate nodes to obtain power system load anomaly node data.

[0038] This invention effectively achieves dynamic binding between distribution cabinet sensor data and topology by constructing a spatiotemporal feature map of power system nodes, giving the sensor data structural semantics and enhancing spatiotemporal perception capabilities. Introducing node-branch relationships in the spatial mapping and time window processing stages provides energy efficiency analysis with a basis in low-voltage structure, improving analysis accuracy. Aggregated energy efficiency index extraction unifies multi-source indicators such as voltage, current, and active power into a single entity, enhancing data consistency and facilitating feature modeling. Behavioral feature extraction based on the energy efficiency attribute map significantly improves the ability to characterize node operating states, especially suitable for identifying nodes with small but frequent load fluctuations in low-voltage areas. The energy efficiency clustering process strengthens the analysis of behavioral differences between nodes, improving sensitivity to structural deviations. Anomaly factor detection integrates deviation analysis and confidence scoring mechanisms, filtering out occasional fluctuation interference and effectively pinpointing the source nodes of load anomalies. The overall process considers both low-voltage operating characteristics and energy efficiency pattern evolution, making it suitable for anomaly monitoring and refined operation and maintenance in high-density sensing scenarios.

[0039] Optionally, step S25 specifically includes:

[0040] Step S251: Based on the feature vectors of each cluster center in the power system energy efficiency model data, construct the energy efficiency behavior residual matrix of each node, and calculate the Euclidean distance between the node energy efficiency behavior features and the corresponding node energy efficiency behavior residual matrix to obtain the node energy efficiency behavior residual data.

[0041] Step S252: Based on the residual data of node energy efficiency behavior, perform slope fitting and mutation detection, calculate the abnormal factor score, and obtain the abnormal factor score data; set the dynamic threshold range of the abnormal score according to the abnormal factor score data;

[0042] Step S253: Evaluate the energy efficiency behavior coordination degree of adjacent nodes in the dynamic topology of the power system to obtain the energy efficiency behavior coordination degree of adjacent nodes;

[0043] Step S254: Combine the dynamic threshold range of anomaly scores and the synergy of energy efficiency behavior of adjacent nodes to construct a weighted anomaly distribution map, identify and extract nodes with high anomaly factor scores, and obtain potential load anomaly candidate node data.

[0044] Step S255: Evaluate the node confidence of the potential load anomaly candidate node data, and set the scoring confidence threshold to [0.6, 0.9] to screen high-confidence nodes, thereby obtaining the power system load anomaly node data.

[0045] This invention, by calculating the Euclidean distance between node energy efficiency behavior characteristics and the behavior residual matrix, can accurately measure the difference between node energy efficiency performance and expected behavior, effectively identifying nodes with energy efficiency deviations. It is particularly suitable for small fluctuations caused by load changes in power systems, enhancing sensitivity to abnormal behavior. Slope fitting and abrupt change detection techniques provide accurate criteria for scoring anomaly factors, ensuring rapid response to sudden changes. Simultaneously, setting dynamic threshold ranges allows the system to adapt to different load and environmental changes, improving system robustness. Through the evaluation of energy efficiency behavior synergy, potential behavioral correlations between adjacent nodes can be further captured, optimizing the accuracy of anomaly detection. Combining the construction of a weighted anomaly distribution map with node synergy and anomaly factor scoring, nodes with high anomaly factors are accurately extracted, avoiding false positives. By setting a scoring confidence threshold, the reliability of the selected load anomaly nodes is ensured.

[0046] Optionally, the analysis of the dominant factors of load anomalies in step S3 specifically includes:

[0047] The path backtracking depth is set to [3,5] layers to dynamically backtrack the upstream and downstream paths of the load anomaly node data in the power system, construct the load anomaly path graph, and extract the structural position of each anomaly node in the dynamic topology of the power system and the correlation degree with adjacent nodes to obtain the load anomaly topology path feature data.

[0048] Based on the topology path feature data of load anomalies, and combined with the node energy efficiency index data, a topology-energy efficiency composite feature vector matrix is ​​constructed.

[0049] The structural influence factor embedding training is performed on the topology-energy efficiency composite feature vector matrix to generate load anomaly structural influence feature data.

[0050] Historical load fluctuation sequences and multidimensional sensor data of load fluctuations of distribution cabinets are obtained, and feature factor correlation analysis is performed in combination with power system load anomaly node data to obtain load anomaly physical factor weight data.

[0051] By integrating the structural influence feature data of load anomalies and the weight data of physical factors of load anomalies, multidimensional factor compression and reduction are performed, and factor combinations with a cumulative contribution rate threshold greater than or equal to 0.85 are retained to obtain the load anomaly influence factor spectrum.

[0052] Based on the load anomaly impact factor spectrum, significance ranking and dominant factor identification are performed. The factor contribution rate threshold is set to 0.15 to extract the dominant factors of the abnormal nodes and generate the load anomaly dominant factors.

[0053] This invention, by setting the path backtracking depth to [3,5] layers, dynamically traces the upstream and downstream paths of load anomaly nodes in a power system. This accurately captures the upstream and downstream relationships of load anomaly nodes within the system and their impact on the global power system, constructing a load anomaly path diagram that effectively reveals the propagation path and influence of anomaly nodes. Extracting the structural location of nodes and their correlation with adjacent nodes helps identify the key roles of anomaly nodes in the power system. Combining topology and energy efficiency indicators to construct a composite feature vector matrix enhances the system's multidimensional analysis capability of node behavior, providing a more comprehensive basis for anomaly identification. Embedding and training structural influence factors accurately extracts key impact features of load anomalies, improving the accuracy of anomaly prediction. Correlation analysis of feature factors in load fluctuation sequences and sensor data provides the system with more detailed physical-level factor analysis, further strengthening the ability to identify anomaly sources. Multidimensional factor compression reduction and setting factor contribution rate thresholds help remove redundant information, highlight key factors, and improve the system's response speed and accuracy to anomaly patterns.

[0054] Optionally, the three-phase current balance assessment of the distribution cabinet in step S4 specifically includes:

[0055] Extract the three-phase current time series data of the load abnormality affected node from the sensor data of the distribution cabinet, perform periodic segmentation, missing compensation and distortion point cleaning on the three-phase current time series data, and obtain calibrated three-phase current standardized time series data.

[0056] Based on the calibrated three-phase current standardized time-series data, a three-phase current time-domain waveform tensor model is constructed, and the three-phase current phase offset rate, phase sequence inconsistency rate and time-domain imbalance fluctuation rate are extracted using the three-phase current time-domain waveform tensor model to obtain the three-phase current imbalance characteristic data.

[0057] Based on the three-phase current imbalance characteristic data and the dynamic topology of the power system, a node-level three-phase current imbalance distribution map is constructed; the current offset tension of each abnormal node is calculated using the node-level three-phase current imbalance distribution map to obtain the current tension propagation weight matrix.

[0058] By fusing the current tension propagation weight matrix and the three-phase current imbalance characteristic data, the three-phase current balance index is calculated to obtain the current balance data of abnormal nodes.

[0059] This invention extracts three-phase current time-series data of nodes affected by load anomalies and performs periodic segmentation, missing data compensation, and distortion point cleaning to ensure data accuracy and continuity, avoiding interference from missing data or noise in subsequent analysis. The calibrated three-phase current standardized time-series data effectively eliminates measurement differences between different nodes, resulting in highly reliable analysis results. Imbalance feature data extracted based on the three-phase current time-domain waveform tensor model accurately reflects current phase shifts, phase sequence inconsistencies, and time-domain imbalance fluctuations, providing a precise basis for subsequent anomaly analysis. By constructing a node-level three-phase current imbalance distribution map and calculating current offset tension, the degree of current imbalance at each node can be clearly identified, aiding in the diagnosis of potential current imbalance problems in power systems, especially in weak current control areas, enabling early warning of risks caused by current imbalance.

[0060] Optionally, the abnormal node classification in step S4 specifically includes:

[0061] The current balance data of abnormal nodes is divided into balance intervals to obtain graded current balance label data.

[0062] Based on the dynamic topology of the power system, the topological adjacency information and branch connection relationship of nodes affected by load anomalies are extracted, and the inter-node connection matrix is ​​constructed.

[0063] By combining the node phase connection matrix and graded current balance label data, the three-phase structure distribution of each node affected by load anomalies is analyzed to obtain three-phase phase sequence consistency characteristic data.

[0064] Topological structure projection and constraint rule matching are performed on the three-phase phase sequence consistency characteristic data, and wiring pattern identification is performed to obtain improper wiring identification data;

[0065] By integrating graded current balance label data and improper wiring identification data for joint logic discrimination, the nodes affected by load anomalies are divided into three-phase load offset nodes and three-phase improper wiring nodes.

[0066] This invention ensures quantitative analysis and accurate classification of current balance by dividing the current balance data of abnormal nodes into balance intervals. This helps identify potential current imbalance problems in power systems, especially in low-voltage control areas. Through the extraction of hierarchical label data, fine-grained management of different current balance levels can be achieved. Based on the dynamic topology of the power system, the topological adjacency information and branch connection relationships of nodes are extracted, accurately depicting the electrical connection relationships of each node in the power system and providing a foundation for further analysis. By combining the node phase-to-phase connection matrix with current balance label data, the three-phase structure distribution of load-abnormal nodes can be comprehensively analyzed, helping to identify three-phase phase sequence consistency issues and improper wiring in the power system, avoiding power system instability caused by wiring errors. Through joint logic discrimination, nodes affected by load anomalies are divided into three-phase load offset nodes and three-phase improper wiring nodes, which helps to accurately locate the problem source, respond quickly, and take effective measures.

[0067] Optionally, this specification also provides a power system operation and maintenance system for a low-voltage control intelligent distribution cabinet, used to execute the power system operation and maintenance method for the low-voltage control intelligent distribution cabinet as described above. The power system operation and maintenance system for the low-voltage control intelligent distribution cabinet includes:

[0068] The topology analysis module is used to acquire power operation data and sensor data of the distribution cabinet, and to perform power system topology analysis based on the power operation data of the distribution cabinet to obtain the dynamic topology of the power system.

[0069] The potential load anomaly analysis module is used to identify the power system energy efficiency pattern from the distribution cabinet sensor data through the dynamic topology of the power system, and obtain the power system energy efficiency pattern data; and to perform potential load anomaly analysis on the power system energy efficiency pattern data, and obtain the power system load anomaly node data.

[0070] The abnormal impact structure division module is used to analyze the load abnormality dominant factors based on the load abnormality node data of the power system, and divide the load abnormality impact structure of the dynamic topology of the power system according to the load abnormality dominant factors to obtain the load abnormality impact nodes.

[0071] The three-phase current balance assessment module is used to assess the three-phase current balance of the distribution cabinet based on the sensor data of the distribution cabinet to identify nodes affected by abnormal loads, and to classify abnormal nodes based on the current balance of the distribution cabinet to obtain three-phase load offset nodes and improper three-phase wiring nodes.

[0072] The abnormal node power redistribution module is used to reconstruct the power distribution of three-phase load offset nodes and improperly connected three-phase nodes, obtain abnormal node power distribution data, and transmit it to the distribution cabinet management platform to execute power distribution tasks.

[0073] The present invention relates to a power system operation and maintenance system for a smart distribution cabinet for low-voltage control. This system can implement any of the power system operation and maintenance methods of the smart distribution cabinet for low-voltage control according to the present invention. It is used to connect the operation and signal transmission media between various modules to complete the power system operation and maintenance method of the smart distribution cabinet for low-voltage control. The internal modules of the system cooperate with each other, thereby improving the low-voltage control efficiency and overall energy efficiency of the distribution cabinet. Attached Figure Description

[0074] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0075] Figure 1 This is a flowchart illustrating the steps of the power system operation and maintenance method for the intelligent distribution cabinet for low-voltage control according to the present invention.

[0076] Figure 2 This is a detailed flowchart of step S1 in the present invention;

[0077] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0078] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0079] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0080] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides a power system operation and maintenance method for an intelligent distribution cabinet for low-voltage control, the method comprising the following steps:

[0081] Step S1: Obtain the power operation data and sensor data of the distribution cabinet, and perform power system topology analysis based on the power operation data of the distribution cabinet to obtain the dynamic topology of the power system;

[0082] In this embodiment, a three-phase power metering module (such as a DTSD1352 meter supporting the Modbus RTU protocol) and multiple types of sensor nodes (including current, voltage, temperature, humidity, etc.) installed inside the distribution cabinet collect real-time data on the electrical operating status of the distribution cabinet and the sensor data of the equipment every 5 seconds. The data is aggregated and processed by an edge computing unit (such as a Raspberry Pi CM4 equipped with an EMQX lightweight MQTT server), and outlier removal, time alignment, and multi-source synchronization are completed by a data preprocessing module. Subsequently, a preliminary connection graph is constructed based on the current data, and branch edges are filtered using a set current direction threshold q = 1.2A. On the basis of meeting the connection connectivity conditions, a graph neural network (GraphSAGE) combined with the voltage distribution hierarchy is used to complete the hierarchical reasoning of topology nodes. The topology structure is optimized through three rounds of iteration, and finally, a dynamic topology structure graph of the power system containing buses, branches, and load terminals (encoded using an adjacency matrix method) is output as the structural basis input for subsequent steps.

[0083] Step S2: Identify the power system energy efficiency pattern by analyzing the power distribution cabinet sensor data through the power system dynamic topology to obtain power system energy efficiency pattern data; perform potential load anomaly analysis on the power system energy efficiency pattern data to obtain power system load anomaly node data.

[0084] In this embodiment, the constructed dynamic topology of the power system is used as a structural constraint. Based on the time window sliding sequence (with a window width of 15 minutes and a step size of 5 minutes) in the distribution cabinet sensor data, the energy efficiency indicators (unit power consumption, current fluctuation rate, phase sequence stability, etc.) of each node are structurally aligned in time and space. Then, an energy efficiency feature vector is constructed for each topology node, and a pattern recognition algorithm based on spectral clustering is used to classify the nodes into typical energy efficiency pattern clusters (e.g., "high load, low fluctuation," "periodic oscillation type," etc.). For each pattern, a center vector is constructed, and the residual score between each node and the center of its cluster is calculated. If the node's residual score exceeds 0.75 (normalized score), it is marked as a potential candidate node for load anomalies. A second confirmation is then performed by combining the node's load fluctuation curve slope change threshold s≥20% over the past 24 hours. Finally, load anomaly nodes that meet both structural deviation and behavioral fluctuation conditions are extracted, providing target objects for the next stage of anomaly factor analysis.

[0085] Step S3: Analyze the load anomaly dominant factors based on the load anomaly node data of the power system, and divide the load anomaly impact structure of the dynamic topology of the power system according to the load anomaly dominant factors to obtain the load anomaly affected nodes.

[0086] In this embodiment, path backtracking is performed on the obtained load anomaly nodes in the dynamic topology, with a maximum backtracking depth of 3 hops. The Dijkstra algorithm is used to extract all upstream and downstream associated nodes and branches, generating a topology path graph. Subsequently, the path graph nodes and their energy efficiency index vectors are concatenated to form a topology-energy efficiency composite feature vector matrix. This matrix is ​​trained using a Graph Convolutional Network (GCN) with 200 training epochs and a learning rate of 0.001, extracting the anomaly propagation structure influence factor vector. Simultaneously, historical load data and multi-dimensional sensor data (such as temperature rise and current distortion rate) from the distribution cabinet over the past 48 hours are retrieved. The correlation between these physical indicators and anomaly labels is calculated using the maximum information coefficient (MIC), with a significance threshold of 0.05. Highly correlated physical factors are selected and weighted according to their influence level, generating a load anomaly physical factor weight vector. Finally, the structural influence factor and physical factor are fused, PCA compression is performed and more than 85% of the cumulative variance is retained, an anomaly factor spectrum is output, and the contribution rate threshold θ = 0.15 is set to extract the dominant anomaly factor of each anomaly node.

[0087] Step S4: Based on the sensor data of the distribution cabinet, evaluate the three-phase current balance of the distribution cabinet for nodes affected by abnormal loads, and classify abnormal nodes according to the current balance of the distribution cabinet to obtain the three-phase load offset nodes and the improper three-phase wiring nodes.

[0088] In this embodiment, the three-phase current data of the target node affected by load anomalies within the most recent 30 minutes are extracted from the sensor data of the distribution cabinet. The sampling period is 2 seconds, the sliding window length is 60 seconds, and period segmentation and missing data compensation (using linear interpolation) are performed. Then, wavelet denoising is used to remove distortion points, and a three-phase time-series waveform tensor (dimension 60×3) is constructed. In the feature extraction stage, the offset rate between the three-phase average currents, the phase sequence inconsistency rate (based on time delay cross-correlation analysis), and the imbalance fluctuation rate represented by the coefficient of variation are calculated respectively, and the three are combined into a current imbalance feature vector. Based on the dynamic topology of the power system, the adjacent propagation intensity of each node is calculated. The influence diffusion of the node is simulated by the current tension propagation weight matrix (normalized tension index λ = 0.6), and the current balance index CBI is calculated based on the propagation results (formula CBI = 1 - ||propagation tensor|| / θ, θ is taken as 1.0). Finally, the balance level is classified according to the CBI value, where CBI < 0.7 is a severely unbalanced node. Combining phase-to-phase connection methods with standard phase sequence rules, a wiring consistency matching analysis is performed. If phase sequence inconsistencies or line bridging are found, they are marked as wiring anomalies. Finally, based on the rule set: only imbalance → load offset; wiring anomaly present → wiring anomaly; both present → wiring anomaly predominant, the abnormal node type is classified.

[0089] Step S5: Reconstruct the power distribution for the three-phase load offset nodes and improperly connected three-phase nodes to obtain the power distribution data for the abnormal nodes, and transmit it to the distribution cabinet management platform to execute the power distribution task.

[0090] In this embodiment, the identified "three-phase load offset nodes" and "three-phase improper wiring nodes" are stored in the database, and a binding relationship is established between them and their topological location and branch load data. For offset nodes, a current redistribution model based on linear programming is adopted, setting the current adjustment target of the branch where the node is located as a three-phase balance target (error ≤ ±5%), and using the adjacent spare branch access scheme as an adjustment variable, setting the maximum phase offset correction range to 10A. For improper wiring nodes, the construction wiring drawings and phase sequence identification records are retrieved, and a wiring replacement suggestion list (such as AC swapping) is automatically generated and pushed to the platform operation interface. The platform sends the task data to the field intelligent circuit breaker control unit based on the SCADA system interface (such as IEC 61850 or DL / T645), triggering the current limiting switching or phase sequence reconnection operation process. The entire process adopts a control command execution feedback mechanism, setting the response timeout time to 3 seconds to ensure closed-loop execution of commands and implementation of the allocation strategy. Finally, allocation execution status data is generated through event logging for subsequent maintenance and verification.

[0091] Optionally, step S1 specifically includes:

[0092] Step S11: Obtain the power operation data and sensor data of the power distribution cabinet, and perform data preprocessing on the power operation data and sensor data of the power distribution cabinet to obtain the power operation data and sensor data of the power distribution cabinet to be analyzed.

[0093] In this embodiment, power operation data and sensor data of the distribution cabinet are acquired by a three-phase power metering module (such as a DTSD1352 meter supporting the Modbus RTU protocol) and multiple types of sensor nodes (including current, voltage, temperature, humidity, etc.) installed inside the distribution cabinet. Power operation data is uploaded to an edge processing terminal (EdgeBox-RK3399) via an RS-485 bus, with a sampling frequency of 1Hz and a data window length of 60 seconds. Environmental sensors (temperature, humidity, door magnetic sensors, smoke detectors) have a sampling period of 5 seconds. After all data enters the data receiving module, timestamp alignment is performed first, and the maximum time offset correction threshold is set to ±1 second. Then, the voltage and current values ​​are normalized using the Z-score normalization method, and data points exceeding the set abnormal range (such as voltage deviation > ±15%) are removed. Missing segments are completed using cubic spline interpolation, and a minimum effective sampling ratio threshold of 90% is set; sample data below this ratio are not used. The final output format is uniformly structured JSON, with fields including timestamp, phase, voltage value, current value, sampling quality, etc., which serve as data input for subsequent analysis steps.

[0094] Step S12: Extract the voltage distribution characteristics and current flow characteristics of the distribution cabinet based on the power operation data to be analyzed;

[0095] In this embodiment, three-phase voltage time series and three-phase current time series are constructed for the processed power operation data to be analyzed. To extract the voltage distribution characteristics of the distribution cabinet, a 5-minute sliding window period is used to calculate the average voltage value and standard deviation of each phase within each window, which is used as part of the node voltage feature vector. At the same time, the maximum offset ΔV between the voltages of each phase is calculated. If ΔV>20V, it is marked as a voltage imbalance node. For the current flow characteristics, the flow direction is determined by the positive and negative directions of the current phase angle. A directional threshold q=1.5A is set, that is, when the current difference between a node and its adjacent node is greater than the threshold and the directions are consistent, a directed edge is constructed. The connection relationships between all nodes that meet the conditions are used to form an initial flow direction matrix, which is encoded in a weighted directed graph manner (the edge weight is the current amplitude) for subsequent topology reasoning analysis.

[0096] Step S13: Perform power system backbone topology analysis based on the voltage distribution characteristics of the distribution cabinet to obtain the power system backbone topology structure;

[0097] In this embodiment, based on the obtained voltage distribution characteristics, bus hierarchy determination and trunk path extraction are performed at this stage. First, a voltage hierarchy threshold range [400V, 220V] is set, and nodes with voltage values ​​within this range are selected as trunk candidate nodes. Then, a bus-load connectivity graph is constructed, and a depth-first search (DFS) algorithm is used to identify the connectivity of all nodes in the graph, retaining only the largest connected subgraph portion that meets the trunk condition. This subgraph is then topologically sorted, the relative potential difference between nodes is calculated, and node groups with potential differences less than 15V are selected as trunk clusters. Finally, the set of bus nodes in the trunk structure and their corresponding branch connections are output, forming a preliminary power system trunk topology structure to assist in the next step of flow direction reasoning.

[0098] Step S14: Use the current flow characteristics of the distribution cabinet to perform global topology completion on the main topology of the power system, thereby obtaining the power system topology;

[0099] In this embodiment, based on the backbone topology, a current flow weighted graph is used for global topology completion analysis. First, all backbone nodes and edges are extracted, and node alignment and edge mapping are performed with the current flow graph. If an edge with significant current transmission but missing connections in the backbone structure is found (current amplitude > 1.5A, voltage difference < 5V), it is marked as an edge to be completed. A graph structure inference model (such as a GAT-based edge prediction network) is used to score the existence of these candidate edges, setting a prediction confidence threshold of 0.7. Edges exceeding the threshold are included in the topology. Newly added edges are uniformly encoded, and node connectivity and edge weights are recalculated. The updated global topology is jointly expressed using an adjacency matrix and an edge weight matrix. To ensure structural rationality, a topology loop closure verification is performed to ensure that no newly added branches introduce redundant loops or isolated nodes, ultimately forming a complete and interconnected power system topology.

[0100] Step S15: Perform power flow calculation on the power system topology and construct the dynamic topology of the power system based on the power flow calculation results.

[0101] In this embodiment, after constructing the power system topology, the Newton-Raphson method is used to perform power flow calculations on the entire topology. The specific process includes setting the initial bus voltage to 400V, the phase angle to 0°, and setting the power flow calculation accuracy error threshold ε = 10⁻³. Power injection, branch current, and voltage sag at each node are solved using power flow equations. During the calculation, a constant impedance model is used for the load model, and line parameters are obtained from sensor measurements (such as branch resistance and reactance). The actual power flow calculation is performed in the MATPOWER (version 7.1) environment. After the calculation is completed, the results are mapped back to the original topology to generate a topology attribute matrix containing real-time active power, reactive power, voltage phase angle, and current amplitude. The load status of each node is classified (light load / heavy load / overload) based on the power flow results. Simultaneously, voltage flow diagrams and current transmission diagrams are constructed as the final expression of the dynamic topology of the power system, providing fundamental support for subsequent energy efficiency analysis and anomaly detection.

[0102] Optionally, step S13 specifically includes:

[0103] Step S131: Classify the voltage levels according to the voltage distribution characteristics of the distribution cabinet, set the voltage threshold range to [400V, 220V] to partition the bus nodes, and construct the bus layer structure to obtain the bus level data.

[0104] In this embodiment, the average three-phase voltage extracted from the power operation data of the distribution cabinet is used as the basis for the voltage distribution characteristics of the nodes. A fixed voltage threshold range [400V, 220V] is used for bus hierarchy classification. Nodes with voltages greater than or equal to 380V are classified as high-voltage buses, and those with voltages between 220V and 380V are classified as low-voltage buses. The K-means++ clustering algorithm is used to perform two-class clustering of node voltages, with initial center values ​​set at 420V and 230V, and an iteration convergence threshold set at 0.001. The hierarchy to which each node belongs is labeled according to the clustering results, forming a bus hierarchy mapping table. Furthermore, a connection weight matrix is ​​established for adjacent nodes in the same hierarchy to calculate the internal connectivity between bus hierarchies. Simultaneously, the node number, corresponding voltage hierarchy, and hierarchical relationship are recorded to form a structured bus hierarchy data file for subsequent branch selection and topology construction.

[0105] Step S132: Calculate the potential difference between different bus levels based on the bus level data, select lines with potential difference constraints less than or equal to [5V, 20V] as main branches, perform topology consistency verification, and generate initial main branch data;

[0106] In this embodiment, the potential difference between each pair of upper and lower bus nodes is calculated based on the generated bus hierarchy data. An absolute voltage difference model ΔV = |V_upper - V_lower| is used, and the filtering threshold range is set to [5V, 20V], meaning only connections with potential differences within this range are retained as candidate trunk branches. During the calculation, for each pair of reachable inter-layer node paths, if the voltage difference meets the set constraints and the impedance of the line between them is less than 0.5Ω, it is determined to be a valid connection. All candidate branch information (starting point number, ending point number, potential difference, resistance, current directionality) is uniformly stored in the candidate branch dataset. Subsequently, a graph traversal algorithm is used to verify the consistency of the branch set, eliminating invalid paths that form loops or duplicate connections, and retaining the branch combinations with the strongest structural connectivity. Finally, an initial trunk branch data table is output for subsequent trunk graph construction.

[0107] Step S133: Construct a preliminary backbone topology graph using the bus nodes in the bus hierarchy data as vertices and the initial backbone branch data as edges, and perform connectivity analysis to remove isolated nodes and generate a preliminary backbone topology structure.

[0108] In this embodiment, the selected main branches are used as edge information of the graph, and the nodes in the bus-level data are used as vertices to construct a preliminary backbone topology graph. An adjacency matrix is ​​used to represent the graph structure, and connectivity analysis is performed. A breadth-first search (BFS) algorithm is used, starting from the high-voltage bus nodes, to mark all reachable nodes. Nodes not traversed in the connected graph are defined as isolated nodes and removed. The removal criterion is that the sum of the node's out-degree and in-degree is less than 1. The retained backbone portion of the graph undergoes normalized renumbering, with node IDs uniformly converted to structured numbers starting with M (e.g., M001, M002…) to ensure data readability and system accessibility of the topology structure. Finally, a preliminary backbone topology graph and structured tables are generated, with fields including node number, adjacent nodes, connection voltage difference, current amplitude, and other topology element information.

[0109] Step S134: Based on the voltage distribution characteristics and current flow characteristics of the distribution cabinet, perform power flow direction analysis, and adjust the branch weights of the preliminary trunk topology according to the power flow direction to obtain the optimized trunk topology.

[0110] In this embodiment, based on the initial backbone topology, power flow direction analysis is performed on each backbone branch by combining voltage distribution characteristics and current flow direction characteristics. The power direction determination formula P = VIcosθ is used, with direction determined by the positive or negative power flow, and a minimum effective power threshold of 50W is set. If the current direction of two adjacent nodes is consistent, the voltage attenuation is reasonable (not exceeding 10%), and the power flow is positive, then the branch is considered a normal power flow branch. Power direction identifiers are established for all branches, and the directional weight of the branch is calculated based on the power flow direction and amplitude, using the formula W = αP + βI, where α = 0.7 and β = 0.3. The updated weight values ​​are written into the topology edge attributes for branch priority ranking. Finally, an optimized backbone topology diagram is output, achieving dynamic adjustment of branch direction weights while maintaining the node connection logic, facilitating subsequent optimization of path selection in power dispatching or simulation.

[0111] Step S135: Perform topology mapping on the optimized backbone topology to obtain the power system backbone topology.

[0112] In this embodiment, based on the optimized backbone topology, topology mapping is implemented to obtain a standardized power system backbone topology. First, each node in the optimized structure is physically mapped according to its actual physical location and device number, establishing a one-to-one correspondence table between logical nodes and device addresses. GIS coordinates are used to constrain node positions, with a coordinate error tolerance threshold of ±2 meters. Weighted shortest path correction is applied to nodes with positional offsets. Next, a topology closed-loop consistency check is performed, including three verification standards: branch direction rationality, current closure check, and load power supply continuity. If there are breakpoints, floating loads, or abnormal loops in the topology, it will automatically roll back to the previous structure version and issue an alarm. The final output power system backbone topology includes a node topology table, a standardized connectivity graph, a GIS spatial location map, and a topology attribute configuration file, formatted uniformly in both XML and JSON formats. It supports importing into mainstream power system modeling and simulation platforms such as DIgSILENT or CYME for system-level deployment and testing.

[0113] Optionally, step S14 specifically includes:

[0114] Step S141: Perform weighted directed graph modeling on the current flow characteristics of the distribution cabinet, set the current directionality threshold to [0.3A, 3.0A], and thus construct the current flow correlation graph;

[0115] In this embodiment, the current directionality characteristics of each line in the distribution cabinet sensor data are extracted, and a weighted directed graph model is constructed using the current flow direction identifier (forward / reverse) and amplitude information. To ensure the validity of the modeling data, the current directionality threshold range is set to [0.3A, 3.0A], meaning only lines within this range are included in the modeling. For all node pairs that meet the conditions, a current flow association graph G(I) = (V, E) is constructed, using the node number as the graph vertex, the current directionality as the directed edge direction, and the current amplitude as the edge weight. The Python NetworkX library is used for graph modeling and weight configuration, and the edge weights are standardized to the interval [0, 1]. The graph structure does not contain isolated nodes, and all nodes must have at least one incoming or outgoing edge to ensure connectivity. The final output is a current flow association graph file, including fields such as node ID, current direction, and edge weight, and supports importing into a graph inference engine for structural completion analysis.

[0116] Step S142: Perform node alignment and edge mapping analysis on the current flow correlation diagram and the power system backbone topology to identify the missing connection parts in the power system backbone topology and obtain the node edge data to be completed in the backbone structure.

[0117] In this embodiment, the constructed current flow correlation graph and the generated power system backbone topology are read, and node alignment and edge mapping analysis are performed. A dual comparison is conducted between node numbers and physical locations (GIS coordinates), allowing a maximum numbering error of ±2 digits and a maximum coordinate deviation of ±1.5 meters to achieve precise node alignment in the graph. After alignment, an edge overlap analysis strategy is used to detect connections in the backbone topology that differ from the current graph, specifically identifying connections between nodes missing in the backbone topology but present in the current flow graph. The criteria for judging the difference of each edge include whether there is a connection between nodes, whether the connection direction is consistent, and whether the edge weight difference exceeds 20%. Finally, a table of node edge data to be completed in the backbone structure is output, with fields including: starting node, ending node, current direction, consistency score, etc., for subsequent graph reasoning and completion.

[0118] Step S143: Perform graph structure reasoning on the edge data of the nodes to be completed in the backbone structure, perform missing connection completion, and obtain global topology prediction data;

[0119] In this embodiment, a Graph Neural Network (GNN) is used for graph structure inference based on the identified backbone structure node edge data to be completed. The GraphSAGE model is used for training, with input data including node features (such as voltage, current values, and node degree) and edge features (such as current directionality and current difference). The training set consists of the known backbone structure portion, and the edges to be completed serve as the test set for edge existence prediction. The sampling batch is set to 128 edges, the learning rate to 0.001, and the training epochs to 500, using the Adam optimizer. The edge prediction output is the edge existence probability. A completion threshold of 0.7 is set, meaning edges with a prediction probability greater than 0.7 will be included in the structure as completed edges. The prediction output is global topology prediction data, including the start and end points, edge probabilities, and prediction sources of newly added connecting edges, in both JSON and CSV formats for structure fusion.

[0120] Step S144: Perform edge weight fusion and unified node encoding on the global topology prediction data and the power system backbone topology to obtain the preliminary power system topology.

[0121] In this embodiment, global topology prediction data is read and fused with the existing power system backbone topology. First, edges in both topologies are merged, and nodes are recoded using a unified node numbering rule (prefix "M" + 6-digit number). Simultaneously, the weights of newly added edges are weighted and averaged with the weights of the original backbone edges. The fusion weight calculation method is: W_new = αW_existing + (1-α)W_predicted, where α = 0.6. If a newly added edge conflicts with an existing edge (inconsistent direction or duplicate node ID), the predicted edge takes precedence, and weight updates and edge redirection are performed. Finally, a preliminary power system topology diagram is formed, including complete node connections, a weight matrix, and a unified encoding table. The output supports DOT and GEXF formats for subsequent consistency verification.

[0122] Step S145: Perform consistency verification and closed-loop verification on the preliminary power system topology to obtain the power system topology.

[0123] In this embodiment, the generated power system topology undergoes consistency verification, primarily including three types of verification: 1) node connectivity verification, ensuring no isolated nodes or floating branches exist in the topology; 2) edge direction consistency verification, determining whether the power flow direction matches the current direction, with a maximum allowable deviation angle of ±15°; and 3) loop closure verification, detecting the existence of illegal passive closed loops. A custom topology verification engine executes the above logical rules. If an abnormal connection is detected, it is marked as a red edge and an error report is output. After all verifications pass, a closed-loop verification simulation is performed, using a power flow calculation model (based on the Newton-Raphson method) to test voltage and current distribution, with error standards set at ±0.01V and ±0.05A. After all verifications pass, the final power system topology file is output, containing topology graph structure, edge weights, power flow verification results, and other data. The file is output in XML and graph database formats for system deployment.

[0124] Optionally, step S2 specifically includes:

[0125] Step S21: Perform spatial mapping and time window segmentation on the sensor data of the distribution cabinet to be analyzed, and bind the time window sensor-topology node according to the node-branch relationship in the dynamic topology of the power system to obtain the spatiotemporal feature map of the power system node.

[0126] In this embodiment, spatial mapping is performed on the sensor data of the distribution cabinet to be analyzed. Based on the physical connection relationship between each node and branch in the dynamic topology of the power system, a one-to-one mapping table between sensors and nodes is constructed. The mapping relationship must meet the constraints that the spatial distance between sensors is less than 1.5 meters and the signal belongs to a unique entity. Subsequently, the sensor data is divided into sliding time windows in 15-minute units, using a fixed step-size window segmentation strategy with a step size of 5 minutes to enhance the robustness of overlapping time-series samples. The data in each time window is bound to its corresponding topology node. The binding method uses a key-value pair structure for storage (the key is the node ID, and the value is the sensor vector in the corresponding time window), thereby constructing a spatiotemporally unified node data structure. Finally, it is integrated into a spatiotemporal feature map of power system nodes, and the output is a multi-dimensional time series tensor with dimensions of number of nodes × number of time windows × feature dimensions (such as voltage, current, active power, and reactive power, a total of 4 channels).

[0127] Step S22: Extract weighted aggregated energy efficiency indicators based on the spatiotemporal feature map of power system nodes to obtain node energy efficiency indicator data;

[0128] In this embodiment, based on the constructed spatiotemporal feature map of nodes, multi-index weighted aggregate energy efficiency calculation is performed. Specifically, basic indicators such as voltage, current, active power, and reactive power of each node are extracted across all time windows, and a time-weighted attenuation coefficient (set to γ ​​= 0.85) is introduced to assign higher weights to more recent time windows. The formula η = ∑(γ^t × x_t), where x_t is the indicator value at time t, is used to calculate node-level weighted energy efficiency vectors for the four types of indicators, with a vector dimension of 4. To further normalize the differences in the dimensions of different indicators, a Z-score standardization strategy is used to normalize all node indicators, with the mean and standard deviation based on the distribution statistics of all nodes. The final output is a node energy efficiency indicator data table, with fields including node ID, voltage factor, active power factor, reactive power factor, and current factor, which serves as input for the next step of behavioral analysis.

[0129] Step S23: Construct an energy efficiency attribute map using node energy efficiency index data, and extract node energy efficiency behavior features based on the energy efficiency attribute map;

[0130] In this embodiment, an energy efficiency attribute graph is constructed using a graph structure modeling approach based on the obtained node energy efficiency index data. The graph vertices are node IDs, and the edges are derived from the adjacent connections in the dynamic topology. The edge weights are defined according to the Euclidean distance between two nodes, with the formula w(i,j) = 1 / (1+d(i,j)), where d(i,j) is the Euclidean distance between nodes i and j in the energy efficiency index vector space. Using this graph structure, a node embedding method (such as Node2Vec, with a walk step size of 10, 80 walks, and an embedding dimension of 16) is further applied to generate an embedded representation of the energy efficiency behavior features of each node. This feature vector reflects the node's hybrid representation ability in terms of topological association and energy efficiency features, ultimately forming a node energy efficiency behavior feature set for cluster analysis and subsequent anomaly identification.

[0131] Step S24: Perform energy efficiency behavior clustering based on node energy efficiency behavior characteristics to obtain power system energy efficiency pattern data;

[0132] In this embodiment, unsupervised clustering analysis is performed using the generated node energy efficiency behavior feature vectors. The K-means algorithm is selected for clustering, with an initial cluster size K = 5. The number of clusters is dynamically adjusted through silhouette coefficient analysis, and the maximum silhouette coefficient (greater than 0.72) is obtained when the number of clusters stabilizes at K = 4. The clustering feature space is iteratively calculated based on Euclidean distance, with a maximum number of iterations set to 300 and a convergence threshold ε of 0.001. After clustering, each cluster represents a stable energy efficiency pattern, containing a set of topological nodes with similar behavior. The output is a power system energy efficiency pattern data table, containing information such as node ID, cluster ID, energy efficiency center vector, and time period, for use in the next step of anomaly detection.

[0133] Step S25: Perform anomaly factor detection and model deviation analysis on the power system energy efficiency model data, extract potential load anomaly candidate nodes, and score the confidence of the potential load anomaly candidate nodes to obtain power system load anomaly node data.

[0134] In this embodiment, outlier detection is performed on the energy efficiency model data using the LOF (Local Outlier Factor) method. With a neighbor count of k=20, deviations in the local density of each node are calculated. Nodes with an LOF value higher than 2.0 are initially considered potential outliers. Subsequently, the node offset is calculated for each cluster center. If the offset exceeds twice the standard deviation within its cluster, it is also marked as a pattern-deviation node. The union of these two sets of results is used as a potential load anomaly candidate node. A confidence scoring mechanism is further introduced, with scoring items including the degree of local density deviation, offset distance, and anomaly frequency within the corresponding time window. A weighted comprehensive method is used for scoring, with weights of 0.5, 0.3, and 0.2. A scoring threshold of 0.65 is set; nodes with scores higher than the threshold are marked as final power system load anomaly nodes. The output is an anomaly node data table, with fields including node ID, anomaly score, scoring details, and label type.

[0135] Optionally, step S25 specifically includes:

[0136] Step S251: Based on the feature vectors of each cluster center in the power system energy efficiency model data, construct the energy efficiency behavior residual matrix of each node, and calculate the Euclidean distance between the node energy efficiency behavior features and the corresponding node energy efficiency behavior residual matrix to obtain the node energy efficiency behavior residual data.

[0137] In this embodiment, energy efficiency feature vectors (e.g., voltage, active power, reactive power, and current, totaling four dimensions) of each cluster center in the power system energy efficiency model data are extracted as representative behavior templates. For each node to be analyzed, the difference between its true energy efficiency behavior feature vector and the vector of its respective cluster center is calculated to construct a node energy efficiency behavior residual vector. Further, a residual matrix is ​​constructed in the residual space with nodes as samples and feature dimensions as axes; the matrix shape is the number of nodes × 4. The Euclidean distance formula is used to evaluate the residual magnitude of the node vector and its residual vector. After standardization, the energy efficiency behavior residual data for each node is obtained. This data reflects the degree of node offset outside of clustering behavior, providing a quantitative basis for subsequent slope fitting and abrupt change detection.

[0138] Step S252: Based on the residual data of node energy efficiency behavior, perform slope fitting and mutation detection, calculate the abnormal factor score, and obtain the abnormal factor score data; set the dynamic threshold range of the abnormal score according to the abnormal factor score data;

[0139] In this embodiment, based on the obtained residual data of node energy efficiency behavior, the residual time series of each node is selected for linear slope fitting. The least squares method is used to calculate the residual trend slope of each node, and the sign and magnitude of the slope are used to identify the direction and intensity of energy efficiency shift. For node residual sequences with significant abrupt changes (e.g., slopes greater than 0.25 or less than -0.25), the Pelt algorithm is introduced for abrupt change detection, with a penalty parameter β set to 1.5. The anomaly factor score of each node is obtained by weighted summation of three factors: slope intensity, abrupt change frequency, and residual mean, with weights set to 0.4, 0.4, and 0.2, respectively. Finally, the anomaly factor score data is output. An adaptive dynamic threshold interval for anomaly scores is constructed based on the score distribution, and the upper and lower limit intervals are calculated using the 95th percentile method.

[0140] Step S253: Evaluate the energy efficiency behavior coordination degree of adjacent nodes in the dynamic topology of the power system to obtain the energy efficiency behavior coordination degree of adjacent nodes;

[0141] In this embodiment, based on the dynamic topology of the power system, adjacent node pairs with physical connections are selected, with a count exceeding 800 pairs. For each pair of nodes, its energy efficiency behavior feature vector is extracted, and the cosine similarity between them is calculated as a quantitative indicator of energy efficiency behavior synergy. If the synergy is greater than 0.85, it is considered highly synergistic; if it is less than 0.6, it is considered inconsistent. Considering the time evolution characteristics, the trend of synergistic behavior changes over the past 3 hours is also compared. If the short-term synergy fluctuation is less than 0.05, it is marked as a stable synergistic pair. The output is a synergy matrix of adjacent node pairs, where each item records the node pair ID, current synergy value, and historical synergy fluctuation value, used for subsequent anomaly distribution map construction.

[0142] Step S254: Combine the dynamic threshold range of anomaly scores and the synergy of energy efficiency behavior of adjacent nodes to construct a weighted anomaly distribution map, identify and extract nodes with high anomaly factor scores, and obtain potential load anomaly candidate node data.

[0143] In this embodiment, anomaly factor scores and energy efficiency synergy are fused to construct a weighted anomaly distribution map. The vertices of the graph are topological nodes, the edges represent synergy relationships, the node attributes are anomaly scores, and the edge attributes are synergy weights. During graph construction, the node anomaly score weighting factor is set to 0.7, and the edge synergy factor is set to 0.3. A GAT (Graph Attention) neural network model is used for feature propagation modeling to learn the critical paths of anomaly propagation in the graph. During propagation, clusters of nodes with high scores are identified, densely distributed anomaly regions are extracted, and nodes with anomaly factor scores higher than the upper limit of the dynamic threshold are designated as preliminary candidate nodes. After the final anomaly factor score data is output, a dynamic threshold interval is constructed based on its overall distribution using the quantile method. Specifically, the distribution of all node anomaly score data is statistically analyzed, and a confidence level of 95% is set. The upper threshold of the score distribution is calculated as the 95th percentile (P95), and the lower threshold is calculated as the 5th percentile (P5), forming the dynamic threshold interval. For example, when the P95 value of all scores is 0.87, the upper limit of the dynamic threshold for anomaly factor scores can be set to 0.87. This threshold is then used as the scoring threshold for the initial candidate nodes. The final output is data on potential load anomaly candidate nodes, with fields including node ID, anomaly score, collaborative propagation weight, and neighborhood cluster density.

[0144] Step S255: Evaluate the node confidence of the potential load anomaly candidate node data, and set the scoring confidence threshold to [0.6, 0.9] to screen high-confidence nodes, thereby obtaining the power system load anomaly node data.

[0145] In this embodiment, a confidence score is evaluated for the extracted potential load anomaly candidate nodes. The scoring dimensions include three aspects: (1) the weight of the anomaly factor score ratio is set to 0.5; (2) the weight of the neighborhood collaborative average score is set to 0.3; and (3) the weight of the stability of the energy efficiency cluster to which it belongs is set to 0.2. After the scoring results are normalized to the [0,1] interval, all candidate nodes are sorted, and the confidence threshold interval for the scoring is set to [0.6,0.9]. Nodes with a confidence score higher than the lower limit are selected as the final power system load anomaly nodes. The output data includes node ID, confidence score, score composition ratio, and anomaly status label, providing a basis for the next step of load management or alarm distribution.

[0146] Optionally, the analysis of the dominant factors of load anomalies in step S3 specifically includes:

[0147] The path backtracking depth is set to [3,5] layers to dynamically backtrack the upstream and downstream paths of the load anomaly node data in the power system, construct the load anomaly path graph, and extract the structural position of each anomaly node in the dynamic topology of the power system and the correlation degree with adjacent nodes to obtain the load anomaly topology path feature data.

[0148] In this embodiment, the path backtracking depth is set to [3,5] layers. Based on the directed edge relationships of the dynamic topology of the power system, a backtracking operation is performed on the upstream and downstream nodes of each load anomaly node to identify structural path nodes and branches that are directly or indirectly connected to the anomaly node within the backtracking depth range. The Dijkstra algorithm is used to calculate the distance weights of different paths, and the path directionality is determined by combining the branch current flow direction to construct a load anomaly path graph dominated by the anomaly node. Based on this, the structural level, upstream and downstream distribution density, number of adjacent nodes, and connection edge weight distribution of the anomaly node in the topology are extracted to generate load anomaly topology path feature data corresponding to each anomaly node, which is used to characterize its anomaly propagation potential and structural influence in the power grid.

[0149] Based on the topology path feature data of load anomalies, and combined with the node energy efficiency index data, a topology-energy efficiency composite feature vector matrix is ​​constructed.

[0150] In this embodiment, the obtained topology path feature data is bound one-to-one with the energy efficiency index data (including node power factor, energy consumption per unit power, load utilization rate, etc.) corresponding to the power system nodes. A structure-energy efficiency composite feature vector matrix is ​​constructed by concatenating feature vectors. Each row of this matrix represents an abnormal node, and its columns include 28 composite features such as topology path depth, path connectivity, node current variance, voltage stability index, and energy efficiency utilization rate. To enhance the robustness of subsequent training, Z-score normalization is performed on the feature matrix to make its mean 0 and variance 1, thereby improving its sensitivity to various influencing factors.

[0151] The structural influence factor embedding training is performed on the topology-energy efficiency composite feature vector matrix to generate load anomaly structural influence feature data.

[0152] In this embodiment, the constructed topology-energy efficiency composite feature vector matrix is ​​input into a graph embedding-based structural factor influence modeling network. The network structure employs a two-layer GCN (Graph Convolutional Network) with a residual connection mechanism, and uses node anomaly labels as weak supervision signals. Training is performed using a loss function based on node clustering embedding centers and node distribution similarity. The training iterations are set to 500 epochs with a learning rate of 0.001. The structural correlation and energy efficiency feature interaction between nodes are modeled during graph convolution, thereby extracting the load anomaly structural influence features of each node and obtaining a 16-dimensional load anomaly structural influence feature vector.

[0153] Historical load fluctuation sequences and multidimensional sensor data of load fluctuations of distribution cabinets are obtained, and feature factor correlation analysis is performed in combination with power system load anomaly node data to obtain load anomaly physical factor weight data.

[0154] In this embodiment, the power load fluctuation sequence of the distribution cabinet over the past 180 days and multi-dimensional sensor data on five types of load fluctuations (temperature, voltage, current, harmonics, and power factor) collected at a 10-minute frequency are obtained through the distribution cabinet management platform. The above multi-dimensional time-series data are mapped to the abnormal nodes, and the Pearson correlation coefficient, mutual information entropy value, and Granger causality test results between each physical factor and the anomaly score are calculated to establish a factor correlation matrix. The factor correlation matrix is ​​then fused using a weighted average method to generate the average influence strength score for each physical factor, ultimately constructing a load anomaly physical factor weighted data containing 12 types of factors.

[0155] By integrating the structural influence feature data of load anomalies and the weight data of physical factors of load anomalies, multidimensional factor compression and reduction are performed, and factor combinations with a cumulative contribution rate threshold greater than or equal to 0.85 are retained to obtain the load anomaly influence factor spectrum.

[0156] In this embodiment, the obtained structural influence feature data and the obtained physical factor weight data are multidimensionally concatenated, and principal component analysis (PCA) is used to reduce the dimensionality of the fused features. A cumulative contribution rate retention threshold of 0.85 is set, meaning that the top few principal components are selected such that their cumulative variance contribution rate is not less than 85%, ensuring the effectiveness and representativeness of the anomalous factor features. In the final obtained load anomalous influence factor spectrum, each factor corresponds to a compressed principal component coefficient and contribution rate.

[0157] Based on the load anomaly impact factor spectrum, significance ranking and dominant factor identification are performed. The factor contribution rate threshold is set to 0.15 to extract the dominant factors of the abnormal nodes and generate the load anomaly dominant factors.

[0158] In this embodiment, the obtained anomaly impact factor spectrum is sorted in descending order according to factor contribution rate. Key factors with a contribution rate greater than 0.15 are extracted as dominant factors, including core variables such as topological path connectivity, load fluctuation sensitivity, energy efficiency utilization disturbance intensity, and node current mean square error. Finally, through logical discrimination rules, these dominant factors are used as the basis for judging various types of load anomalies, and the most important causal features for each anomaly node are output, forming a power system load anomaly dominant factor dataset.

[0159] Optionally, the three-phase current balance assessment of the distribution cabinet in step S4 specifically includes:

[0160] Extract the three-phase current time series data of the load abnormality affected node from the sensor data of the distribution cabinet, perform periodic segmentation, missing compensation and distortion point cleaning on the three-phase current time series data, and obtain calibrated three-phase current standardized time series data.

[0161] In this embodiment, the distribution nodes identified as affected during the load anomaly detection phase are extracted, and historical data from the corresponding three-phase current sensors in the distribution cabinets are retrieved. A continuous three-phase current time series is then created, divided into segments with a 5-minute time granularity. For the extracted data series, a variable-period Fourier window analysis is used to segment it into periodic features. Each segment is set to the length of a complete power frequency cycle (i.e., 160 sampling points within 20ms, with a sampling frequency of 8kHz). For windows with missing data, a spline function based on multidimensional local interpolation is used for compensation, with the interpolation range limited to ±2 sampling points. Subsequently, a combined method of sliding median difference and gradient boundary detection is applied to the current waveform in each cycle segment to identify distortion points, cleaning high-frequency distortion values ​​such as arc disturbances and inductive disturbances, resulting in periodically complete and distortion-free standardized three-phase current time series data.

[0162] Based on the calibrated three-phase current standardized time-series data, a three-phase current time-domain waveform tensor model is constructed, and the three-phase current phase offset rate, phase sequence inconsistency rate and time-domain imbalance fluctuation rate are extracted using the three-phase current time-domain waveform tensor model to obtain the three-phase current imbalance characteristic data.

[0163] In this embodiment, the obtained three-phase current standardized data is organized into a third-order tensor form, constructing a three-phase current time-domain waveform tensor model I∈R3×N×T, where 3 represents the three phases A, B, and C, N is the number of period segments, and T is the number of sampling points per period. Inter-phase difference analysis is performed on this tensor model, using relative mean square error (RMSE) to calculate the inter-phase offset rate with a threshold set at 0.15A. The phase sequence change rate of each period waveform is calculated based on the principal component phase alignment method to identify the proportion of non-standard ABC phase sequences. Furthermore, the time-domain imbalance fluctuation rate is evaluated using the current variance fluctuation coefficient within the period. The three-phase current imbalance feature dataset obtained through the above feature extraction is denoted as F. imb It includes the phase offset rate matrix, the phase sequence inconsistency rate vector, and the volatility vector.

[0164] Based on the three-phase current imbalance characteristic data and the dynamic topology of the power system, a node-level three-phase current imbalance distribution map is constructed; the current offset tension of each abnormal node is calculated using the node-level three-phase current imbalance distribution map to obtain the current tension propagation weight matrix.

[0165] In this embodiment, based on the real-time topology diagram of the power system at the current moment, the corresponding three-phase imbalance characteristic data is injected into each node, and a node-level three-phase current imbalance distribution diagram is constructed according to the branch connection direction. In the diagram, the degree of three-phase imbalance propagated along the branch direction is represented by the gradient difference between nodes, and the current offset tension index T is defined using the graph structure tension analytical method. ij =∥F imb,i -F imb,j∥2, Traverse all connecting edges and construct the current tension propagation weight matrix W between nodes. T =[T ij [The method] retains only the tension paths with Tij>0.2 to eliminate interference from propagation paths in stable regions.

[0166] By fusing the current tension propagation weight matrix and the three-phase current imbalance characteristic data, the three-phase current balance index is calculated to obtain the current balance data of abnormal nodes.

[0167] In this embodiment, the aforementioned current tension propagation weight matrix WT and the node three-phase current imbalance characteristic data F are combined. imb The fusion process employs a structure-aware aggregation algorithm based on weighted averaging to calculate the three-phase current balance index for each node. The balance index ranges from [0,1], with values ​​closer to 1 indicating a closer state of equilibrium. Finally, nodes with current imbalance below a set threshold (which can be set to 0.75) are marked as current-unbalanced nodes, and the current balance data for each abnormal node is output as a key evaluation indicator for subsequent wiring structure correction and anomaly source tracing.

[0168] Of particular importance, the calculation of the three-phase current balance index is specifically as follows:

[0169] The current tension propagation weight matrix is ​​normalized, and the propagation attenuation coefficient is set to 0.8 to generate a normalized current tension propagation matrix.

[0170] In this embodiment, the current tension propagation weight matrix is ​​normalized, and the resulting current tension propagation weight matrix W is then processed. T =[T ij Perform row-by-row normalization using standardized formulas. The normalized propagation attenuation coefficient α is set to 0.8, indicating that the influence of each propagation stage will decrease by 80%, used to simulate the nonlinear attenuation effect of current imbalance along branch paths in the power topology. The normalized matrix W... norm This represents the normalized influence weight of each node on its neighboring nodes in the propagation of three-phase current imbalance, providing a structural basis for subsequent propagation of node feature information. To improve processing stability, a minimum retention weight threshold of 0.01 is set; path influences smaller than this value are truncated to ensure that anomaly propagation only occurs between related nodes.

[0171] The node features of the three-phase current imbalance characteristic data are expanded to construct a node imbalance feature vector matrix.

[0172] In this embodiment, taking each distribution node as a unit, the three types of characteristic values ​​corresponding to its three-phase current phase offset rate, phase sequence inconsistency rate, and time-domain fluctuation rate are expanded into a three-dimensional feature vector. in θ represents the average difference between the currents in phases A and B at this node, in amperes; i σ represents the maximum phase sequence displacement angle within a period, in degrees; i Let F be the variance ratio fluctuation rate within the three-phase current cycle. Combine the three-phase imbalance eigenvectors of all nodes to form the node imbalance eigenvector matrix F = [f1, f2, ..., f...]. n ] T , where n is the number of anomaly candidate nodes participating in the analysis. To ensure the numerical stability of subsequent propagation modeling, all features are uniformly standardized using Z-score, making each feature comparable across different dimensions.

[0173] Based on the normalized current tension propagation matrix and the node imbalance feature vector matrix, the modeling of the imbalance feature propagation between nodes is performed to obtain the initial imbalance propagation influence matrix;

[0174] In this embodiment, a single-weighted propagation model is used to model the impact of imbalance between nodes. The propagation model has the form F′=W norm F′ is the propagated node feature vector matrix, representing the new imbalance-aware characteristics formed by each node under the tension influence of adjacent nodes. This propagation process can be viewed as a node "sensing" the imbalance state of its neighboring nodes in the topology and adaptively correcting its local current balance, reflecting the multi-point coupling characteristics of three-phase imbalance in actual power systems. During propagation modeling, the output feature vector of each node is considered as the overall imbalance-aware influence it receives, and these are ultimately combined into the initial imbalance propagation influence matrix F′, used for subsequent current balance calculations.

[0175] Based on the initial imbalance propagation influence matrix, the node current balance index is calculated. The node current balance index of all nodes is truncated, normalized, and mapped to intervals. The balance level is divided into intervals, and abnormal node current balance data is generated.

[0176] In this embodiment, for each node eigenvector in the initial imbalance propagation influence matrix F′, its mean square deviation value in the three-phase dimension is first calculated and denoted as the node initial imbalance index. in This is the mean of the feature vector after propagation at this node. For all nodes, U... i The index is globally normalized, mapped to the [0,1] interval, and then processed in reverse to generate the final current balance index. The closer the value is to 1, the more balanced it is. The balance level is divided into the following ranges: [0.9, 1.0] for "high balance", [0.7, 0.9] for "medium balance", [0.5, 0.7] for "low balance", and [0.0, 0.5] for "severe imbalance". Each abnormal node is assigned a corresponding level label, and the final abnormal node current balance data table is output for reference in subsequent fault handling and structural reconstruction.

[0177] Optionally, the abnormal node classification in step S4 specifically includes:

[0178] The current balance data of abnormal nodes is divided into balance intervals to obtain graded current balance label data.

[0179] In this embodiment, the current balance data B is analyzed. i The distribution of current balance is divided into four levels according to the set balance levels: [0.9, 1.0], [0.7, 0.9], [0.5, 0.7], and [0.0, 0.5]. The current balance value of each node is matched with its corresponding label using the set interval criteria, generating a graded label dataset L. i For example, if the current balance B at node i i =0.92, then the node is marked as "high balance" level, with a label of 1; if B i If the current balance is 0.68, it is marked as "Medium Balance" level with a label of 2. This process ensures the applicability of current balance label data under different power grid conditions, supports accurate classification of load anomaly types, and facilitates subsequent fault diagnosis and handling.

[0180] Based on the dynamic topology of the power system, the topological adjacency information and branch connection relationship of nodes affected by load anomalies are extracted, and the inter-node connection matrix is ​​constructed.

[0181] In this embodiment, the topology information of the power system is used to extract the neighboring nodes and branch connections of abnormal nodes. A distance threshold d for branch connections is set. thresh =3km is used as the association judgment condition. If the physical distance between two nodes is less than this threshold and they are within the same power grid section, then a direct electrical connection is considered to exist between the two nodes. Therefore, a node connection matrix A = [a ij ], where a ij =1 indicates that node i and node j have a direct electrical connection, otherwise it is 0. This connection matrix provides important topological constraint information for subsequent structural analysis.

[0182] By combining the node phase connection matrix and graded current balance label data, the three-phase structure distribution of each node affected by load anomalies is analyzed to obtain three-phase phase sequence consistency characteristic data.

[0183] In this embodiment, based on the node phase-to-phase connection matrix A and the node current balance label data L... i The consistency of the three-phase currents on the phase-to-phase lines at each node is analyzed. For each node, a phase sequence consistency threshold θ is set. thresh =15. If the phase angle deviation of the interphase current waveform is greater than this threshold, then the node is considered to have a phase sequence inconsistency problem. Calculate the phase sequence consistency characteristic C for each node. i =max(Δθ) i ), where Δθ i This represents the average phase sequence offset of the node. Finally, based on these characteristic values, phase sequence consistency feature data for the node is generated for subsequent detection and repair of improper wiring. Alternatively, one can first check if all three phases are connected to the node; if only one or two phases are connected, it is marked as "phase missing"; then check if multiple branches are simultaneously connected to the same phase; if so, it is marked as "phase duplicate"; finally, check if there are cross connections that do not conform to the topology rules (such as a branch connecting phase A to phase B); such nodes will be marked as "phase sequence disorder". Through the above structural analysis, three phase sequence consistency features are extracted: phase integrity, phase balance, and connection standardization, which are represented in vector form to provide structural feature support for subsequent wiring pattern recognition.

[0184] Topological structure projection and constraint rule matching are performed on the three-phase phase sequence consistency characteristic data, and wiring pattern identification is performed to obtain improper wiring identification data;

[0185] In this embodiment, a three-phase connection pattern of the power system is constructed based on topological projection and matched with known connection rules. For example, assuming the power system adopts a "star connection" pattern, the phase sequence consistency characteristic C at each node is... i If the phase sequence pattern does not match that under "star connection", the node is marked as a potentially improperly connected node. Set the threshold θ for wiring rule matching. match =30. When the phase sequence consistency deviation exceeds this threshold, the node will be marked as improperly wired. Through the process of matching wiring rules, a set of improper wiring identification data R is obtained. i Each node is determined to have improper wiring based on its three-phase sequence consistency characteristics. Rules may include, but are not limited to: no duplicate phase connections in a single node; three-phase connections in the same node should be evenly distributed; and phases should not be cross-connected. By matching the three-phase structure of each node against the topology path rules item by item, if a node is found to violate any rule, it is marked as an "improperly wired node." The identification results are output in Boolean form (1 for abnormal, 0 for normal), along with a description of the violation type, for subsequent classification decisions and power system maintenance prompts.

[0186] By integrating graded current balance label data and improper wiring identification data for joint logic discrimination, the nodes affected by load anomalies are divided into three-phase load offset nodes and three-phase improper wiring nodes.

[0187] In this embodiment, a logical combination discrimination model is established based on the aforementioned two key anomaly indicators: current balance label and wiring anomaly identifier. The rules are as follows: if the wiring structure of a node conforms to the specifications (improper wiring identifier is 0), but the balance label is in the state of slight or moderate imbalance, it is judged as a "three-phase load offset node"; if the node is identified as improperly wired or its balance label is severely unbalanced, it is marked as a "three-phase improper wiring node".

[0188] Of particular importance, the joint logic judgment is specifically as follows:

[0189] The graded current balance label data is coded with label type. The threshold for the low balance interval is set to 0.85 and the threshold for the severe imbalance interval is set to 0.65. Nodes with current balance higher than 0.85 are classified as normal nodes, nodes with current balance lower than 0.85 but higher than 0.65 are classified as slightly offset nodes, and nodes with current balance lower than 0.65 are classified as severely offset nodes, thus generating structured current balance status data.

[0190] In this embodiment, the current balance values ​​of abnormal load nodes in the power system are coded with labels. The balance classification criteria are set as follows: nodes with a current balance value P greater than 0.85 are categorized as "normal nodes," with a label code of 0; nodes in the interval (0.65, 0.85) are considered "slightly offset nodes," with a label code of 1; and nodes less than or equal to 0.65 are marked as "severely offset nodes," with a label code of 2. This classification uses the pandas.cut function in Python for interval segmentation and encoding, ensuring consistency and automation in label generation. Finally, each abnormal node and its corresponding balance status label are combined into a structured tabular data format, with fields including: node ID, current balance value, and balance level label, facilitating subsequent fusion analysis with wiring status information.

[0191] Based on the improper wiring identification data, mark the wiring status type of each load abnormality affected node, set the identification variable fi∈{0,1}, where fi=1 indicates that there is improper wiring behavior, and construct the wiring status vector;

[0192] In this embodiment, the set of improperly wired nodes identified in the previous steps is used to set an identifier variable fi to distinguish whether a node has a wiring abnormality. Specifically, an empty vector is created for each abnormal node, with its corresponding number, and initialized to fi = 0. The node numbers in the improper wiring identifier data are traversed; if a node appears in the abnormality list, the corresponding fi is assigned a value of 1, indicating that the node has a wiring pattern that does not conform to the standard specifications, such as phase sequence disorder or phase repetition. This wiring status vector is in the form of a one-dimensional array F = [f1, f2, ..., fn], where n is the number of nodes. It can be subsequently fused with current balance structured data to achieve logical discrimination under multi-dimensional conditions.

[0193] Logical rules are used to combine the structured current balance state data and the wiring state vector, and the following joint discrimination rule is set:

[0194] If the load anomaly affects the balance level of a node and the node's identifier variable is 0, then the node is marked as a three-phase load offset node.

[0195] If the identifier variable of the node affected by the abnormal load is 1, then the node is marked as an improper three-phase wiring node;

[0196] If a load anomaly affects a node and simultaneously meets the conditions of severe imbalance level and the node's identifier variable is 1, then the node will be marked as an improperly connected three-phase power node.

[0197] In this embodiment, a joint logic discrimination is performed based on the node's current balance status label (0,1,2) and wiring status vector (0 / 1) to construct an anomaly type identification model. The specific logic rules are as follows: if a node's current balance label is 2 (severe offset) and wiring status fi = 0, it is initially judged as a "three-phase load offset node"; if a node's wiring status fi = 1, regardless of the current balance status, it is uniformly marked as a "three-phase improper wiring node"; if the node's current balance label is 2 and fi = 1 simultaneously, it is still marked as a "three-phase improper wiring node" according to the priority processing principle. This joint logic can be implemented through conditional priority encoding, using Python logic discrimination statements or constructing a decision tree model to complete automatic judgment and classification label output. During logic discrimination, node label classification is performed through a preset logic rule chain. Using a data table as input, the system checks the "balance label" and "wiring identification variable" of each node row by row, storing them in the variables `level` and `fi` respectively. Then, nested judgments are performed according to rules: if `level == 2` and `fi == 0`, the output node type is "load offset"; if `fi == 1`, the output type is set to "improper wiring" regardless of the `level`; if both `level == 2` and `fi == 1` are satisfied, "improper wiring" is still output to ensure that the fault identification results cover the most critical issues. Priority decision trees or multi-condition mapping matrices are used to improve efficiency during the logical combination judgment process. The results are output in tabular form, including node number, balance label, wiring identification, and final judgment result.

[0198] Perform a discrimination logic mapping on nodes affected by load anomalies to generate three-phase load offset nodes and three-phase improper wiring nodes.

[0199] In this embodiment, during the final judgment and execution phase, a standardized result set is generated from the classification results of each node obtained through logical combination analysis. The result output is saved in the form of two node lists: List A contains the numbers and status information of all nodes judged as "three-phase load offset nodes"; List B contains the numbers, anomaly types, and wiring characteristic descriptions of all nodes judged as "improper three-phase wiring nodes". This structured result can be used for subsequent equipment inspection, abnormal node re-inspection, and wiring adjustment operations in the power supply system, and can also be synchronously input into the power operation and maintenance management platform for abnormal work order generation.

[0200] Optionally, this specification also provides a power system operation and maintenance system for a low-voltage control intelligent distribution cabinet, used to execute the power system operation and maintenance method for the low-voltage control intelligent distribution cabinet as described above. The power system operation and maintenance system for the low-voltage control intelligent distribution cabinet includes:

[0201] The topology analysis module is used to acquire power operation data and sensor data of the distribution cabinet, and to perform power system topology analysis based on the power operation data of the distribution cabinet to obtain the dynamic topology of the power system.

[0202] The potential load anomaly analysis module is used to identify the power system energy efficiency pattern from the distribution cabinet sensor data through the dynamic topology of the power system, and obtain the power system energy efficiency pattern data; and to perform potential load anomaly analysis on the power system energy efficiency pattern data, and obtain the power system load anomaly node data.

[0203] The abnormal impact structure division module is used to analyze the load abnormality dominant factors based on the load abnormality node data of the power system, and divide the load abnormality impact structure of the dynamic topology of the power system according to the load abnormality dominant factors to obtain the load abnormality impact nodes.

[0204] The three-phase current balance assessment module is used to assess the three-phase current balance of the distribution cabinet based on the sensor data of the distribution cabinet to identify nodes affected by abnormal loads, and to classify abnormal nodes based on the current balance of the distribution cabinet to obtain three-phase load offset nodes and improper three-phase wiring nodes.

[0205] The abnormal node power redistribution module is used to reconstruct the power distribution of three-phase load offset nodes and improperly connected three-phase nodes, obtain abnormal node power distribution data, and transmit it to the distribution cabinet management platform to execute power distribution tasks.

[0206] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A power system operation and maintenance method for an intelligent distribution cabinet for low-voltage control, characterized in that, Includes the following steps: Step S1: Obtain the power operation data and sensor data of the distribution cabinet, and perform power system topology analysis based on the power operation data of the distribution cabinet to obtain the dynamic topology of the power system; Step S2: Use the power system dynamic topology to identify the power system energy efficiency pattern from the distribution cabinet sensor data to obtain power system energy efficiency pattern data; Potential load anomaly analysis is performed on power system energy efficiency model data to obtain power system load anomaly node data; Step S3: Analyze the load anomaly dominant factors based on the load anomaly node data of the power system, and divide the load anomaly impact structure of the dynamic topology of the power system according to the load anomaly dominant factors to obtain the load anomaly affected nodes. Step S4: Based on the sensor data of the distribution cabinet, evaluate the three-phase current balance of the distribution cabinet for nodes affected by abnormal loads, and classify abnormal nodes according to the current balance of the distribution cabinet to obtain the three-phase load offset nodes and the improper three-phase wiring nodes. Step S5: Reconstruct the power distribution for the three-phase load offset nodes and improperly connected three-phase nodes to obtain the power distribution data for the abnormal nodes, and transmit it to the distribution cabinet management platform to execute the power distribution task.

2. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 1, characterized in that, Step S1 is as follows: Step S11: Obtain the power operation data and sensor data of the power distribution cabinet, and perform data preprocessing on the power operation data and sensor data of the power distribution cabinet to obtain the power operation data and sensor data of the power distribution cabinet to be analyzed. Step S12: Extract the voltage distribution characteristics and current flow characteristics of the distribution cabinet based on the power operation data to be analyzed; Step S13: Perform power system backbone topology analysis based on the voltage distribution characteristics of the distribution cabinet to obtain the power system backbone topology structure; Step S14: Use the current flow characteristics of the distribution cabinet to perform global topology completion on the main topology of the power system, thereby obtaining the power system topology; Step S15: Perform power flow calculation on the power system topology and construct the dynamic topology of the power system based on the power flow calculation results.

3. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 2, characterized in that, Step S13 is as follows: Step S131: Classify the voltage levels according to the voltage distribution characteristics of the distribution cabinet, set the voltage threshold range to [400V, 220V] to partition the bus nodes, and construct the bus layer structure to obtain the bus level data. Step S132: Calculate the potential difference between different bus levels based on the bus level data, select lines with potential difference constraints less than or equal to [5V, 20V] as main branches, perform topology consistency verification, and generate initial main branch data; Step S133: Construct a preliminary backbone topology graph using the bus nodes in the bus hierarchy data as vertices and the initial backbone branch data as edges, and perform connectivity analysis to remove isolated nodes and generate a preliminary backbone topology structure. Step S134: Based on the voltage distribution characteristics and current flow characteristics of the distribution cabinet, perform power flow direction analysis, and adjust the branch weights of the preliminary trunk topology according to the power flow direction to obtain the optimized trunk topology. Step S135: Perform topology mapping on the optimized backbone topology to obtain the power system backbone topology.

4. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 2, characterized in that, Step S14 is as follows: Step S141: Perform weighted directed graph modeling on the current flow characteristics of the distribution cabinet, set the current directionality threshold to [0.3A, 3.0A], and thus construct the current flow correlation graph; Step S142: Perform node alignment and edge mapping analysis on the current flow correlation diagram and the power system backbone topology to identify the missing connection parts in the power system backbone topology and obtain the node edge data to be completed in the backbone structure. Step S143: Perform graph structure reasoning on the edge data of the nodes to be completed in the backbone structure, perform missing connection completion, and obtain global topology prediction data; Step S144: Perform edge weight fusion and unified node encoding on the global topology prediction data and the power system backbone topology to obtain the preliminary power system topology. Step S145: Perform consistency verification and closed-loop verification on the preliminary power system topology to obtain the power system topology.

5. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 1, characterized in that, Step S2 is as follows: Step S21: Perform spatial mapping and time window segmentation on the sensor data of the distribution cabinet to be analyzed, and bind the time window sensor-topology node according to the node-branch relationship in the dynamic topology of the power system to obtain the spatiotemporal feature map of the power system node. Step S22: Extract weighted aggregated energy efficiency indicators based on the spatiotemporal feature map of power system nodes to obtain node energy efficiency indicator data; Step S23: Construct an energy efficiency attribute map using node energy efficiency index data, and extract node energy efficiency behavior features based on the energy efficiency attribute map; Step S24: Perform energy efficiency behavior clustering based on node energy efficiency behavior characteristics to obtain power system energy efficiency pattern data; Step S25: Perform anomaly factor detection and model deviation analysis on the power system energy efficiency model data, extract potential load anomaly candidate nodes, and score the confidence of the potential load anomaly candidate nodes to obtain power system load anomaly node data.

6. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 5, characterized in that, Step S25 is as follows: Step S251: Based on the feature vectors of each cluster center in the power system energy efficiency model data, construct the energy efficiency behavior residual matrix of each node, and calculate the Euclidean distance between the node energy efficiency behavior features and the corresponding node energy efficiency behavior residual matrix to obtain the node energy efficiency behavior residual data. Step S252: Based on the residual data of node energy efficiency behavior, perform slope fitting and mutation detection, calculate the abnormal factor score, and obtain the abnormal factor score data; set the dynamic threshold range of the abnormal score according to the abnormal factor score data; Step S253: Evaluate the energy efficiency behavior coordination degree of adjacent nodes in the dynamic topology of the power system to obtain the energy efficiency behavior coordination degree of adjacent nodes; Step S254: Combine the dynamic threshold range of anomaly scores and the synergy of energy efficiency behavior of adjacent nodes to construct a weighted anomaly distribution map, identify and extract nodes with high anomaly factor scores, and obtain potential load anomaly candidate node data. Step S255: Evaluate the node confidence of the potential load anomaly candidate node data, and set the scoring confidence threshold to [0.6, 0.9] to screen high-confidence nodes, thereby obtaining the power system load anomaly node data.

7. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 1, characterized in that, The specific factors for analyzing load anomalies mentioned in step S3 are: The path backtracking depth is set to [3,5] layers to dynamically backtrack the upstream and downstream paths of the load anomaly node data in the power system, construct the load anomaly path graph, and extract the structural position of each anomaly node in the dynamic topology of the power system and the correlation degree with adjacent nodes to obtain the load anomaly topology path feature data. Based on the topology path feature data of load anomalies, and combined with the node energy efficiency index data, a topology-energy efficiency composite feature vector matrix is ​​constructed. The structural influence factor embedding training is performed on the topology-energy efficiency composite feature vector matrix to generate load anomaly structural influence feature data. Historical load fluctuation sequences and multidimensional sensor data of load fluctuations of distribution cabinets are obtained, and feature factor correlation analysis is performed in combination with power system load anomaly node data to obtain load anomaly physical factor weight data. By integrating the structural influence feature data of load anomalies and the weight data of physical factors of load anomalies, multidimensional factor compression and reduction are performed, and factor combinations with a cumulative contribution rate threshold greater than or equal to 0.85 are retained to obtain the load anomaly influence factor spectrum. Based on the load anomaly impact factor spectrum, significance ranking and dominant factor identification are performed. The factor contribution rate threshold is set to 0.15 to extract the dominant factors of the abnormal nodes and generate the load anomaly dominant factors.

8. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 1, characterized in that, The three-phase current balance assessment of the distribution cabinet mentioned in step S4 specifically includes: Extract the three-phase current time series data of the load abnormality affected node from the sensor data of the distribution cabinet, perform periodic segmentation, missing compensation and distortion point cleaning on the three-phase current time series data, and obtain calibrated three-phase current standardized time series data. Based on the calibrated three-phase current standardized time-series data, a three-phase current time-domain waveform tensor model is constructed, and the three-phase current phase offset rate, phase sequence inconsistency rate and time-domain imbalance fluctuation rate are extracted using the three-phase current time-domain waveform tensor model to obtain the three-phase current imbalance characteristic data. Based on the three-phase current imbalance characteristic data and the dynamic topology of the power system, a node-level three-phase current imbalance distribution map is constructed; the current offset tension of each abnormal node is calculated using the node-level three-phase current imbalance distribution map to obtain the current tension propagation weight matrix. By fusing the current tension propagation weight matrix and the three-phase current imbalance characteristic data, the three-phase current balance index is calculated to obtain the current balance data of abnormal nodes.

9. The power system operation and maintenance method for the intelligent distribution cabinet for weak current control according to claim 1, characterized in that, The abnormal node classification mentioned in step S4 is specifically as follows: The current balance data of abnormal nodes is divided into balance intervals to obtain graded current balance label data. Based on the dynamic topology of the power system, the topological adjacency information and branch connection relationship of nodes affected by load anomalies are extracted, and the inter-node connection matrix is ​​constructed. By combining the node phase connection matrix and graded current balance label data, the three-phase structure distribution of each node affected by load anomalies is analyzed to obtain three-phase phase sequence consistency characteristic data. Topological structure projection and constraint rule matching are performed on the three-phase phase sequence consistency characteristic data, and wiring pattern identification is performed to obtain improper wiring identification data; By integrating graded current balance label data and improper wiring identification data for joint logic discrimination, the nodes affected by load anomalies are divided into three-phase load offset nodes and three-phase improper wiring nodes.

10. A power system operation and maintenance system for an intelligent distribution cabinet for low-voltage control, characterized in that, The power system operation and maintenance method for executing the intelligent distribution cabinet for weak current control as described in claim 1, wherein the power system operation and maintenance system for the intelligent distribution cabinet for weak current control includes: The topology analysis module is used to acquire power operation data and sensor data of the distribution cabinet, and to perform power system topology analysis based on the power operation data of the distribution cabinet to obtain the dynamic topology of the power system. The potential load anomaly analysis module is used to identify the power system energy efficiency pattern from the distribution cabinet sensor data through the dynamic topology of the power system, and obtain the power system energy efficiency pattern data; and to perform potential load anomaly analysis on the power system energy efficiency pattern data, and obtain the power system load anomaly node data. The abnormal impact structure division module is used to analyze the load abnormality dominant factors based on the load abnormality node data of the power system, and divide the load abnormality impact structure of the dynamic topology of the power system according to the load abnormality dominant factors to obtain the load abnormality impact nodes. The three-phase current balance assessment module is used to assess the three-phase current balance of the distribution cabinet based on the sensor data of the distribution cabinet to identify nodes affected by abnormal loads, and to classify abnormal nodes based on the current balance of the distribution cabinet to obtain three-phase load offset nodes and improper three-phase wiring nodes. The abnormal node power redistribution module is used to reconstruct the power distribution of three-phase load offset nodes and improperly connected three-phase nodes, obtain abnormal node power distribution data, and transmit it to the distribution cabinet management platform to execute power distribution tasks.

Images