Power equipment maintenance plan intelligent decision-making system based on spatial-temporal characteristics and application method

Abstract

The invention relates to the technical field of electrical equipment maintenance, in particular to an electrical equipment maintenance plan intelligent decision-making system based on spatial-temporal characteristics and an application method. Comprising a space-time knowledge graph construction unit; the spatial-temporal feature extraction unit is used for extracting equipment spatial distribution features, load time sequence features and maintenance resource time constraint features from the power grid operation data, constructing a multi-dimensional feature vector, and realizing feature extraction based on a graph traversal algorithm and an LSTM neural network; an intelligent decision-making unit; a main distribution network space-time cooperation unit; and a space-time risk assessment unit. According to the invention, the splicing of the main network and the distribution network CIM model is realized through the main and distribution network space-time cooperation unit, the full-voltage class fusion topological graph is generated, and the influence range of main network maintenance on the power supply path of the distribution network can be automatically identified based on a cross-voltage class topology tracking method; the mechanism effectively reduces the problem of repeated power failure or forwarding path conflict caused by insufficient coordination of the main network and the distribution network.

Description

Technical Field

[0001] This invention relates to the field of power equipment maintenance technology, and more specifically, to an intelligent decision-making system and application method for power equipment maintenance planning based on spatiotemporal characteristics. Background Technology

[0002] In the field of power system maintenance management, traditional maintenance planning relies on manual analysis of static topology and historical experience, making it difficult to simultaneously consider the coordinated dispatch of the main and distribution networks, equipment conflict verification, and real-time risk assessment. This leads to pain points in scenarios such as the Yuxi Power Dispatch and Control Center, including high staffing pressure, incomplete identification of grid weak points, and high risk of repeated power outages. Existing technologies lack in-depth support for the fusion of spatiotemporal characteristics, dynamic risk quantification, and the linkage of main and distribution network topologies. Therefore, there is an urgent need to improve the scientific rigor and reliability of maintenance plans through intelligent technologies.

[0003] For example, Chinese patent CN202110920543.5 discloses a power equipment maintenance decision generation method based on reinforcement learning. This invention can indirectly realize communication between multiple power devices, has high data utilization, and has a low application threshold in professional fields. This scheme solves maintenance strategies through Markov decision processes and reinforcement learning, but only focuses on modeling the operating status of a single device, without constructing a topological association model of the main and distribution networks, and cannot analyze the chain effects of maintenance operations across voltage levels. In addition, it relies on the calculation of preset weights for the "first cut set" and "second cut set", lacks topological reasoning capabilities based on knowledge graphs, and is difficult to adaptively handle maintenance conflict verification under complex network structures. Another example is Chinese patent CN201611089917.9, which discloses a power equipment maintenance decision-making method using equal degradation theory and equipment risk. This invention introduces equal degradation theory into the power industry to fully consider the changes in equipment failure probability before and after maintenance, analyzes and compares the changes in equipment risk before and after maintenance, and considers equipment maintenance costs to obtain the optimal maintenance plan. This invention analyzes equipment risk based on the theory of equal degradation, but it only optimizes maintenance costs from an asset management perspective, without incorporating the temporal and spatial dimensions of load timing characteristics and real-time operational data. Therefore, it cannot dynamically assess the impact of load fluctuations during maintenance periods on grid stability. Furthermore, its risk assessment does not combine N-1 fault simulation and graph traversal algorithms, making it difficult to quantify the scope of power outages and their impact on users, and lacking a visual early warning capability for grid security risks.

[0004] Despite the technical advantages of the above-mentioned technical solutions, the existing technical solutions also have the following common technical defects: First, the main and distribution network maintenance coordination analysis lacks a topology linkage mechanism. In power system maintenance management, traditional technical solutions usually treat the main network and distribution network as independent systems and lack the ability to conduct linkage analysis of the main and distribution network topology relationship. This fragmented topology analysis model means that the coordinated maintenance plan for the main and distribution networks still relies on manual experience for coordination, which can easily lead to problems such as repeated power outages or insufficient power transfer. Secondly, insufficient fusion of spatiotemporal features limits the accuracy of decision-making. Existing maintenance decision-making technologies mostly focus on single-dimensional features (such as equipment health or historical maintenance cycles), lacking the fusion analysis of spatial topology relationships and time series data. At the same time, traditional methods do not model equipment health data according to the time dimension, making it difficult to predict changes in the probability of failure before and after maintenance, affecting the scientific nature of decision-making. Thirdly, dynamic risk assessment lacks visualization and real-time simulation capabilities. Traditional technical solutions in power grid risk assessment mostly rely on static rule bases or offline calculations, lacking dynamic simulation capabilities based on real-time topology. In addition, the risk assessment results lack visualization (such as risk area heat maps, early warning signs for key equipment), resulting in low efficiency of risk identification during the maintenance decision-making process and difficulty in generating degradation solutions in a timely manner. In view of this, we propose an intelligent decision-making system and application method for power equipment maintenance planning based on spatiotemporal features. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent decision-making system and application method for power equipment maintenance planning based on spatiotemporal characteristics, so as to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, one objective of this invention is to provide an intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics, comprising:

[0007] The spatiotemporal knowledge graph construction unit is used to parse the power grid CIM model into a graph database structure containing spatial topological relationships and temporal dimension attributes, and to generate a spatiotemporally related device network model through topology pruning optimization.

[0008] The spatiotemporal feature extraction unit is used to extract equipment spatial distribution features, load time series features, and maintenance resource time constraint features from power grid operation data, construct multi-dimensional feature vectors, and realize feature extraction based on graph traversal algorithm and LSTM neural network;

[0009] The intelligent decision-making unit, based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit and the multidimensional feature vector constructed by the spatiotemporal feature extraction unit, generates a maintenance plan scheduling scheme that takes into account both spatial topological constraints and time resource balance through a genetic algorithm.

[0010] The main and distribution network spatiotemporal coordination unit is used to stitch together the CIM models of the main network and the distribution network to generate a fused topology map of all voltage levels. Based on the topology analysis, it analyzes the impact of main network maintenance on the power supply path of the distribution network and adjusts the maintenance time window.

[0011] The spatiotemporal risk assessment unit identifies the power loss range of N-1 faults based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit, and predicts the fault probability by combining the equipment health time series and generates a risk degradation plan.

[0012] As a further improvement to this technical solution, the spatiotemporal knowledge graph construction unit includes a spatial topology modeling module and a temporal attribute annotation module, wherein:

[0013] The spatial topology modeling module is used to parse the substations, lines, transformers, circuit breakers, disconnectors, and busbars in the power grid CIM model into graph nodes, and construct a spatial topology network with electrical connection relationships as edges. The graph nodes include attributes such as equipment type, voltage level, and physical location coordinates. The spatial topology network supports hierarchical display based on voltage level, and the edges between nodes include electrical parameters such as impedance value and line length.

[0014] The time attribute annotation module is used to annotate the time dimension attributes (such as maintenance cycle, historical fault time series, last maintenance timestamp, etc.) of equipment nodes and form a device-time association table.

[0015] As a further improvement to this technical solution, the spatiotemporal knowledge graph construction unit also includes a topology pruning optimization module and a graph database storage module, wherein:

[0016] The topology pruning optimization module generates and executes a topology pruning strategy based on the spatial topology network constructed by the spatial topology modeling module.

[0017] Furthermore, the topology pruning strategy specifically includes:

[0018] Endpoint identification rule: Identify endpoint nodes in the spatial topology network that are directly connected to the connection points. The endpoint node is a device that is connected to other device nodes only through one edge.

[0019] Layered cropping strategy:

[0020] For equipment nodes with voltage levels of 220kV and above, the complete topology is retained without pruning.

[0021] For equipment nodes with voltage levels of 110kV and below, cut off the endpoint nodes that are directly connected to the connection point, and merge the electrical parameters (such as impedance values) of the cut-off endpoints into the remaining connection edges.

[0022] Core path retention mechanism: Based on the importance weight of equipment nodes in the spatial topology network, the topology path connected to core equipment nodes such as main transformers and busbars is retained, and the importance weight is determined by the load impact degree and power supply range level of the equipment.

[0023] The graph database storage module uses a graph database to store spatiotemporal knowledge graphs.

[0024] Furthermore, the storage structure is as follows:

[0025] Equipment node: includes equipment identifier, name, type, voltage level, physical location and time attribute, where the time attribute includes maintenance cycle and historical fault time information;

[0026] Connection edge: Includes connection endpoint identification, connection relationship type and electrical parameters, including impedance value, line length, current carrying capacity and related parameters.

[0027] As a further improvement to this technical solution, the spatiotemporal feature extraction unit includes a spatial feature extraction module, a temporal feature extraction module, and a feature fusion module, wherein:

[0028] The spatial feature extraction module is used to extract the spatial distribution features of equipment from the power grid operation data. It uses a depth-first search algorithm to traverse the spatiotemporal knowledge graph and extract the features of equipment adjacency, power supply range hierarchy, and spatial topology connectivity. Then, based on the physical location coordinates of the equipment, it calculates the Euclidean distance and spatial clustering coefficient between the equipment to generate a spatial location feature vector.

[0029] The time feature extraction module is used to extract load time sequence features and maintenance resource time constraint features from power grid operation data. It uses an LSTM neural network to train historical load data. The LSTM network contains two hidden layers with 64 neurons in each layer to capture the time dependence and periodic features of load data. It also extracts the available time window of maintenance resources and resource allocation time constraints to generate time constraint feature vectors.

[0030] The feature fusion module is used to fuse the spatial feature vector generated by the spatial feature extraction module and the temporal feature vector generated by the temporal feature extraction module. It generates an initial multidimensional feature vector by concatenating the features according to the dimensions using a feature concatenation method, and normalizes the initial multidimensional feature vector using the Z-score normalization method to eliminate the difference in dimensions, thereby generating the final multidimensional feature vector.

[0031] As a further improvement to this technical solution, the intelligent decision-making unit includes a constraint analysis module, a genetic algorithm optimization module, and a scheme evaluation module, wherein:

[0032] The constraint parsing module is used to parse the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit and the multidimensional feature vector generated by the spatiotemporal feature extraction unit, and extract the constraint conditions required for maintenance plan arrangement.

[0033] Furthermore, the constraint resolution module extracts spatial topological constraints and temporal resource constraints, specifically including:

[0034] Spatial dependency constraint identification:

[0035] Based on the adjacency relationships of devices in the spatiotemporal knowledge graph (such as the connection relationship between circuit breakers and disconnect switches) and the power supply range hierarchy (such as the power supply area of ​​a 110kV line), the spatial correlation of maintenance operations is analyzed. For example, when device A is under maintenance, if adjacent device B has a direct electrical connection with A (such as sharing a busbar), then electrical isolation operations must be performed simultaneously (spatial dependency constraint).

[0036] Time resource balance constraint identification:

[0037] By combining the equipment time attributes (maintenance cycle, last maintenance timestamp) with the maintenance resource time constraint features output by the spatiotemporal feature extraction unit (such as the available time window of maintenance teams and the dispatch cycle of work vehicles), the upper limit of the number of maintenance personnel and the upper limit of equipment dispatch capacity that can be deployed within the same time window (such as 8:00-18:00 on a single day) are calculated (time resource balance constraint).

[0038] The genetic algorithm optimization module generates a maintenance plan arrangement scheme based on the spatiotemporal constraints extracted by the constraint parsing module and the improved genetic algorithm. The improved genetic algorithm integrates an elite retention strategy, a dynamic parameter adjustment mechanism and a dual-objective fitness function.

[0039] The scheme evaluation module is used to evaluate the maintenance plan arrangement scheme generated by the genetic algorithm optimization module using multiple indicators. If the evaluation result does not meet the preset engineering threshold, the genetic algorithm optimization module is triggered to iterate and optimize again.

[0040] As a further improvement to this technical solution, the genetic algorithm optimization module generates a maintenance plan scheduling scheme through an improved genetic algorithm, including the following steps:

[0041] S320.1 Population Initialization:

[0042] Integer encoding is used, and each chromosome contains N gene loci, equal to the total number of devices to be repaired. Each gene locus is g. i The gene value is an integer in the range [0, T] corresponding to the start time of maintenance of the i-th device (T is the preset total length of the time window); M chromosomes of the preset population size are randomly generated to form the initial population;

[0043] S320.2, Fitness Calculation:

[0044] For each chromosome, based on the maintenance time allocation encoded by it, calculate the two-objective fitness function F:

[0045] F = ω1·(1 - V) + ω2·(1 - B);

[0046] Where, V is the violation degree of spatial topology constraint, and the calculation formula is: Where, Ω i is the set of adjacent devices of device i (extracted from the spatio-temporal knowledge graph); I(g i , g j ) is the conflict indication function: if the maintenance time windows of devices i and j overlap, then I(g i , g j ) = 1, otherwise I(g i , g j ) = 0; B is the time resource balance degree, and the calculation formula is: Where, R t is the amount of maintenance resources allocated within the time window t, σ(R1, R2,..., R T ) is the standard deviation of the sequence of the amount of maintenance resources R t within the time window t; is the average value of the amount of maintenance resources R t within the time window t; ω1 and ω2 are weight coefficients, satisfying ω1 + ω2 = 1, respectively representing the priorities of spatial topology constraint and time resource balance;

[0047] S320.3, Elite Retention Strategy:

[0048] Select the top E individuals (E is the preset elite number and E < M) with the highest fitness from the current population and directly copy them to the next-generation population to ensure that the optimal solution is not destroyed by genetic operations;

[0049] S320.4, Dynamic Parameter Adjustment:

[0050] The crossover probability P c and the mutation probability P m are dynamically adjusted with the iteration number k:

[0051] The crossover probability P c : linearly decreases from the initial value P c0 to the minimum threshold P c1 , and the formula is:

[0052]

[0053] Where, K is the preset maximum iteration number;

[0054] The mutation probability Pm : linearly increase from the initial value P m0 to the maximum threshold P m1 , and the formula is:

[0055]

[0056] where K is the preset maximum number of iterations;

[0057] S320.5. Genetic operation execution:

[0058] Selection: Use the tournament selection method. Each time, randomly select T s individuals (T s is the preset tournament scale) from the population, and select the individual with the highest fitness to enter the mating pool. Repeat the operation until the size of the mating pool reaches M - E;

[0059] Crossover: Randomly pair the individuals in the mating pool according to P c and generate offspring using two-point crossover:

[0060] Randomly select two crossover points p1, p2 (1 ≤ p1 < p2 ≤ n) and exchange the gene segment of the parental chromosome in the interval [p1, p2];

[0061] Mutation: Randomly select a gene position on the offspring chromosome according to P m and replace its value with a random integer within the range of [0, T];

[0062] S320.6. Construction of the new generation population:

[0063] Combine the elite individuals and the mutated offspring individuals to form a new generation population;

[0064] S320.7. Judgment of termination conditions:

[0065] If any of the following conditions is met, terminate the algorithm and output the optimal solution (the chromosome with the highest fitness):

[0066] The change in the optimal fitness of the population for G consecutive generations (G is the preset stagnant generation number) is less than the threshold ε (ε is the preset convergence threshold);

[0067] The number of iterations k reaches the preset maximum number of iterations K.

[0068] As a further improvement of this technical solution, the main distribution network spatio-temporal coordination unit includes a model splicing module, a topology analysis module, and a time window adjustment module, where:

[0069] The model splicing module is used to splice the main network and distribution network CIM models to generate a fusion topology map of all voltage levels. By parsing the main network and distribution network CIM models, it extracts equipment, connection relationships and electrical parameters, establishes entity mapping (such as mapping the main network "substation" to the distribution network "bus") and attribute mapping of the main network and distribution network equipment to handle differences in naming rules and topology granularity, and finally merges the main network and distribution network equipment nodes and connection edges to construct a fusion topology map containing electrical attributes and spatial location information.

[0070] The topology analysis module analyzes the impact of main grid maintenance on distribution network power supply paths based on the fused topology graph. By marking the main grid maintenance equipment and updating the topology connection status in the fused topology graph, it identifies distribution network equipment and power supply paths that are electrically connected to the main grid maintenance equipment or affected by changes in power supply paths. It calculates the number of distribution network power supply path interruptions caused by main grid maintenance, the total amount of load affected, the number of users, and the load rate of backup paths (such as the load percentage of backup transformers).

[0071] The time window adjustment module adjusts the main network maintenance time window based on the topology analysis results of the topology analysis module. It identifies the load trough period based on the historical load curve of the distribution network, determines the period with sufficient reserve resources by combining the status of the distribution network equipment, and selects the window that covers both the load trough period and the period with sufficient reserve resources from the candidate maintenance time windows of the main network. If there are multiple windows that meet the conditions, the window with the smallest distribution network load impact index is selected as the final maintenance time window.

[0072] As a further improvement to this technical solution, the topology analysis module adopts a cross-voltage level topology tracing method when identifying the impact of main grid maintenance on the distribution network power supply path, specifically including:

[0073] Starting from the electrical connection point of the equipment under maintenance in the main grid, traverse the integrated topology across voltage levels (e.g., trace from the 500kV equipment in the main grid to the 10kV and 0.4kV equipment in the distribution network);

[0074] Prioritize identifying ring network switching paths on the distribution network side (such as switchable backup feeders of ring network cabinets), and determine the priority of load transfer based on the real-time status (closed / open) of the distribution network sectionalizing switches (prioritize paths with switches in the "operable" state).

[0075] As a further improvement to this technical solution, the spatiotemporal risk assessment unit includes, wherein:

[0076] The fault range analysis module is based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit. It constructs a whole power grid model that includes the electrical connection relationship and spatial distribution of equipment. By simulating a single equipment outage scenario (N-1 scenario), it identifies the power loss range and extracts the number of distribution transformer areas, the number of affected users, the total power loss load, and the load importance level (such as the proportion of first-level, second-level, and third-level loads) within the power loss area.

[0077] The fault probability prediction module combines the health time series of the evaluated equipment, collects its historical health data (including aging degree, defect records, and online monitoring abnormal data), constructs a multi-dimensional health assessment system, and outputs the fault probability prediction value of the evaluated equipment within a preset time window based on data time correlation analysis.

[0078] The risk degradation scheme generation module quantifies equipment risk based on the failure probability and the scope of power outage, through a quantitative correlation between the failure probability and the impact of power outage load (the impact of power outage load is calculated by combining the total power outage load and the load importance coefficient), prioritizes equipment according to risk value, and generates differentiated risk management strategies (high-risk equipment triggers maintenance plan adjustments, medium-risk equipment enhances monitoring, and low-risk equipment sets early warnings).

[0079] The second objective of this invention is to provide an application method for an intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics. Based on the aforementioned intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics, the method includes the following steps:

[0080] S100. The power grid CIM model is parsed into a graph database structure containing spatial topology relationships and time dimension attributes, and a spatiotemporally related device network model is generated through topology pruning optimization.

[0081] S200: Extract equipment spatial distribution features, load time series features, and maintenance resource time constraint features from power grid operation data, construct multi-dimensional feature vectors, and implement feature extraction based on graph traversal algorithm and LSTM neural network;

[0082] S300: Based on the generated spatiotemporal knowledge graph and multidimensional feature vectors, an improved genetic algorithm is used to generate a maintenance schedule arrangement scheme that takes into account both spatial topological constraints and time resource balance.

[0083] S400, splicing the main grid and distribution network CIM models to generate a fusion topology map of all voltage levels, and based on the topology analysis, analyzing the impact of main grid maintenance on distribution network power supply paths and adjusting maintenance time windows;

[0084] S500 identifies the power loss range of N-1 faults based on spatiotemporal knowledge graphs, and predicts the fault probability by combining equipment health time series and generates risk degradation schemes.

[0085] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0086] 1. This invention uses a spatiotemporal coordination unit to stitch together the CIM models of the main grid and distribution network, generating a fused topology map across all voltage levels. Based on a cross-voltage level topology tracing method, it can automatically identify the impact range of main grid maintenance on the power supply path of the distribution network. This mechanism changes the traditional independent analysis mode of the main grid and distribution network, realizing topology linkage analysis. It upgrades the coordination of main grid and distribution network maintenance from manual experience to topology-driven intelligent assessment, effectively reducing the problems of repeated power outages or power transfer path conflicts caused by insufficient coordination of the main grid and distribution network.

[0087] 2. This invention utilizes a spatiotemporal knowledge graph construction unit to parse the power grid CIM model into a graph database structure containing spatial topology and temporal dimension attributes, and integrates equipment spatial distribution characteristics and load temporal characteristics through a spatiotemporal feature extraction unit. This fusion mechanism forms a multi-dimensional feature vector, enabling maintenance decisions to simultaneously consider spatial topology constraints and temporal resource balance, overcoming the limitations of traditional single-dimensional analysis and improving the matching degree between maintenance plan scheduling and the actual operating state of the power grid;

[0088] 3. This invention constructs a whole-power grid model based on a spatiotemporal risk assessment unit, identifies the power outage range by simulating an N-1 fault scenario, and predicts the fault probability by combining equipment health time series data. The system can quantify the number of distribution transformer areas, the number of affected users, and the load importance level within the power outage area, generating differentiated risk management strategies. This mechanism realizes an upgrade from static risk assessment to dynamic simulation, and assists decision-making through visualized risk analysis. It changes the problems of low risk identification efficiency and delayed generation of degradation solutions in traditional technologies, providing real-time risk quantification support for maintenance decisions.

[0089] 4. The intelligent decision-making unit of this invention generates maintenance plans through an improved genetic algorithm, achieving multi-objective optimization between spatial topology constraints and time resource balance, thereby improving the feasibility of plan scheduling; the topology pruning optimization module optimizes the spatial topology network, reduces redundant connections, improves the storage and computing efficiency of the graph database, and ensures the real-time performance of large-scale power grid topology analysis. Attached Figure Description

[0090] Figure 1 This is a system framework diagram of the present invention;

[0091] Figure 2 This is a schematic diagram of the method steps of the present invention;

[0092] The meanings of the labels in the diagram are as follows:

[0093] 100. Spatiotemporal knowledge graph construction unit; 110. Spatial topology modeling module; 120. Temporal attribute annotation module; 130. Topology pruning and optimization module; 140. Graph database storage module;

[0094] 200. Spatiotemporal feature extraction unit; 210. Spatial feature extraction module; 220. Temporal feature extraction module; 230. Feature fusion module;

[0095] 300. Intelligent Decision-Making Unit; 310. Constraint Analysis Module; 320. Genetic Algorithm Optimization Module; 330. Solution Evaluation Module;

[0096] 400. Spatiotemporal Coordination Unit for Main and Distribution Networks; 410. Model Assembly Module; 420. Topology Analysis Module; 430. Time Window Adjustment Module;

[0097] 500. Spatiotemporal risk assessment unit; 510. Fault range analysis module; 520. Fault probability prediction module; 530. Risk degradation scheme generation module. Detailed Implementation

[0098] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0099] Example 1

[0100] like Figure 1 As shown, this embodiment provides an intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics, including:

[0101] The spatiotemporal knowledge graph construction unit 100 is used to parse the power grid CIM model into a graph database structure containing spatial topological relationships and temporal dimension attributes, and to generate a spatiotemporally related device network model through topology pruning optimization.

[0102] In this embodiment, the spatiotemporal knowledge graph construction unit 100 includes a spatial topology modeling module 110 and a temporal attribute annotation module 120, wherein:

[0103] The spatial topology modeling module 110 is used to parse the substations, lines, transformers, circuit breakers, disconnectors, and busbars in the power grid CIM model into graph nodes, and construct a spatial topology network with electrical connection relationships as edges. The graph nodes contain attributes such as equipment type, voltage level, and physical location coordinates. The spatial topology network supports hierarchical display based on voltage level, and the edges between nodes contain electrical parameters such as impedance value and line length.

[0104] The time attribute annotation module 120 is used to annotate the time dimension attributes (such as maintenance cycle, historical fault time series, last maintenance timestamp, etc.) of equipment nodes and form an equipment-time association table.

[0105] In this embodiment, the spatiotemporal knowledge graph construction unit 100 further includes a topology pruning optimization module 130 and a graph database storage module 140, wherein:

[0106] The topology pruning optimization module 130 generates and executes a topology pruning strategy based on the spatial topology network constructed by the spatial topology modeling module 110.

[0107] Furthermore, the topology pruning strategy specifically includes:

[0108] Endpoint identification rules: Identify endpoint nodes in a spatial topology network that are directly connected to connection points. An endpoint node is a device that is connected to other device nodes only through one edge.

[0109] Layered cropping strategy:

[0110] For equipment nodes with voltage levels of 220kV and above, the complete topology is retained without pruning.

[0111] For equipment nodes with voltage levels of 110kV and below, cut off the endpoint nodes that are directly connected to the connection point, and merge the electrical parameters (such as impedance values) of the cut-off endpoints into the remaining connection edges.

[0112] Core path retention mechanism: Based on the importance weight of equipment nodes in the spatial topology network, the topology path connected to core equipment nodes such as main transformers and busbars is retained, and the importance weight is determined by the load impact degree and power supply range level of the equipment.

[0113] The graph database storage module 140 uses a graph database to store spatiotemporal knowledge graphs.

[0114] Furthermore, the storage structure is as follows:

[0115] Equipment node: includes equipment identifier, name, type, voltage level, physical location and time attribute, where the time attribute includes maintenance cycle and historical fault time information;

[0116] Connection edge: Includes connection endpoint identification, connection relationship type and electrical parameters, including impedance value, line length, current carrying capacity and related parameters.

[0117] As a further explanation of this embodiment, the Jena toolkit can be used to parse the OWL format CIM model, and DOM4J can be used to traverse the XML tree structure to extract device nodes and connection relationships; the specific process is as follows:

[0118] Load the CIM / XML file into memory and obtain the root node through DocumentElement;

[0119] Traverse the child nodes and extract attributes such as the type, voltage level, and physical location coordinates of equipment such as power plants and lines;

[0120] Establish a mapping between device nodes and electrical connections to generate the initial structure of the spatial topology network.

[0121] As a further explanation of this embodiment, the time attribute annotation module 120 in this embodiment annotates the time dimension attribute of the device node by the following steps:

[0122] First, obtain the maintenance cycle and the last maintenance timestamp from the equipment ledger system, and extract the historical fault time series from the SCADA system.

[0123] Then, use IBM SPSS's Define Date function to generate time series variables, converting time data into a computable timestamp format;

[0124] Finally, establish a relationship table between device nodes and time attributes in the graph database. For example, store the maintenance cycle (e.g., "maintenance cycle = 3 years") through Neo4j node attributes, and record the historical fault time series through relationship attributes.

[0125] The spatiotemporal feature extraction unit 200 is used to extract equipment spatial distribution features, load time series features, and maintenance resource time constraint features from power grid operation data, construct multi-dimensional feature vectors, and realize feature extraction based on graph traversal algorithm and LSTM neural network;

[0126] In this embodiment, the spatiotemporal feature extraction unit 200 includes a spatial feature extraction module 210, a temporal feature extraction module 220, and a feature fusion module 230, wherein:

[0127] The spatial feature extraction module 210 is used to extract the spatial distribution features of equipment from the power grid operation data. It uses a depth-first search algorithm to traverse the spatiotemporal knowledge graph and extract the features of equipment adjacency, power supply range hierarchy, and spatial topology connectivity. Then, based on the physical location coordinates of the equipment, it calculates the Euclidean distance and spatial clustering coefficient between the equipment to generate a spatial location feature vector.

[0128] As a further explanation of this embodiment, this embodiment uses Python's NetworkX library to construct a spatiotemporal knowledge graph and traverses the graph structure using a depth-first search algorithm. The specific process is as follows: First, the spatial topology network is queried through the Neo4j database interface, and the device nodes and their electrical connections are imported into the NetworkX graph model; then, the device adjacency relationships and spatial topology connectivity features are extracted using the depth-first search algorithm. For example, taking a 220kV substation as an example, the depth-first search traversal can identify its connection relationships with three 220kV lines and two main transformers, forming a set of adjacent devices. Simultaneously, based on the physical location coordinates (latitude and longitude) of the devices, this embodiment uses the Haversine formula to calculate the Euclidean distance between nodes, a formula applicable to distance measurements on the Earth's surface. The spatial clustering coefficient is calculated using a built-in NetworkX function, reflecting the density of the spatial distribution of devices.

[0129] The time feature extraction module 220 is used to extract load time series features and maintenance resource time constraint features from power grid operation data. It uses an LSTM neural network to train historical load data. The LSTM network contains two hidden layers with 64 neurons in each layer to capture the time dependence and periodic features of load data. It also extracts the available time window of maintenance resources and resource allocation time constraints to generate time constraint feature vectors.

[0130] As a further explanation of this embodiment, a two-layer LSTM model with hidden layers can be constructed using the TensorFlow framework. Each layer contains 64 neurons. The input layer receives historical load data with a 15-minute sampling interval, and the output layer is the load prediction value for the next 24 hours. The LSTM model uses the Adam optimizer with mean squared error (MSE) as the loss function. The training data is divided into training and test sets in a 7:3 ratio and processed using Min-Max normalization. Meanwhile, the extraction process for the available time window and resource allocation time constraints of maintenance resources in this embodiment includes: obtaining the available time window of resources from the maintenance work order system, parsing it into timestamp values ​​using the ISO8601 standard, and extracting features such as window length and the time required for resource allocation. For example, the available time window for a maintenance team is 7 hours, and the allocation time is 1 hour, forming a time constraint feature vector [7,1].

[0131] The feature fusion module 230 is used to fuse the spatial feature vector generated by the spatial feature extraction module 210 with the temporal feature vector generated by the temporal feature extraction module 220. The initial multidimensional feature vector is generated by concatenating the features according to the dimensions using the feature concatenation method. The initial multidimensional feature vector is then normalized using the Z-score normalization method to eliminate the difference in dimensions and generate the final multidimensional feature vector.

[0132] As a further explanation of this embodiment, the standardizer is first fitted using scikit-learn's StandardScaler to historical spatiotemporal feature data. This standardizer performs Z-score standardization on spatial features (such as Euclidean distance and clustering coefficients) and temporal features (such as load forecasts and maintenance time window lengths) to eliminate dimensional differences. Then, the feature vectors extracted in real time are transformed. Subsequently, the standardized spatial and temporal features are concatenated and stitched together in a dimensional order of "spatial features first, temporal features second" to form a multidimensional feature vector. The multidimensional feature vector is stored in a Parquet columnar storage format, partitioned by device type and date, and supplemented with metadata such as device ID and timestamp, supporting efficient reading and writing by the Apache Spark distributed computing framework. Finally, the stored multidimensional feature vector is directly input into the genetic algorithm optimization module 320 to provide spatiotemporal constraint input for the intelligent arrangement of maintenance time windows.

[0133] The intelligent decision-making unit 300, based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit 100 and the multidimensional feature vector constructed by the spatiotemporal feature extraction unit 200, generates a maintenance plan scheduling scheme that takes into account both spatial topological constraints and time resource balance through a genetic algorithm.

[0134] In this embodiment, the intelligent decision-making unit 300 includes a constraint parsing module 310, a genetic algorithm optimization module 320, and a scheme evaluation module 330, wherein:

[0135] The constraint parsing module 310 is used to parse the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit 100 and the multidimensional feature vector generated by the spatiotemporal feature extraction unit 200, and extract the constraint conditions required for maintenance plan arrangement.

[0136] Furthermore, the constraint resolution module 310 extracts spatial topological constraints and temporal resource constraints, specifically including:

[0137] Spatial dependency constraint identification:

[0138] Based on the adjacency relationships of devices in the spatiotemporal knowledge graph (such as the connection relationship between circuit breakers and disconnect switches) and the power supply range hierarchy (such as the power supply area of ​​a 110kV line), the spatial correlation of maintenance operations is analyzed. For example, when device A is under maintenance, if adjacent device B has a direct electrical connection with A (such as sharing a busbar), then electrical isolation operations must be performed simultaneously (spatial dependency constraint).

[0139] Time resource balance constraint identification:

[0140] By combining the equipment time attributes (maintenance cycle, last maintenance timestamp) with the maintenance resource time constraint features output by the spatiotemporal feature extraction unit (such as the available time window of maintenance teams and the dispatch cycle of work vehicles), the upper limit of the number of maintenance personnel and the upper limit of equipment dispatch capacity that can be deployed within the same time window (such as 8:00-18:00 on a single day) are calculated (time resource balance constraint).

[0141] As a further explanation of this embodiment, the spatial topology constraint extraction process in this embodiment includes: querying the device adjacency relationship through the graph database, specifically: using Cypher statements to extract the set of adjacent devices of the target device from Neo4j, for example, querying the disconnecting switch or line directly electrically connected to a certain circuit breaker; the power supply range level is automatically identified according to the device voltage level, with 220kV and above devices defined as Level 1 main grid level, and 110kV and below devices defined as Level 2 distribution network level, and the level division is achieved by matching the voltage level field.

[0142] As a further explanation of this embodiment, the parsing logic of time resource constraints in this embodiment includes: obtaining the available time window of the work team from the maintenance resource management system and converting the time string into a timestamp interval; the upper limit of the time window capacity is determined according to the work team's personnel configuration and equipment allocation capability. For example, if a work team is configured with 5 maintenance personnel, the maximum number of parallel maintenance tasks per day is set to 3 based on historical work data.

[0143] The genetic algorithm optimization module 320 generates a maintenance plan arrangement scheme based on the spatiotemporal constraints extracted by the constraint parsing module 310 through an improved genetic algorithm. The improved genetic algorithm integrates an elite retention strategy, a dynamic parameter adjustment mechanism, and a dual-objective fitness function.

[0144] The scheme evaluation module 330 is used to evaluate the maintenance plan arrangement scheme generated by the genetic algorithm optimization module 320 in multiple indicators. If the evaluation result does not meet the preset engineering threshold, the genetic algorithm optimization module 320 is triggered to iterate and optimize again.

[0145] As a further explanation of this embodiment, the specific logic of the multi-index evaluation in this embodiment is as follows:

[0146] Spatial constraint violation rate calculation: Extract the electrical connection topology of the equipment (such as the association between circuit breakers and adjacent disconnect switches) from the spatiotemporal knowledge graph construction unit 100, traverse the time window (start time, duration) of each equipment in the maintenance plan, and count the number of conflict pairs with overlapping maintenance times of adjacent equipment (if the difference between the start times of maintenance of two equipment is less than the duration, it is judged as a conflict); compare the number of conflict pairs with the total number of adjacent equipment pairs to form the violation rate (preset threshold ≤15%).

[0147] Time resource balance calculation: Based on the maintenance resource availability time period (such as shift attendance time, vehicle dispatch cycle) parsed by constraint parsing module 310, the resource occupancy (number of personnel, number of vehicles) of each time period of the maintenance plan is statistically analyzed by hourly slice; the ratio of the standard deviation of resource occupancy to the mean is calculated, and the balance is measured by 1-this ratio (preset threshold ≥0.7, the smaller the ratio, the smoother the resource allocation).

[0148] Power outage impact index calculation: Obtain the list of affected users (including number of users and power supply area) corresponding to the equipment under maintenance from the spatiotemporal risk assessment unit 500. Combine the power outage duration (end time - start time) of the maintenance plan, and calculate the impact index by multiplying the number of affected users by the power outage duration (the threshold is set according to the power grid load density, such as ≤1000 users per hour for urban power grids).

[0149] As a further explanation of this embodiment, if the evaluation result does not meet any of the indicator thresholds, the genetic algorithm optimization module 320 is immediately triggered to iterate again: the evaluation result is fed back to the population initialization stage, and the chromosome encoding rules are dynamically adjusted (such as narrowing the random range of the time window) or the optimization parameters are optimized (such as temporarily increasing the mutation probability). For example, when the spatial constraint violation rate exceeds 15%, the genetic algorithm prioritizes adjusting the maintenance time difference between adjacent devices, forcibly widening the time window of conflicting devices, recalculating the fitness and iterating until a maintenance plan that meets all indicators is generated.

[0150] In this embodiment, the genetic algorithm optimization module 320 generates a maintenance plan scheduling scheme through an improved genetic algorithm, including the following steps:

[0151] S320.1 Population Initialization:

[0152] Integer encoding is used, and each chromosome contains N gene loci, equal to the total number of devices to be repaired. Each gene locus is g. i The gene value is an integer in the range [0, T] corresponding to the start time of maintenance of the i-th device (T is the preset total length of the time window); M chromosomes of the preset population size are randomly generated to form the initial population;

[0153] S320.2 Fitness Calculation:

[0154] For each chromosome, calculate the bi-objective fitness function F based on its encoded maintenance time allocation:

[0155] F = ω1·(1-V) + ω2·(1-B);

[0156] Where V represents the degree of violation of spatial topological constraints, and the calculation formula is: Among them, Ω i Let I(g) be the set of neighboring devices of device i (extracted from the spatiotemporal knowledge graph);i ,g j ) is the conflict indication function: if the maintenance time windows of devices i and j overlap, then I(g) i ,g i ) = 1, otherwise I(g) i ,g j ) = 0; B is the time resource balance degree, calculated by the following formula: Among them, R t The amount of maintenance resources allocated within the time window t, σ(R1,R2,...,R T R represents the maintenance resource quantity within the time window t. t The standard deviation of the sequence; R represents the maintenance resource quantity within the time window t. t The average value; ω1 and ω2 are weight coefficients, satisfying ω1+ω2=1, which respectively represent the priority of spatial topological constraints and time resource balance;

[0157] As a further explanation of this embodiment, the spatial topology constraint violation degree V is based on the electrical connection relationship of the devices constructed by the spatiotemporal knowledge graph construction unit 100. It traverses the adjacent device pairs (such as circuit breakers and associated disconnect switches) of the device to be repaired: if the repair time windows (start time and duration) of the two devices overlap, it is determined to be a time conflict; the number of all conflict pairs is counted and compared with the total number of adjacent device pairs to obtain the violation degree spatial topology constraint violation degree V.

[0158] As a further explanation of this embodiment, the time resource balance B relies on the resource availability time period data of the constraint parsing module 310, slices the time axis by hour, and counts the maintenance resource occupancy (such as the number of personnel and vehicles) in each slice; calculates the ratio of the standard deviation of these occupancy to the mean. The smaller the ratio, the more evenly the resources are distributed in the time dimension, that is, the smaller the balance B.

[0159] Furthermore, in this embodiment, the intelligent decision-making unit 300 optimizes the maintenance plan using a genetic algorithm, with the key parameters defined as follows:

[0160] In the bi-objective fitness function F, the weights ω1 and ω2 for spatial constraints and resource balancing are dynamically adjusted according to the number of devices:

[0161]

[0162] Where, n 关键 The number of critical equipment involved in the maintenance; N 总 The total number of equipment involved in the maintenance; constrained to 0.3≤ω1≤0.7 to avoid extreme weighting that could lead to plan imbalance;

[0163] If the resource idle rate is >30% or the space conflict rate is >20%, the weights will be automatically adjusted.

[0164] When the resource idle rate is high, increase ω2;

[0165] When the space conflict rate is high, increase ω1;

[0166] Recalculate the fitness after adjustment until the plan converges.

[0167] S320.3, Elite retention strategy:

[0168] Select the top E individuals (E is the preset elite number and E < M) with the highest fitness from the current population, and directly copy them to the next-generation population to ensure that the optimal solution is not destroyed by genetic operations;

[0169] S320.4, Dynamic parameter adjustment:

[0170] Crossover probability P c and mutation probability P m are dynamically adjusted with the iteration number k:

[0171] Crossover probability P c : linearly decreases from the initial value P c0 to the minimum threshold P c1 , and the formula is:

[0172]

[0173] [[ID=�9]]where K is the preset maximum iteration number;

[0174] Mutation probability P m : linearly increases from the initial value P m0 to the maximum threshold P m1 ,公式为:

[0175]

[0176] where K is the preset maximum iteration number;

[0177] S320.5, Genetic operation execution:

[0178] Selection: Use the tournament selection method. Each time, randomly select T s individuals (T s is the preset tournament scale) from the population, and select the individual with the highest fitness to enter the mating pool. Repeat the operation until the size of the mating pool reaches M - E;

[0179] Crossover: Randomly pair the individuals in the mating pool according to P c and generate offspring using two-point crossover:

[0180] Randomly select two crossover points p1 and p2 (1 ≤ p1 < p2 ≤ N), and exchange the gene segments of the parental chromosome in the interval [p1, p2];

[0181] Mutation: For the offspring chromosome according to P m Randomly select a gene position and replace its value with a random integer within the range of [0, T];

[0182] S320.6, Construction of the new generation population:

[0183] Merge the elite individuals and the mutated offspring individuals to form a new generation population;

[0184] S320.7, Judgment of termination conditions:

[0185] If any of the following conditions is met, terminate the algorithm and output the optimal solution (the chromosome with the highest fitness):

[0186] The change in the optimal fitness of the population for consecutive G generations (G is the preset stagnation generation number) is less than the threshold ε (ε is the preset convergence threshold);

[0187] The iteration number k reaches the preset maximum iteration number K.

[0188] The main and distribution network spatio-temporal coordination unit 400 is used to splice the main network and distribution network CIM models to generate a full voltage level fusion topology map, analyze the impact of main network maintenance on the distribution network power supply path based on the topology, and adjust the maintenance time window;

[0189] In this embodiment, the main and distribution network spatio-temporal coordination unit 400 includes a model splicing module 410, a topology analysis module 420, and a time window adjustment module 430, where:

[0190] The model splicing module 410 is used to splice the main network and distribution network CIM models to generate a full voltage level fusion topology map. By analyzing the main network and distribution network CIM models, it extracts equipment, connection relationships, and electrical parameters, establishes entity mappings (such as the main network's "substation" mapping to the distribution network's "bus") and attribute mappings of the main network and distribution network equipment to handle naming rule and topology granularity differences, and finally merges the main and distribution network equipment nodes and connection edges to construct a fusion topology map containing electrical attributes and spatial location information;

[0191] As a further explanation of this embodiment, the parsing of the CIM model in this embodiment relies on the CIM / XML file parsing specification (following the Q / GDW1517 standard) to extract the main and distribution network equipment, connection relationships, electrical parameters and spatial location data. Then, the correspondence between the main and distribution network equipment and attributes is established through pre-configured mapping rules. The topology of the main network "substation-busbar-line" and the distribution network "feeder-switch-transformer area" is merged. The distribution network feeder is regarded as the sub-node of the main network line. A fused topology map containing electrical attributes and spatial location information is constructed and stored in a graph database (such as Neo4j).

[0192] The topology analysis module 420 analyzes the impact of main grid maintenance on distribution network power supply paths based on the fused topology graph. By marking the main grid maintenance equipment and updating the topology connection status in the fused topology graph, it identifies distribution network equipment and power supply paths that are electrically connected to the main grid maintenance equipment or affected by power supply path changes. It calculates the number of distribution network power supply path interruptions caused by the main grid maintenance, the total affected load, the number of users, and the standby path load rate (e.g., standby transformer load percentage). When identifying the impact of main grid maintenance on distribution network power supply paths, the topology analysis module 420 employs a cross-voltage level topology tracing method, specifically including:

[0193] Starting from the electrical connection point of the equipment under maintenance in the main grid, traverse the integrated topology across voltage levels (e.g., trace from the 500kV equipment in the main grid to the 10kV and 0.4kV equipment in the distribution network);

[0194] Prioritize identifying ring network switching paths on the distribution network side (such as switchable backup feeders of ring network cabinets), and determine the priority of load transfer based on the real-time status (closed / open) of the distribution network sectionalizing switches (prioritize paths with switches in the "operable" state).

[0195] As a further explanation of this embodiment, the specific process of the topology analysis module 420 implementing cross-voltage influence tracking in this embodiment is as follows:

[0196] First, starting from the main grid maintenance equipment, subscribe to SCADA system data through the MQTT protocol to obtain the closed / open status and "operable" mark of the distribution network sectional switches. At the same time, query the real-time load of the distribution network equipment affected by the main grid maintenance from the power consumption information collection system to complete the data preparation.

[0197] Then, starting with the main grid maintenance equipment (such as 220kV circuit breakers), a breadth-first search (BFS) is used to traverse the fusion topology map, identify associated distribution network equipment across voltage levels (from the high voltage level of the main grid to the low voltage level of the distribution network, such as 500kV-220kV-110kV-10kV-0.4kV), prioritize tracing the switching path of the distribution network ring network, and determine the load transfer priority by combining the real-time status of the distribution network sectional switches (closed / open, operable / inoperable).

[0198] Next, based on the tracking results, the number of power supply path interruptions in the distribution network is counted, the total affected load is calculated by summing the real-time load values ​​of the affected distribution network equipment, the number of affected users is counted by combining the power supply range of the distribution network equipment, and the rated capacity (obtained from the equipment ledger) and real-time load (obtained from the SCADA system) of the distribution network standby equipment are queried. The standby path load rate is calculated by "(current load of standby path ÷ rated capacity of standby equipment) × 100%", thus completing the analysis and calculation of the impact of main grid maintenance on the power supply path of the distribution network.

[0199] The time window adjustment module 430 adjusts the main network maintenance time window based on the topology analysis results of the topology analysis module 420. It identifies the load trough period based on the historical load curve of the distribution network, determines the period with sufficient reserve resources by combining the status of the distribution network equipment, and selects the window that covers both the load trough period and the period with sufficient reserve resources from the candidate maintenance time windows of the main network. If there are multiple windows that meet the conditions, the window with the smallest distribution network load impact index is selected as the final maintenance time window.

[0200] As a further explanation of this embodiment, the specific process of window optimization performed by the time window adjustment module 430 is as follows:

[0201] First, retrieve the historical load curves of the distribution network equipment for the past year from storage systems such as InfluxDB time series database. Then, using data analysis tools such as Pandas, identify the load trough intervals within 24 hours that match the duration of the main grid maintenance (e.g., if the main grid maintenance lasts for 4 hours, then filter for consecutive 4 hours). The load in this interval is significantly lower than the daily average load (e.g., more than 30% lower).

[0202] Then, query the equipment ledger of the distribution network backup equipment (backup transformers, backup feeders, etc.) to obtain its rated capacity, and combine it with the real-time load data of the SCADA system to calculate "remaining capacity = rated capacity - real-time load". The time period with remaining capacity ≥ 0 is determined as the time period with sufficient backup resources.

[0203] Next, the intersection of the candidate maintenance time windows of the main network (derived from the maintenance plan) with the identified "off-peak load periods" and "periods with sufficient backup resources" is used to filter out feasible windows that meet the conditions simultaneously.

[0204] Finally, if multiple feasible windows exist, calculate the "distribution network load impact index (affected load × duration)" for each window, and select the window with the smallest index as the final main network maintenance time window.

[0205] The spatiotemporal risk assessment unit 500 identifies the power loss range of N-1 faults based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit 100, and predicts the fault probability by combining the equipment health time series and generates a risk degradation scheme.

[0206] In this embodiment, the spatiotemporal risk assessment unit 500 includes, wherein:

[0207] The fault range analysis module 510 is based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit 100. It constructs a whole power grid model that includes the electrical connection relationship and spatial distribution of equipment. By simulating a single equipment outage scenario (N-1 scenario), it identifies the power loss range and extracts the number of distribution transformer areas, the number of affected users, the total power loss load and the load importance level (such as the proportion of first-level, second-level and third-level loads) in the power loss area.

[0208] As a further explanation of this embodiment, when constructing a full power grid model that includes the electrical connection relationships and spatial distribution of equipment, firstly, the electrical connection relationships (such as the connection between busbars and lines, and the association between transformers and distribution areas) and spatial distribution information (equipment latitude and longitude, power supply area boundaries) of the equipment are extracted from the spatiotemporal knowledge graph construction unit 100; secondly, relying on a graph database (such as Neo4j), the equipment is abstracted as nodes (marked with voltage level and equipment type), and the electrical connection is abstracted as edges (marked with parameters such as impedance and current carrying capacity), and a full voltage level topology model covering the main grid and distribution network is constructed.

[0209] As a further explanation of this embodiment, after constructing the complete power grid topology model, in order to analyze the cascading impact of a single device failure on the power grid supply, it is necessary to simulate equipment outage scenarios to locate the power loss range. Specifically, the power loss area identification for the N-1 scenario is performed through the following steps: First, traverse the key equipment of the power grid (lines, transformers, circuit breakers, etc.) and perform a "virtual outage" operation on each device in sequence; Second, starting from the out-of-service equipment, use the breadth-first search (BFS) algorithm to trace the affected power supply path across voltage levels (e.g., from the outage of the 220kV main grid line, trace layer by layer to the 10kV feeder and transformer substation of the distribution network, and mark the power loss nodes); Finally, filter the nodes that have been restored to power through the backup path to determine the final power loss area.

[0210] Furthermore, after completing the construction of the entire power grid topology model and the N-1 scenario simulation, and identifying the power outage area, it is necessary to accurately statistically analyze the various impacts caused by the power outage in order to provide data support for subsequent risk assessment and strategy formulation. The specific statistical process for power outage impact indicators is as follows: First, connect to the power consumption information system and count the number of affected users according to the distribution transformer areas within the power outage area (accumulate the number of power supply users in each transformer area); second, call the SCADA real-time load data, accumulate the load value of the power outage equipment, and calculate the total power outage load; finally, extract the load type (level 1, level 2, level 3) from the equipment ledger, and statistically analyze the proportion of load importance level (such as the percentage of level 1 load in the total power outage), and output indicators such as the number of distribution transformer areas, the number of users, the total load, and the proportion of level.

[0211] The fault probability prediction module 520 combines the health time series of the evaluated equipment, collects its historical health data (including aging degree, defect records, and abnormal data from online monitoring), constructs a multi-dimensional health assessment system, and outputs the fault probability prediction value of the evaluated equipment within a preset time window based on data time correlation analysis.

[0212] As a further explanation of this embodiment, the fault probability prediction module 520 in this embodiment is used for time-series modeling of equipment health. Its design purpose is to predict the probability of failure within a future time window by combining historical equipment health data. The fault probability prediction module 520 collects multi-dimensional health data from online monitoring platforms such as the Equipment Management System (PMS), partial discharge monitoring, and oil chromatography monitoring. This data covers equipment aging (e.g., transformer commissioning years, insulation paper polymerization degree), defect records (including historical defect types, processing status, and number of repeated defects), and online monitoring anomalies (e.g., the number of times partial discharge exceeds the standard, and abnormal trends in oil chromatography data). Based on this, an LSTM neural network is used to construct a health time-series model. The input layer concatenates features such as equipment aging, defect frequency, and monitoring anomalies statistically analyzed weekly / monthly. A two-layer LSTM (64 neurons per layer) learns the degradation pattern of health over time. The output layer uses a Sigmoid function to map and obtain the fault probability within a future time window T (e.g., 7 days, 30 days) (outputting a probability value in the 0-1 range), providing crucial data support for predicting equipment failure probability and subsequent risk assessment.

[0213] Specifically, when constructing a multi-dimensional health assessment system, a quantitative assessment system is built that includes aging level, defect records, and online monitoring anomalies. The formula for calculating health level H is:

[0214] H = 0.3 × A + 0.5 × D + 0.2 × M;

[0215] Among them, A is the aging degree score (such as the transformer's years of operation / design life, normalized to 0-1); D is the defect record score (historical defect count and severity weighted, normalized to 0-1); M is the online monitoring anomaly score (number of partial discharge exceedances and abnormal trends in oil chromatography, normalized to 0-1); the weights of 0.3, 0.5, and 0.2 refer to the DL / T596 industry standard and can be adjusted as needed.

[0216] The risk degradation scheme generation module 530 quantifies equipment risk based on the failure probability and the scope of power outage, through a quantitative correlation between the failure probability and the impact of power outage load (the impact of power outage load is calculated by combining the total power outage load and the load importance coefficient), prioritizes equipment according to risk value, and generates differentiated risk management strategies (high-risk equipment triggers maintenance plan adjustments, medium-risk equipment enhances monitoring, and low-risk equipment sets early warnings).

[0217] As a further explanation of this embodiment, the risk degradation scheme generation module 530 in this embodiment generates a control scheme adapted to equipment with different risk levels based on the power outage impact identified by the fault range analysis module 510 and the equipment failure probability output by the fault probability prediction module 520, through risk quantification modeling and hierarchical strategy formulation. The specific implementation process is as follows:

[0218] First, to accurately measure the threat posed by equipment failure to the power grid, the equipment risk value R is defined as a comprehensive mapping between the failure probability and the impact of power outage, as shown in the formula:

[0219] R = P × I;

[0220] Where P is the equipment failure probability output by the failure probability prediction module 520; I is the power loss load impact degree, which needs to be corrected by combining the total power loss load and the load importance. The formula is expanded as: I = L × K, where L is the total power loss load counted by the failure range analysis module 510, and K is the load importance coefficient.

[0221] Then, based on the magnitude of the risk value R, the equipment is divided into three risk levels: high, medium, and low (the threshold can be adjusted according to the actual operation requirements of the power grid):

[0222] High risk: R≥0.8 (both the probability of failure and the impact of power loss are significant, requiring emergency intervention);

[0223] Medium risk: 0.3≤R<0.8 (the possibility or impact of failure is relatively controllable, but monitoring needs to be strengthened);

[0224] Low risk: R<0.3 (low probability of failure, small impact, mainly based on early warning and prevention);

[0225] Finally, the following tiered disposal strategy was developed for equipment with different risk levels to ensure precise resource allocation:

[0226] For high-risk equipment, prioritize ensuring power supply safety and immediately adjust maintenance plans or activate backup equipment; specific operations include:

[0227] Trigger the emergency maintenance process: Prioritize equipment failure repair to the highest level and insert it into the current maintenance schedule;

[0228] Activate backup equipment: If the main equipment outage has a significant impact, switch to backup transformers, backup lines, etc., to temporarily replace and ensure power supply;

[0229] For medium-risk equipment, increase monitoring frequency to detect early signs of malfunction. Specific operations include:

[0230] Encrypted online monitoring: The frequency of data collection for partial discharge, oil chromatography, vibration, and other monitoring data has been increased from "once a day" to "once every 2 hours";

[0231] Increase manual inspections: Arrange maintenance personnel to conduct daily on-site inspections, focusing on checking for abnormalities such as equipment overheating, unusual noises, and leaks;

[0232] For low-risk devices, set early warning thresholds and monitor the trend of health degradation in real time. Specific operations include:

[0233] Define health status warning indicators: such as triggering an early warning when the degree of polymerization of transformer insulation paper is lower than 500, or triggering an early warning when the number of mechanical operations of a circuit breaker exceeds the rated value by 80%;

[0234] Connect to the smart grid early warning platform: Automatically push early warning information when indicators are abnormal to remind operation and maintenance personnel to pay attention.

[0235] Specifically, in this embodiment, the spatiotemporal knowledge graph construction unit 100, spatiotemporal feature extraction unit 200, intelligent decision-making unit 300, main and distribution network spatiotemporal coordination unit 400, and spatiotemporal risk assessment unit 500 form a closed-loop system for power grid maintenance decision-making through data interaction and functional collaboration. The specific interaction logic is as follows: In this embodiment, the spatiotemporal knowledge graph construction unit 100 generates a spatiotemporal knowledge graph containing electrical connection relationships and spatial distribution information of equipment. This provides spatial topology constraint data for the constraint parsing module 310 of the intelligent decision-making unit 300, provides the basic topology for the main and distribution network CIM model fusion of the model splicing module 410 of the main and distribution network spatiotemporal coordination unit 400, and also provides electrical correlation data for N-1 fault simulation for the fault range analysis module 510 of the spatiotemporal risk assessment unit 500; the spatiotemporal feature extraction unit 200 extracts multi-dimensional features such as equipment maintenance cycle and resource availability time, inputs them into the constraint parsing module 310 of the intelligent decision-making unit 300, and parses the time resource balance constraints; the intelligent decision-making unit 300... Based on the aforementioned constraints, the genetic algorithm optimization module 320 generates a maintenance plan, which is then evaluated by the scheme evaluation module 330 using multiple indicators. If the threshold is not met, iterative optimization is triggered. The main and distribution network spatiotemporal coordination unit 400 integrates the main and distribution network topology through the model splicing module 410. The topology analysis module 420 analyzes the impact of main network maintenance on the distribution network power supply path based on the integrated topology, outputting the power outage range and load data, which are fed back to the intelligent decision-making unit 300 to adjust the maintenance time window. The fault probability prediction module 520 of the spatiotemporal risk assessment unit 500 combines the equipment health data of the spatiotemporal knowledge graph construction unit 100 to predict the fault probability. After the risk degradation scheme generation module 530 quantifies the risk, the high-risk equipment information is fed back to the intelligent decision-making unit 300, triggering the adjustment of the maintenance plan. Through data flow and functional collaboration of the spatiotemporal knowledge graph, each unit forms a closed loop of "topology construction - constraint resolution - plan generation - collaborative verification - risk feedback," ensuring that maintenance decisions meet the needs of power grid safety, resource balance, and risk prevention and control.

[0236] Example 2

[0237] like Figure 2 As shown, this embodiment also provides an application method for an intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics. Based on the aforementioned intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics, the method includes the following steps:

[0238] S100. The power grid CIM model is parsed into a graph database structure containing spatial topology relationships and time dimension attributes, and a spatiotemporally related device network model is generated through topology pruning optimization.

[0239] S200: Extract equipment spatial distribution features, load time series features, and maintenance resource time constraint features from power grid operation data, construct multi-dimensional feature vectors, and implement feature extraction based on graph traversal algorithm and LSTM neural network;

[0240] S300: Based on the generated spatiotemporal knowledge graph and multidimensional feature vectors, an improved genetic algorithm is used to generate a maintenance schedule arrangement scheme that takes into account both spatial topological constraints and time resource balance.

[0241] S400, splicing the main grid and distribution network CIM models to generate a fusion topology map of all voltage levels, and based on the topology analysis, analyzing the impact of main grid maintenance on distribution network power supply paths and adjusting maintenance time windows;

[0242] S500 identifies the power loss range of N-1 faults based on spatiotemporal knowledge graphs, and predicts the fault probability by combining equipment health time series and generates risk degradation schemes.

[0243] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0244] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics, characterized in that, include: The spatiotemporal knowledge graph construction unit (100) is used to parse the power grid CIM model into a graph database structure containing spatial topological relationships and time dimension attributes, and to generate a spatiotemporally related device network model through topology pruning optimization. The spatiotemporal feature extraction unit (200) is used to extract equipment spatial distribution features, load time series features and maintenance resource time constraint features from power grid operation data, construct multi-dimensional feature vectors, and realize feature extraction based on graph traversal algorithm and LSTM neural network; The intelligent decision-making unit (300) generates a maintenance plan arrangement scheme that takes into account both spatial topological constraints and time resource balance based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit (100) and the multidimensional feature vector constructed by the spatiotemporal feature extraction unit (200). The main and distribution network spatiotemporal coordination unit (400) is used to splice the CIM models of the main network and the distribution network to generate a fusion topology map of all voltage levels, and to analyze the impact of the main network maintenance on the power supply path of the distribution network and adjust the maintenance time window based on the topology analysis. The spatiotemporal risk assessment unit (500) identifies the power loss range of N-1 faults based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit (100), and predicts the fault probability by combining the equipment health time series and generates a risk degradation scheme.

2. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 1, characterized in that, The spatiotemporal knowledge graph construction unit (100) includes a spatial topology modeling module (110) and a temporal attribute annotation module (120), wherein: The spatial topology modeling module (110) is used to parse the power plants, lines, transformers, circuit breakers, disconnectors and busbars in the power grid CIM model into graph nodes, and construct a spatial topology network with electrical connection relationships as edges; The time attribute annotation module (120) is used to annotate the time dimension attributes of the device nodes and form a device-time association table.

3. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 2, characterized in that, The spatiotemporal knowledge graph construction unit (100) further includes a topology pruning optimization module (130) and a graph database storage module (140), wherein: The topology pruning optimization module (130) generates and executes a topology pruning strategy based on the spatial topology network constructed by the spatial topology modeling module (110). The graph database storage module (140) uses a graph database to store spatiotemporal knowledge graphs.

4. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 3, characterized in that, The spatiotemporal feature extraction unit (200) includes a spatial feature extraction module (210), a temporal feature extraction module (220), and a feature fusion module (230), wherein: The spatial feature extraction module (210) is used to extract the spatial distribution features of equipment from the power grid operation data. It uses a depth-first search algorithm to traverse the spatiotemporal knowledge graph and extract the adjacent relationships of equipment, the power supply range level, and the spatial topology connectivity features. Then, based on the physical location coordinates of the equipment, it calculates the Euclidean distance and spatial clustering coefficient between the equipment to generate a spatial location feature vector. The time feature extraction module (220) is used to extract load time series features and maintenance resource time constraint features from the power grid operation data, train the historical load data using an LSTM neural network to capture the time-dependent relationship and periodic features of the load data; and extract the available time window of the maintenance resources and the resource allocation time constraint to generate a time constraint feature vector; The feature fusion module (230) is used to fuse the spatial feature vector generated by the spatial feature extraction module (210) and the time feature vector generated by the time feature extraction module (220), generate an initial multi-dimensional feature vector by concatenating according to dimensions through the feature concatenation method, and perform normalization processing on the initial multi-dimensional feature vector using the Z-score normalization method to eliminate the dimension difference and generate a final multi-dimensional feature vector.

5. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 4, characterized in that, The intelligent decision-making unit (300) includes a constraint parsing module (310), a genetic algorithm optimization module (320), and a solution evaluation module (330), where: The constraint parsing module (310) is used to parse the spatio-temporal knowledge graph generated by the spatio-temporal knowledge graph construction unit (100) and the multi-dimensional feature vector generated by the spatio-temporal feature extraction unit (200), and extract the constraint conditions required for the maintenance plan scheduling; The genetic algorithm optimization module (320) generates a maintenance plan scheduling solution based on the spatio-temporal constraint conditions extracted by the constraint parsing module (310), and the improved genetic algorithm integrates an elite retention strategy, a dynamic parameter adjustment mechanism, and a double-objective fitness function; The solution evaluation module (330) is used to perform multi-index evaluation on the maintenance plan scheduling solution generated by the genetic algorithm optimization module (320). If the evaluation result does not meet the preset engineering threshold, it triggers the genetic algorithm optimization module (320) to re-iterate and optimize.

6. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 5, characterized in that, The genetic algorithm optimization module (320) generates a maintenance plan scheduling solution through an improved genetic algorithm, including the following steps: S320.

1. Population initialization: Integer encoding is used, and each chromosome contains N gene loci, equal to the total number of devices to be repaired. Each gene locus is g. i The gene value is an integer in the range [0, T] corresponding to the start time of maintenance of the i-th device; M chromosomes of a preset population size are randomly generated to form the initial population; S320.

2. Fitness calculation: For each chromosome, calculate the double-objective fitness function F based on the maintenance time allocation encoded by it; S320.

3. Elite retention strategy: Select the top E individuals with the highest fitness from the current population and directly copy them to the next-generation population to ensure that the optimal solution is not destroyed by genetic operations; S320.

4. Dynamic parameter adjustment: Crossover probability P c And the probability of mutation P m Dynamically adjusts with the number of iterations k: Crossover probability P c : From the initial value P c0 Linearly decrease to the minimum threshold P c1 The formula is: where K is the preset maximum number of iterations; Probability of mutation P m : From the initial value P m0 Linearly increasing to the maximum threshold P m1 The formula is: where K is the preset maximum number of iterations; S320.

5. Genetic operation execution: Selection: A tournament selection method is used, with T randomly selected from the population each time. s For each individual, the one with the highest fitness is selected to enter the mating pool, and the operation is repeated until the size of the mating pool reaches ME; Crossover: Individuals in the mating pool are cross-referenced according to P... c Random pairing, using two-point crossover to generate offspring: Randomly select two crossover points p1, p2 (1 ≤ p1 < p2 ≤ N), and exchange the gene bit segments of the parental chromosomes in the [p1, p2] interval; Variation: Chromosomes in offspring are processed according to P... m Randomly select a gene locus and replace its value with a random integer in the range [0, T]. S320.

6. New-generation population construction: Merge the elite individuals and the mutated offspring individuals to form a new-generation population; S320.

7. Termination condition judgment: If any of the following conditions is met, terminate the algorithm and output the optimal solution: The change in the optimal fitness of the population for consecutive G generations is less than the threshold ε; The number of iterations k reaches the preset maximum number of iterations K.

7. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 6, characterized in that, The main distribution network spatiotemporal coordination unit (400) includes a model stitching module (410), a topology analysis module (420), and a time window adjustment module (430), wherein: The model splicing module (410) is used to splice the main network and distribution network CIM models to generate a fusion topology map of all voltage levels. By parsing the main network and distribution network CIM models, it extracts equipment, connection relationships and electrical parameters, establishes entity mapping and attribute mapping of main network and distribution network equipment to handle the differences in naming rules and topology granularity, and finally merges the main network and distribution network equipment nodes and connection edges to construct a fusion topology map containing electrical attributes and spatial location information. The topology analysis module (420) analyzes the impact of main network maintenance on distribution network power supply paths based on the fused topology diagram. By marking the main network maintenance equipment and updating the topology connection status in the fused topology diagram, it identifies distribution network equipment and power supply paths that are electrically connected to the main network maintenance equipment or affected by changes in power supply paths, and calculates the number of distribution network power supply path interruptions caused by main network maintenance, the total amount of load affected, the number of users, and the load rate of backup paths. The time window adjustment module (430) adjusts the main network maintenance time window according to the topology analysis results of the topology analysis module (420), identifies the load trough period based on the historical load curve of the distribution network, determines the period with sufficient reserve resources in combination with the status of the distribution network equipment, and selects the window that covers both the load trough period and the period with sufficient reserve resources from the candidate maintenance time window of the main network; if there are multiple windows that meet the conditions, the window with the smallest distribution network load impact index is selected as the final maintenance time window.

8. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 7, characterized in that, When identifying the impact of main grid maintenance on the distribution network power supply path, the topology analysis module (420) adopts a cross-voltage level topology tracing method, specifically including: Starting from the electrical connection points of the equipment under maintenance on the main network, the fusion topology is traversed across voltage levels. Prioritize identifying ring network switching paths on the distribution network side, and determine the priority of load transfer based on the real-time status of distribution network sectional switches.

9. The intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics according to claim 8, characterized in that, The spatiotemporal risk assessment unit (500) includes, wherein: The fault range analysis module (510) constructs a whole power grid model containing the electrical connection relationship and spatial distribution of equipment based on the spatiotemporal knowledge graph generated by the spatiotemporal knowledge graph construction unit (100). It identifies the power outage range by simulating a single equipment outage scenario and extracts the number of distribution transformer areas, the number of affected users, the total power outage load and the load importance level within the power outage area. The fault probability prediction module (520) combines the health time series of the evaluated equipment, collects its historical health data, constructs a multi-dimensional health evaluation system, and outputs the fault probability prediction value of the evaluated equipment within a preset time window based on the data time correlation analysis. The risk degradation scheme generation module (530) quantifies equipment risk based on the failure probability and power outage range, through the quantitative correlation between failure probability and the impact of power outage load, prioritizes equipment according to risk value, and generates differentiated risk management strategies.

10. An application method for an intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics, based on the intelligent decision-making system for power equipment maintenance planning based on spatiotemporal characteristics as described in any one of claims 1-9, characterized in that, Includes the following steps: S100. The power grid CIM model is parsed into a graph database structure containing spatial topological relationships and time dimension attributes, and a spatiotemporally related device network model is generated through topology pruning optimization. S200: Extract equipment spatial distribution features, load time series features, and maintenance resource time constraint features from power grid operation data, construct multi-dimensional feature vectors, and implement feature extraction based on graph traversal algorithm and LSTM neural network; S300: Based on the generated spatiotemporal knowledge graph and multidimensional feature vectors, an improved genetic algorithm is used to generate a maintenance schedule arrangement scheme that takes into account both spatial topological constraints and time resource balance. S400, splicing the main grid and distribution network CIM models to generate a fusion topology map of all voltage levels, and based on the topology analysis, analyzing the impact of main grid maintenance on distribution network power supply paths and adjusting maintenance time windows; S500 identifies the power loss range of N-1 faults based on spatiotemporal knowledge graphs, and predicts the failure probability by combining equipment health time series and generates risk degradation schemes.

Images