A power distribution network feasibility study report automatic generation method based on LDA probability model

Abstract

The application provides a power distribution network feasibility report automatic generation method based on an LDA probability model, including: obtaining power grid equipment account information and electrical topology connection relationship, generating an account document containing equipment name, type, position, connection relationship information; taking the obtained account attribute and electrical topology relationship as a benchmark grid graph structure, and marking out standby line and standby equipment data, generating a marking document containing the benchmark grid structure and marking information; training an LDA model, setting the number of topics, performing iterative training, extracting the topic distribution of each document and the vocabulary distribution of each topic; based on the redundancy information, a feasibility report generation model is constructed, the feasibility report generation model is used for analyzing potential topics and modes in the redundancy information, and a power distribution network redundancy feasibility report is generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a power distribution network feasibility study report automatic generation method based on LDA probability model. BACKGROUND

[0002] Evaluating the redundancy of a power distribution network is crucial because it directly affects the reliability and resilience of the power system, enabling rapid restoration of power supply through backup paths in the event of a failure, reducing the impact of power outages, ensuring uninterrupted social and economic activities, and enhancing the overall power grid's ability to withstand unexpected events, ensuring the safety and stability of power supply, and laying a solid foundation for sustainable economic development. However, the challenges of evaluating the redundancy of a power distribution network include significant differences in power distribution networks in different regions, with varying equipment models and parameters, making it difficult to construct a benchmark power grid model. Furthermore, while high redundancy can improve the reliability of a power distribution network, it also means lower resource utilization. Balancing reliability and economy and finding a more optimal level of redundancy is a technical challenge. Secondly, generating a power distribution network redundancy feasibility report requires advanced algorithm models and data mining techniques, such as the LDA probability model. How to intelligently generate reports is also a technical difficulty. Therefore, how to systematically solve these problems, i.e., while considering evaluation accuracy, efficiency, and practicality, effectively addressing the diversity of model construction, the depth of data analysis, the integrated innovation of algorithm application, and the intelligent demand of report generation, is a technical problem that needs to be solved. SUMMARY

[0003] The present application provides a power distribution network feasibility study report automatic generation method based on LDA probability model, mainly including:

[0004] Obtain the grid equipment account information and its electrical topology connection relationship, generate an account document containing device name, type, location, and connection relationship information;

[0005] Construct a graph structure for the benchmark power grid based on the obtained account attributes and electrical topology relationship, and label the backup line and backup device data to generate a labeled document containing the benchmark power grid structure and labeling information;

[0006] According to the backup line and backup device data in the benchmark power grid structure, combined with the number of alternative paths between different devices and lines, calculate the redundancy of each node in the power distribution network, generate an electrical topology graph containing redundancy information and a redundancy information document containing redundancy calculation results;

[0007] Based on the calculated redundancy index of the nodes, analyze the fault recovery capability of the power distribution network under different redundancy levels, evaluate the switching and recovery speed when a fault occurs, and generate a fault analysis document containing the fault recovery capability analysis results and the evaluation results of the switching and recovery speed when a fault occurs.

[0008] According to the historical fault data of the power distribution network, the resource utilization of the power distribution network is evaluated in combination with the redundancy index, the idle condition of the standby equipment and lines in the non-fault state is judged, and an evaluation document containing the historical fault data and the evaluation result of the resource utilization is generated;

[0009] The account document, the labeling document, the redundancy information document, the fault analysis document and the evaluation document of the power grid equipment are summarized, preprocessed, a vocabulary table is constructed, each document is represented as a word frequency vector of the vocabulary in the vocabulary table, a document-word frequency matrix is generated, the row of the matrix represents the document, the column represents the vocabulary, and the value in the matrix represents the frequency of the vocabulary appearing in the document;

[0010] The LDA model is trained, the number of topics is set, iterative training is performed, the topic distribution of each document and the vocabulary distribution of each topic are extracted;

[0011] Based on the redundancy information, a feasibility report generation model is constructed, the feasibility report generation model is used for analyzing the potential topics and patterns in the redundancy information, and a redundancy feasibility report of the power distribution network is generated.

[0012] The technical scheme provided by the embodiments of the present application can include the following beneficial effects:

[0013] The application discloses a power distribution network feasibility study report automatic generation method based on an LDA probability model, records the ledger information of power grid equipment and the electrical topology connection relationship in detail, and generates a ledger document containing the equipment name, type, position and connection relationship, so that the problem of scattered equipment information and difficult management in the traditional method is solved; a marking document containing the benchmark grid structure and marking information is constructed, and then the redundancy of each node is calculated based on the standby line and equipment data in the marking document, the redundancy information is integrated into the electrical topology graph, so that a detailed redundancy information document is formed, and the redundancy of each node in the power distribution network can be calculated and the fault recovery capability of the power distribution network can be evaluated accordingly; further, the application uses historical fault data and the redundancy index to jointly evaluate the resource utilization rate of the power distribution network, generates an evaluation document, the document not only records the historical fault data, but also analyzes the idle condition of standby equipment and lines in the non-fault state, and provides a scientific basis for power grid management; in combination with the above documents, the application generates a document-term frequency matrix by constructing a vocabulary and representing the documents as term frequency vectors, and then trains an LDA model to extract the topic distribution and vocabulary distribution of the documents; the application generates a power distribution network redundancy feasibility report including the calculation result of the redundancy, resource utilization rate analysis and feasibility evaluation, and performs rule checking, and the report helps the power grid managers to adjust and maintain the basic data and graph model data, so that the operation efficiency and stability of the power grid are improved; in a word, the application integrates and analyzes the power grid data, not only improves the management efficiency and transparency of the power grid, but also enhances the fault response capability and resource allocation efficiency of the power grid, and realizes the optimization and upgrading of the power distribution network management. BRIEF DESCRIPTION OF DRAWINGS

[0014] Fig. 1 A flowchart of a power distribution network feasibility study report automatic generation method based on an LDA probability model.

[0015] Fig. 2 A schematic diagram of a power distribution network feasibility study report automatic generation method based on an LDA probability model. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be described clearly and in detail below with reference to the drawings in the embodiments of the application. The described embodiments are only some of the embodiments of the application.

[0017] As Figs. 1-2 , the power distribution network feasibility study report automatic generation method based on the LDA probability model specifically can include:

[0018] S101, obtain the ledger information of power grid equipment and the electrical topology connection relationship, and generate a ledger document containing the equipment name, type, position and connection relationship information.

[0019] The process involves acquiring and analyzing diagram files, extracting equipment symbols and text annotations to obtain the names, types, and locations of power grid equipment, and constructing an equipment attribute table. It also involves analyzing the graphic elements of connecting lines and arrows in the diagram files, combining this with the equipment location information in the equipment attribute table to determine the electrical connections between devices and constructing an equipment topology connection table. The equipment attribute table and the equipment topology connection table are then integrated to form a ledger document. Attribute analysis is performed on the generated ledger document. Based on the attribute analysis results, implicit relationships and frequent patterns between attributes are identified, and relevant attribute combinations that can form contextual logical connections and thought chains are found. Using the identified relevant attribute combinations, combined with the business characteristics and professional knowledge of power grid equipment, the ledger document is expanded and / or refined. Finally, the expanded and / or refined ledger document undergoes logical checks and consistency verification.

[0020] Specifically, according to the graph model file derived from the South Grid Smart Panoramic System, the graph model file is analyzed, and the name, type, location and other attribute information of the power grid equipment are obtained by extracting the device symbols, text annotations and other information in the graph model, and the device attribute table is constructed. Using the location information in the device attribute table, the electrical connection relationship between devices is judged by analyzing the connection lines, arrows and other graphical elements in the graph model file, and the device topology connection relationship table is constructed to determine the upstream and downstream relationship and electrical connectivity between devices. According to the device attribute table and the topology connection relationship table, the attribute information and connection relationship are integrated by using data fusion techniques such as data mapping and correlation analysis to generate complete power grid equipment account information and structured data sets of electrical topology relationship, forming an account document. The attribute information such as device name, type, location and connection relationship is directly extracted from the structured account document without using natural language processing technology. Using the extracted attribute information, the association rule mining algorithm such as Apriori algorithm or FP-growth algorithm is used to find the implicit relationship and frequent pattern between attributes. By analyzing the association rules of device type, location and connection relationship, the relevant attribute combinations that can form context logical connection and thinking chain are found. According to the discovered relevant attribute combinations, combined with the business characteristics and professional knowledge of power grid equipment, the account document is expanded and refined. According to the association of device type and location, the running state information of the device is supplemented; according to the association of device connection relationship and name, the business attribute information such as maintenance record and fault history of the device is supplemented, enriching the content of the account document. The expanded account document is used to construct a knowledge graph, define concepts and entities such as devices, attributes and relationships, and establish semantic links between them. Reasoning rules such as device type and running state constraint rules, device connection relationship and fault propagation reasoning rules are developed. Using the knowledge graph and reasoning rules, the expanded account document is logically checked and consistency verified to verify whether the logical relationship between attributes is reasonable, ensuring the integrity and accuracy of the account document, forming a high-quality power grid equipment account information and electrical topology relationship account document. First, the graph model file derived from the South Grid Smart Panoramic System is preprocessed using the OpenCV image processing library, such as grayscale and binarization, to improve image quality. Then, the region-based convolutional neural network (R-CNN) is used to detect and recognize the device symbols in the graph model file. Through the trained device symbol model, the transformer, circuit breaker, bus and other devices are recognized with an accuracy of more than 95%. Next, the optical character recognition (OCR) technology such as Tesseract OCR engine is used to extract the text annotation information in the graph model file, and the device name, type and other attributes are associated with the corresponding device symbols to construct the device attribute table. For the recognized connection lines and arrows, the Hough transform algorithm is used for straight line detection, and the upstream and downstream topology relationship of the devices is judged according to the start and end position and direction of the connection lines between the devices to generate the device topology connection matrix.The device attribute table and topology connection matrix are associated, and a structured dataset containing detailed device attributes and connection relationships is generated through primary key matching and foreign key association. Association rule mining is performed on the dataset using the Apriori algorithm, with a minimum support of 10% and a minimum confidence of 80%. Association rules between device type and voltage level are discovered, such as "transformer => 220kV" and "circuit breaker => 110kV," revealing the inherent relationships between device attributes. Based on these association rules, the ledger documents are expanded, such as by adding parameters like transformer capacity, turns ratio, and wiring group, and circuit breaker rated current and rated breaking current. A device knowledge graph is constructed using the Neo4j graph database, storing device, attribute, and connection relationship information as nodes and edges. The Cypher query language is used to query, infer, and verify device attributes and topology relationships, such as "MATCH(d:Device)--".

[0021] The query `(t:Transformer)WHEREd.type = 'Circuit Breaker'RETURNd,t` can retrieve all circuit breakers directly connected to the transformer. The TransE algorithm vectorizes entities and relationships in the knowledge graph, and a representation learning model predicts and completes equipment attributes and connection relationships, further improving the completeness and accuracy of the ledger documents.

[0022] S102. Construct a graph structure of the baseline power grid based on the obtained ledger attributes and electrical topology relationships, and mark the backup lines and backup equipment data to generate a labeling document containing the baseline power grid structure and labeling information.

[0023] Based on the extracted ledger attributes and electrical topology relationships, nodes and relationships are created, with devices as nodes and topology connections as edges, to construct a graph structure for the benchmark power grid. This includes labeling standby lines and standby equipment to distinguish between primary and standby lines and equipment, and labeling standby lines with their corresponding primary line information to form a standby relationship mapping. Key nodes and critical paths in the power grid topology are analyzed to assess their importance and impact range. Based on the topology and device attributes, network characteristic parameters for power supply paths and load distribution are generated. For device nodes in the graph structure, ledger attribute information is associated, and for standby equipment, its commissioning conditions and switching methods are associated to form a standby equipment attribute profile. The graph structure is integrated with the geographical information of the devices, and nodes are mapped onto a map based on the geographical location attributes of the devices to form a power grid geographic topology map to display the spatial distribution and geographical wiring relationships of the power grid. Based on the constructed graph structure of the benchmark power grid, a labeling document containing device nodes, topology connections, and standby labeling information is generated. The labeling document includes visual attributes such as node color and shape to distinguish different types of devices and connections.

[0024] Specifically, according to the extracted account attributes and electrical topology relations, a Neo4j graph database management tool is used to create nodes and relations, with devices as nodes and topology connection relations as edges, to build the graph structure of the benchmark power grid. Through the Cypher query language, flexible queries and analysis of the power grid structure are realized. In the process of building the graph structure, special label attributes such as "backup" and "redundant" are set for backup lines and backup devices. Through attribute labeling, main and backup devices are distinguished, facilitating subsequent identification and management. For backup lines with corresponding main lines, the corresponding main line IDs are labeled; for independent backup lines, backup types, commissioning conditions, and other attributes are labeled. Using the path query function of the graph database, appropriate algorithms are selected according to specific business needs, such as Dijkstra algorithm for solving the shortest path problem, A* algorithm for solving the weighted shortest path problem, and depth-first search (DFS) algorithm for solving the full path problem. When applying the algorithm, appropriate weights, constraint conditions, and other parameters are set to analyze the key nodes and key paths in the power grid topology structure, evaluate the importance and influence range of the nodes, and generate network characteristic parameters such as power supply paths and load distribution based on the topology structure and device attributes. For device nodes in the graph structure, attributes such as device type, rated parameters, and manufacturer are associated to enrich the attribute dimensions of the nodes. For backup devices, their commissioning conditions and switching modes are also associated to form a complete backup device attribute profile. Commissioning conditions can include fault types such as busbar fault and line fault, fault locations such as XX substation and XX line, and load levels such as XX% of rated load; switching modes can include automatic switching such as through relay protection action and manual switching such as through dispatcher remote operation. These attributes can be designed as enumeration types, numerical types, and string types according to actual needs. After completing the graph structure construction and attribute association, the graph structure is integrated with the geographic information system (GIS). First, unique geographic coordinates and spatial geometric objects are assigned to each device in the GIS system, and then these geographic attributes are associated with the device nodes in the graph structure to realize the fusion of spatial data and topology data. Based on the integrated graph structure, graph visualization technology is used to design appropriate map symbols and annotation styles to generate annotated documents containing device nodes, topology connections, and backup annotations. Through the visual attributes of nodes such as color and shape, different types of devices and connections are distinguished, and a legend is provided to facilitate users to intuitively identify and locate devices. The generated benchmark power grid graph data model is continuously maintained and updated according to the actual operation of the power grid, with device status and topology structure dynamically adjusted to ensure that the graph database is consistent with the field at all times. In the Neo4j graph database, CREATE statements are used to create device nodes and topology relation edges, with device node attributes including device ID, device type, and rated voltage, and topology relation edge attributes including conductor material, conductor length, and resistance.For backup equipment, the additional type attribute is marked as "backup", and for backup lines, the backup line id attribute marks the main line id. The shortest power supply path is analyzed using the Dijkstra shortest path algorithm, the substation node is specified as the starting point, the load node is specified as the ending point, and the line impedance is specified as the weight. The shortest power supply distance is calculated. The DFS algorithm is used to traverse the electrical island in the power flow calculation, the traversal depth is defined as 3, and all nodes and edges in the electrical island are marked. The device node is associated with the geographic coordinates using the NEO4J-SPATIAL plug-in, and the SVG graph element is used for the substation. <circle>Labels, using broken line segments for transmission lines, are defined.<polyl ine> Tags. In a GIS system using the WGS84 coordinate system, other devices within a 500-meter radius of the current device can be queried via spatial indexing. The Neo4j version management plugin neo4j-vers ioner-core can also be used to record the change history of the map structure, defining the attribute names, old values, and new values ​​of the modified operations. Each version records a change timestamp and supports version rollback.

[0025] S103. Based on the data of backup lines and backup equipment in the baseline power grid structure, and combined with the number of alternative paths between different equipment and lines, calculate the redundancy of each node in the distribution network, and generate an electrical topology diagram containing redundancy information and a redundancy information document containing the redundancy calculation results.

[0026] Analyze the baseline power grid structure, extract the connection relationships between nodes in the distribution network, and construct an adjacency matrix representing the network topology. Traverse all reachable nodes in the adjacency matrix and record the path from the starting node to each reachable node to obtain a set of alternative paths between all nodes in the distribution network. Each path in the alternative path set represents a potential power supply scheme. Based on the attribute information of backup lines and backup equipment, the alternative paths are filtered and sorted, prioritizing paths containing backup lines and backup equipment while excluding paths corresponding to faulty equipment and faulty lines, resulting in a set of effective alternative paths for each node. Combining the effective alternative path set and the baseline power grid structure, calculate the redundancy index for each node. The redundancy index represents... The system assesses the power supply reliability of nodes; based on the calculation results of redundancy indices, it classifies and sorts distribution network nodes, generating a distribution network redundancy level classification table; it correlates the calculation results of node redundancy indices with the geographical topology information of the distribution network, marking nodes with different redundancy levels with different colors or patterns, generating an electrical topology map that integrates redundancy information to display the power supply reliability distribution of the distribution network; it summarizes various data and results from the node redundancy analysis process, including the node adjacency matrix, the set of alternative paths, the set of effective alternative paths, the node redundancy indices, the redundancy classification levels, and the distribution network redundancy level classification table, generating an electrical topology map containing redundancy information and a redundancy information document containing the redundancy calculation results.

[0027] Specifically, by analyzing the benchmark power grid structure in the graph database, the connection relationship between each node in the distribution network is extracted, and an adjacency matrix is constructed to represent the network topology. The elements in the adjacency matrix represent the connectivity between nodes. If there is a line connection between two nodes, the corresponding matrix element value is 1, otherwise it is 0. Depth-first search (DFS) algorithm is used to traverse the adjacency matrix. From each node, all reachable nodes are explored, and the path from the starting node to each reachable node is recorded. DFS algorithm uses stack structure to realize the traversal of the graph, the specific steps include: (1) select the starting node, mark it as visited, and push it into the stack; (2) when the stack is not empty, pop the top node from the stack, and check all its unvisited adjacent nodes; (3) for each unvisited adjacent node, mark it as visited and push it into the stack, and record its connection with the current node; (4) repeat steps (2) and (3) until the stack is empty. Through the DFS algorithm, a set of alternative paths between all nodes in the distribution network is obtained, each path represents a potential power supply scheme. The alternative path set is processed, and according to the attribute information of the backup line and the backup device, the alternative paths are filtered and sorted. The paths containing backup lines and backup devices are preferentially selected, and the paths corresponding to faulty devices and faulty lines are excluded, and finally the effective alternative path set of each node is obtained. In order to consider the power supply capacity and reliability of the alternative path comprehensively, the concept of weighted graph is introduced, and the lines and devices on the alternative path are weighted. The weight can reflect the power supply capacity and reliability, for example, the weight of the line can be set as its current capacity, and the weight of the device can be set as its reliability index. When calculating the redundancy index, not only the number of effective alternative paths is considered, but also the weighted sum of alternative paths. The calculation formula of the redundancy index is: node redundancy=(effective alternative path weighted sum) / (total connection line weighted sum). The higher the redundancy index, the better the power supply reliability of the node. According to the calculation result of the redundancy index, referring to the industry standard and expert experience, a scientific redundancy index threshold is set to classify and sort the distribution network nodes. According to the order from high to low of the redundancy index, it is divided into high redundancy (redundancy index >=0.8), medium redundancy (0.5<=redundancy index

[0028] <0.8), low redundancy (redundancy index <0.5) three levels, get the distribution table of power distribution network redundancy level, and mark and statistics of each level node, generate redundancy distribution chart or redundancy statistics report. The node redundancy information and the geographical topology information of power distribution network are associated, and the nodes of different redundancy levels are marked with different colors or patterns in the electrical topology map, and the electrical topology map with integrated redundancy information is generated, which can intuitively show the distribution of power supply reliability of power distribution network. Using Python Pandas and Matplotl ib libraries, the electrical topology map and other analysis results are sorted, counted and visualized, and a reasonable document structure and layout are designed to generate the redundancy information document containing the electrical topology map and redundancy calculation results. The redundancy information document includes power distribution network basic information, backup line and backup equipment statistics, node redundancy analysis, electrical topology map, conclusion and suggestion, etc., which summarizes the data and results in the process of node redundancy analysis. The adjacency matrix of the power distribution network is extracted from the graph database, with a dimension of 1000*1000, representing 1000 nodes in the power distribution network. The NetworkX library is used to build a graph object G, and the adjacency matrix is converted to the edge set of the graph by the G.add_edges_from() method. The nx.dfs_tree(G, source=node) function is used for depth-first traversal of the graph, where node represents the starting node, and the traversal result is returned in the form of a tree. The tree is converted to a graph by the nx.from_ nx_ tree() function, and the redundancy index of each node is calculated by the nx.degree_ centrality() function. The redundancy index of each node is marked on the electrical topology map, and the redundancy information document is generated.

[0029] nx. all_s imple_paths(G,source=node,target=target) function obtains all simple paths from the starting node to the target node, forming a set of alternative paths. Assuming the number of alternative paths is 500, the Pandas DataFrame object is used to store the alternative paths, with column names "path", "weight", and "reliability" representing the path sequence, power supply capacity, and reliability, respectively. The power supply capacity is calculated by weighting the current capacity of the lines on the path, with a value range of [0, 1]; the reliability is calculated by weighting the reliability indicators of the devices on the path, with a value range of [0, 1]. The product of the path weight and the reliability is used as the basis for sorting, and the top 100 are selected as high-quality alternative paths. The redundancy index is calculated for each node, with a mean value of 0.75 and a standard deviation of 0.2. The Matplotlib library is used to draw a histogram of the redundancy index distribution, with the horizontal axis representing the redundancy index and the vertical axis representing the number of nodes. The redundancy index threshold is set to 0.8 and 0.5, resulting in 200 high-redundancy nodes, 500 medium-redundancy nodes, and 300 low-redundancy nodes. The Geopandas library is used to combine the geographic coordinate information of the distribution network nodes with the redundancy index to generate a geographic topology map of the distribution network redundancy. The Pandas DataFrame object and Matplotlib library are used to generate a redundancy analysis report and export it as an HTML format. The redundancy information document contains four chapters: distribution network basic information, alternative path analysis, redundancy index distribution, and redundancy level classification, each containing 1-2 charts and corresponding analysis text.

[0030] S104, based on the redundancy index calculated by the node, analyze the fault recovery capability of the distribution network under different redundancy levels, evaluate the switching and recovery speed when a fault occurs, and generate a fault analysis document containing the analysis results of the fault recovery capability and the evaluation results of the switching and recovery speed when a fault occurs.

[0031] Based on the power distribution network redundancy level division table, fault simulation analysis is performed for each redundancy level, including calculating the power supply recovery time of the node under the fault condition according to the number of alternative paths and the power supply capacity of the node, obtaining the fault recovery time data under different redundancy levels; analyzing the influence of automatic switching mode and manual switching mode on the fault switching time, obtaining the fault switching time data; using the fault recovery time data and the fault switching time data, evaluating the fault recovery capability of the power distribution network under different redundancy levels, including calculating the fault recovery capability index; analyzing the correlation between the fault recovery capability index and the redundancy index, fitting the functional relationship between the redundancy index and the fault recovery capability index, obtaining the influence law of the redundancy index on the fault recovery capability, forming the quantitative relationship model of the redundancy index and the fault recovery capability; summarizing the evaluation results of the fault recovery capability of the power distribution network under different redundancy levels and the quantitative relationship model of the redundancy index and the fault recovery capability, generating the fault analysis document containing the fault recovery capability analysis results and the evaluation results of the switching and recovery speed when the fault occurs.

[0032] Specifically, based on the distribution network redundancy level classification table, nodes with different redundancy levels have different recovery capabilities and switching speeds when faults occur. For each redundancy level, PSASP power system simulation software is used for fault simulation analysis. According to the distribution network topology and equipment parameters, a simulation model is constructed, fault scenarios and simulation parameters are set, and the situation of node failure is simulated. According to the number of alternative paths and power supply capacity of the node, the power supply recovery time of the node under fault conditions is calculated, the average fault recovery time data under different redundancy levels is obtained, and the corresponding relationship table of distribution network redundancy level and fault recovery time is formed. On the basis of fault simulation analysis, further consideration is given to the fault switching mode of the node, and the automatic switching and manual switching are distinguished. Automatic switching is realized through intelligent switch and fault positioning technology, and manual switching needs manual operation. According to the difference of switching mode, the fault switching time of the node is calculated, and the average fault switching time data under different redundancy levels is obtained. Using fault recovery time and fault switching time data, the fault recovery capability of distribution network under different redundancy levels is comprehensively evaluated. The standard recovery time is defined as a reference recovery time determined according to industry standards or experience value, which represents that the power supply is considered acceptable within this time. The fault recovery capability index is calculated, and the higher the index, the stronger the fault recovery capability of the distribution network under this redundancy level. The calculation formula of fault recovery capability index is: fault recovery capability index = (standard recovery time-average fault recovery time) / standard recovery time × 100%. On the basis of calculating the fault recovery capability index, the correlation between the fault recovery capability index and the redundancy index is further analyzed. Scatter plots of redundancy index and fault recovery capability index are drawn to observe the correlation between them. According to the shape of the scatter plot, a suitable regression model is selected, such as linear regression, polynomial regression, exponential regression, etc. The least squares method is used for model parameter estimation to obtain the coefficients and intercepts of the regression model. The determination coefficient (R 2 ), root mean square error (RMSE), etc. to evaluate the goodness of fit and prediction accuracy of the regression model, and obtain a quantitative relationship model between the redundancy index and the fault recovery capability. By summarizing the evaluation results of the fault recovery capability of the distribution network under different redundancy levels and the quantitative relationship model between the redundancy index and the fault recovery capability, a fault analysis document containing the analysis results of the fault recovery capability and the evaluation results of the switching and recovery speed when a fault occurs is generated. In order to realize real-time monitoring of the redundancy of the distribution network, intelligent meters, miniature circuit breakers, and other monitoring devices are deployed in the distribution network to collect real-time data such as current, voltage, and power of each node through wireless communication or power line communication technology. The collected data is cleaned, stored, and analyzed to calculate the real-time redundancy index of the nodes and form a redundancy monitoring report and early warning information. According to the results of the fault recovery capability analysis and the real-time monitoring of the redundancy, the distribution network is optimized. For key nodes and important loads in the distribution network, targeted optimization strategies are developed based on their redundancy index and fault recovery capability index, such as increasing backup lines and improving device reliability. By optimizing the topology structure and device configuration of the distribution network, the redundancy level of key nodes and important loads is improved, thereby improving their fault recovery capability. The results of the fault recovery capability analysis can also be applied to the planning and operation management of the distribution network. In the planning stage of the distribution network, the redundancy level of the distribution network is set reasonably based on load forecasting and reliability requirements. In the operation stage of the distribution network, the operation mode of the distribution network is dynamically adjusted based on real-time redundancy monitoring data to ensure that the distribution network maintains good fault recovery capability under different redundancy levels. Redundancy data of 500 nodes in the distribution automation system are obtained, including node ID, redundancy index, power supply capacity, and other attributes. The K-means clustering algorithm is used to cluster the nodes with a cluster number k = 3, resulting in three redundancy levels: high, medium, and low, with cluster centers of 0.85, 0.60, and 0.35, respectively. A distribution network simulation model is built on the PSASP simulation platform, including 500 nodes, 750 lines, and 100 switches. Three fault scenarios are set: single-point grounding, two-point grounding, and three-phase short circuit, with 100 random fault points in each scenario. The average fault recovery time of each node is calculated by running the simulation program, with 20 minutes for high redundancy level, 40 minutes for medium redundancy level, and 60 minutes for low redundancy level. For automatically switched nodes, the switching time is calculated as 2 minutes, and for manually switched nodes, the switching time is calculated as 10 minutes. Assuming that the standard recovery time is 30 minutes, the fault recovery capability index of each redundancy level is calculated as 93.33%, 83.33%, and 66.67%, respectively. A scatter plot of the redundancy index and the fault recovery capability index is drawn, showing a significant positive correlation between the two.Linear regression, logarithmic regression, exponential regression and other models were tried, and finally the logarithmic regression model was selected, the regression equation was y = 0.2030ln(x) + 0.8616, where x was the redundancy index, y was the fault recovery capability index. The determination coefficient R of the model. 2 = 0.8762, RMSE = 0.0336, and the goodness of fit was good. In the distribution network, 100 nodes were selected, smart meters and miniature circuit breakers were installed, and data was uploaded once a minute through ZigBee wireless communication. A data platform was built using Hadoop, data cleaning and conversion were performed using Hive, and real-time calculation was performed using SparkStreaming, and the redundancy index was updated every 5 minutes. When the redundancy index was lower than 0.4, the system automatically pushed warning information to prompt the operation and maintenance personnel to handle it in time. For important nodes with a redundancy index lower than 0.2, optimization schemes were developed, such as adding one standby line or replacing one old transformer, and after optimization, the average redundancy index of important nodes increased to 0.56. In the planning of distribution network, according to the load prediction data and reliability requirements, the redundancy target value of each region was set, such as 0.80 for urban areas and 0.60 for rural areas. In the operation of distribution network, the real-time redundancy index was evaluated every 15 minutes, and when the index was lower than the target value, the standby line or standby power supply was automatically switched to improve the power supply reliability.

[0033] S105, according to the historical fault data of the distribution network, combined with the redundancy index, the utilization rate of the distribution network resources was evaluated, the idle condition of the standby equipment and line in the non-fault state was judged, and an evaluation document containing the historical fault data and the evaluation result of the resource utilization rate was generated.

[0034] Obtain power distribution network historical fault records, the fault records including fault occurrence time, fault equipment, fault type, influence range, repair time information; preprocess the power distribution network historical fault records; form a power distribution network fault data set by summarizing the preprocessed power distribution network historical fault records; according to the topology structure and equipment account information of the power distribution network, spatially map the power distribution network fault data set to determine the specific location of each fault event occurrence; associate the specific location with the attribute information of the fault equipment to form the attribute portrait of the fault equipment; analyze the fault characteristics and rules of different types of equipment using the attribute portrait of the fault equipment; based on the redundancy information document, perform correlation analysis on the power distribution network fault data set, calculate the fault rate, fault duration, fault influence range indicators under each redundancy level to evaluate the impact of redundancy level on power distribution network reliability; based on the evaluation results, establish a quantitative relationship between redundancy and reliability; analyze the relationship between faults, types and redundancy to find out the main causes and risk factors of faults; evaluate the utilization efficiency of standby resources, identify the optimal balance point of redundancy and resource utilization rate; construct a fault prediction model for predicting equipment fault probability and type; summarize the analysis results of the fault characteristics and rules of different types of equipment, the quantitative relationship between redundancy and reliability, the main causes and risk factors of faults, the evaluation results of standby resource utilization efficiency, and the fault prediction results, and generate an evaluation report of power distribution network historical fault data analysis and resource utilization rate results.

[0035] Specifically, from the historical database of the distribution automation system, the past year's fault records of the distribution network are extracted, including fault occurrence time, fault equipment, fault type, impact range, repair time, etc., to form a distribution network fault data set. The fault data is cleaned and preprocessed to eliminate incomplete and inconsistent data, and the fault data is standardized to a structured data format. According to the topology structure and equipment account information of the distribution network, the fault data is spatially mapped to determine the specific location of each fault event, and the attribute information of the fault equipment is associated, such as equipment type, rated parameters, manufacturer, etc. Selecting equipment type, fault type, fault frequency, fault duration, etc. Key attributes, attribute values are discretized or normalized to form a standardized attribute feature vector. Then use radar chart, parallel coordinate chart and other visualization methods to intuitively display the feature distribution of the fault equipment in different attributes, and form the attribute portrait of the fault equipment. Combined with the redundancy information of the distribution network, the fault data is analyzed, and the fault rate, fault duration, fault impact range, etc. Indexes are calculated under each redundancy level to evaluate the impact of redundancy level on the reliability of the distribution network. Using statistical methods such as chi-square test, correlation coefficient, etc. Analyze the correlation between fault data and equipment attributes, redundancy and other factors to find out the key factors affecting fault occurrence. The quantitative relationship between redundancy and reliability is obtained. Using association rule mining algorithms such as Apriori algorithm or FP-growth algorithm, the fault data and redundancy information are deeply mined. Set the minimum support to 10%, the minimum confidence to 60%, and mine the frequent item sets and association rules that meet the support and confidence thresholds. Then use lift, chi-square value, etc. Indexes to evaluate the effectiveness and significance of the association rules, and select strong association rules. Evaluate the impact of different factors on fault occurrence, and obtain the main causes and risk factors of fault occurrence. Based on the analysis of fault data, the utilization of standby equipment and standby lines is evaluated. The idle time and idle frequency of standby equipment and standby lines in non-fault state are counted, and the average utilization rate and maximum idle time of standby resources are calculated. Compare the utilization of standby resources under different redundancy levels to find the balance point between redundancy and resource utilization. Select decision tree, random forest, support vector machine, etc. Common classification algorithms to establish a fault prediction model. Input equipment attributes, redundancy, environmental factors, etc. Characteristics, parameter tuning and cross-validation of the model. Use precision, recall, F1 value, ROC curve, etc. Evaluation indexes to evaluate the classification performance of the model. At the same time, use feature importance sorting, feature selection, etc. Methods to optimize the feature space and generalization ability of the model. Summarize the results of distribution network fault data analysis, reliability evaluation, association rule mining, standby resource utilization evaluation, fault prediction, etc. Generate an evaluation document for distribution network historical fault data analysis and resource utilization evaluation.The document content includes fault data statistics, redundancy impact analysis, association rule mining results, backup resource utilization evaluation and fault prediction model evaluation parts; the fault data statistics part includes the time distribution, regional distribution and equipment type distribution of fault occurrence; the redundancy impact analysis part includes the fault rate under different redundancy levels, the comparison of fault duration, the correlation analysis of redundancy and reliability indicators; the association rule mining results part includes the frequent item set and association rule list, the association network graph of key influencing factors; the backup resource utilization evaluation part includes the average utilization rate of backup equipment and backup line, the maximum idle time statistics, the balance analysis of redundancy and resource utilization rate; the fault prediction model evaluation part includes the classification performance indicators of the model, the feature importance ranking and the prediction results of typical fault cases. From the Oracle database of the distribution automation system, 2000 fault records from January 1, 2022 to December 31, 2022 are extracted through SQL query statements. The Pandas library of Python is used to clean the fault data, remove fields with more than 20% missing values, and fill the missing values of fields with less than 20% with mean or mode. The regular expression is used to standardize the fault equipment name, and different models of circuit breakers, transformers and other equipment are normalized to unified equipment types. The fault equipment longitude and latitude coordinates are converted to specific geographic location information using Baidu Map API. The fault frequency and fault duration are normalized to the [0, 1] interval. Scatter plots of fault equipment in fault frequency and fault duration are drawn, and different colors are used to distinguish equipment types. The Apriori algorithm is used to mine association rules from fault data and redundancy information, with a minimum support of 0.1 and a minimum confidence of 0.6. A total of 50 association rules are mined. The lift and chi-square value of each rule are calculated, and the rules with lift greater than 2 and chi-square value greater than 3.841, with degree of freedom 1 and significance level 0.05, are selected as strong association rules, resulting in 10 strong association rules. The idle time of backup equipment and backup line accounts for the proportion of total time, the average idle rate of backup transformer is 35%, and the average idle rate of backup line is 40%. The random forest algorithm is used to establish a fault prediction model, with fault frequency, fault duration, equipment type and redundancy as input features, and fault type as output. 70% of the data is used as the training set and 30% of the data is used as the test set. Through grid search and 5-fold cross-validation, the optimal parameters of random forest are determined as the number of trees 100, the maximum tree depth 10 and the minimum number of samples in the node 5. The model performance is evaluated on the test set, with accuracy 85%, precision 82%, recall 87%, F1 value 0.84 and AUC value 0.91.The evaluation document of the fault data analysis is generated, and the document content includes: the fault data profile of distribution network A, the fault frequency of equipment type A is the highest, accounting for 35% of the total number of faults; the average fault recovery time of distribution network B is the longest, which is 55 minutes. With the redundancy increasing from 0.2 to 0.8, the fault rate decreases from 15% to 3%, and the average fault recovery time shortens from 60 minutes to 20 minutes. The fault type prediction accuracy of the random forest model on the test set is 85%, and the prediction accuracy of transformer fault is the highest, reaching 92%. Suggestions can also be included, such as focusing on monitoring and maintaining transformers, reasonably configuring the number of standby transformers, regularly inspecting and maintaining high-risk lines, and ensuring the safe and stable operation of the distribution network.

[0036] S106, the account documents, the labeling documents, the redundancy information documents, the fault analysis documents and the evaluation documents of the power grid equipment are summarized, preprocessed, and a vocabulary table is constructed; each document is represented as a word frequency vector of the vocabulary in the vocabulary table, a document-word frequency matrix is generated, the rows of the matrix represent the documents, the columns represent the vocabulary, and the values in the matrix represent the frequency of the vocabulary appearing in the document.

[0037] The account documents, the labeling documents, the redundancy information documents, the fault analysis documents and the evaluation documents of the power grid equipment are summarized, and the contents of the summarized documents are converted into a pure text format, while non-text elements including pictures and tables are removed, to obtain a pure text data set; the pure text data set is preprocessed, including word segmentation, stop word removal, and part-of-speech tagging operations; the preprocessed pure text data set is part-of-speech tagged to form a vocabulary table; the word frequency is counted based on the vocabulary table, the words with a frequency exceeding a threshold value are selected as keywords to form a keyword list; by traversing each document, the frequency of each word in the keyword list appearing in the document is counted to form a word frequency vector of the document, the dimension of the word frequency vector is equal to the length of the keyword list, and each element of the word frequency vector represents the frequency of the corresponding keyword appearing in the document; a document-word frequency matrix is constructed based on the word frequency vectors of the documents, the rows of the matrix represent the documents, the columns represent the vocabulary, and the values in the matrix represent the frequency of the vocabulary appearing in the document.

[0038] Specifically, the account documents, labeling documents, redundancy information documents, fault analysis documents and evaluation documents of power grid equipment from different sources are summarized, including account attributes and electrical topology connection relationship, benchmark power grid structure and standby line, standby equipment data, redundancy calculation results, fault recovery capability evaluation, historical fault data and resource utilization evaluation results, etc. The document content is converted to plain text format. Using regular expression matching and document parsing tools, remove non-text elements such as pictures and tables in text data, and extract pure text information. The extracted text data is cleaned and filtered, and Simhash algorithm is used to identify duplicate documents, and regular expression matching is used to identify blank documents and format error documents, which are removed or corrected to obtain the preprocessed text dataset. The preprocessed text dataset is segmented, stop words are removed, and part-of-speech tagging is performed. Use jieba segmentation tool, use forward maximum matching algorithm based on dictionary for segmentation, remove common words in stop word table, and build stop word table based on Chinese stop word corpus. Use NLTK library to perform part-of-speech tagging on the segmentation results, and tag the part-of-speech of each word. The annotation set can use the PKU annotation set of the Institute of Computational Linguistics of Peking University. Based on the preprocessed text data, a vocabulary table is constructed, all non-repeating words in the text data are counted, and a word list is generated. Calculate the frequency of each word in the document set, divide by the total number of documents to get the document frequency (DF) of the word. Calculate the frequency of each word in each document, divide by the document length to get the term frequency (TF) of the word. Calculate the TF-IDF value of the word, sort by TF-IDF value from high to low, select the top-80 words as keywords to form a keyword list. Represent each document as a frequency vector of the word table, and the dimension of the vector is equal to the size of the word table. Traverse each document, count the frequency of each word in the word table in the document, and generate the word frequency vector of the document. Each element of the vector represents the frequency of the corresponding word in the document, and the frequency of the word not appearing is 0. Finally, a document-term frequency matrix is formed, the row represents the document, the list represents the word in the word table, and the element in the matrix represents the frequency of the word in the document. Considering the high-dimensional sparsity of the document-term frequency matrix, the CSR (Compressed Sparse Row) format is used to store the matrix, only the non-zero elements and their position information are stored, which greatly reduces the storage space of the matrix and improves the matrix operation efficiency. The document data from 10 business systems such as equipment account system, operation and maintenance management system, fault diagnosis system, etc. is summarized, a total of 1000 documents, 1GB data. Use regular expression "<img.?>|<table.?>|<chart.*?>" to match and remove non-text elements such as pictures, tables and charts in the text, the removal rate is 95%. Use Simhash algorithm to detect document duplication, with a Hamming distance threshold of 5, remove duplicate documents, removal rate is 8%.The text is segmented using the j ieba segmentation tool, and the Chinese segmentation based on HMM is performed using the Vi terbi algorithm, with an average segmentation accuracy of 96%. The part-of-speech tagging uses the PennTreebank tagging set of the NLTK library, with a tagging accuracy of 94%. A vocabulary table is constructed, containing 20,000 non-redundant words. The TF-IDF value of each word is calculated, and the top-100 words are taken as keywords. A 10000x20000 document-term frequency matrix is generated, stored in CSR format, and the compression rate is.

[0039] S107, training the LDA model, setting the number of topics, iterative training, extracting the topic distribution of each document and the word distribution of each topic.

[0040] According to the document-term frequency matrix, the input of the LDA topic model is constructed, each document is represented as a frequency vector or a weighted vector of words and is normalized to ensure that the sum of the vector elements is 1; the number of topics of the LDA model is set in combination with the document characteristics and domain knowledge; the parameters of the LDA model are initialized, including the topic-word distribution parameter and the document-topic distribution parameter; the LDA model is trained to gradually converge to the optimal topic-word distribution and document-topic distribution; the posterior probability of each topic is calculated, and the topic-word distribution and the document-topic distribution are updated; based on the updated topic-word distribution and the document-topic distribution, the word distribution of each topic and the topic distribution of each document are obtained.

[0041] Specifically, according to the document-term frequency matrix, the input format of the LDA topic model is constructed, each document is represented as a frequency vector or a weight vector of words, and normalization processing is performed to ensure that the sum of vector elements is 1, satisfying the probability distribution constraint condition of the LDA model. The specific format of the document-term matrix can be a TF matrix, a TF-IDF matrix, or a word frequency matrix, which generates a sparse matrix representation by counting the frequency or weight of each word in each document. Set the number of topics parameter of the LDA model, select the appropriate number of topics according to the size of the document set and domain knowledge, usually the value range is 10ˉ100. The optimal number of topics can be determined by cross-validation, perplexity evaluation, topic consistency evaluation, etc. to balance the complexity and interpretability of the model. Cross-validation selects the optimal number of topics by calculating the prediction error of the LDA model on the test set; perplexity evaluation selects the minimum perplexity number of topics by calculating the probability prediction ability of the LDA model on the test set; topic consistency evaluation selects the highest topic consistency number by calculating the semantic similarity between topics. Initialize the parameters of the LDA model, including the topic-word distribution parameter β and the document-topic distribution parameter θ, which can be randomly initialized using the symmetric Dirichlet prior distribution. Set the prior parameters α and η to control the shape and scale of the prior distribution, commonly used values are α = 50 / K, η = 0.1 or 0.01, where K is the number of topics. You can also draw a perplexity curve, that is, take the number of topics K as the horizontal axis and the corresponding perplexity value as the vertical axis, and determine a suitable K value by observing the trend of the curve. The ideal K value is usually located near the inflection point of the perplexity curve, that is, the position where the perplexity decreases slowly. The hyperparameters α and β have an important influence on model training, α controls the prior of document-topic distribution, and β controls the prior of topic-word distribution. A larger α will make each document tend to contain more topics, and a larger β will make each topic tend to contain more words. Hyperparameter optimization can be performed through methods such as grid search. Use Gibbs sampling algorithm to train and infer the LDA model, iteratively update the topic assignment of each word in the document, and gradually converge to the optimal topic-word distribution and document-topic distribution. Gibbs sampling iteratively samples the topic assignment of each word to approximate the posterior distribution. The number of iterations is usually set to 1000ˉ2000, and each iteration needs to traverse all words in the document set. For each document, traverse each word position in the document, calculate the posterior probability of each topic based on the topic assignment of other words and the current model parameters, and perform topic sampling and updating. During sampling, optimization techniques such as collapsed Gibbs sampling and topic block sampling can be used to improve sampling efficiency and convergence speed. In each iteration of Gibbs sampling, the posterior probability of each topic is calculated based on the topic assignment of the current word and the topic assignment of other words, and topic sampling is performed based on the posterior probability to update the topic assignment of the current word and update the corresponding topic-word distribution and document-topic distribution.After the iteration, the parameters of the LDA model are estimated according to the obtained topic-word distribution and document-topic distribution, to obtain the word distribution of each topic and the topic distribution of each document. The probability distribution can be obtained by normalizing the frequency matrix, or the sampling results of Gibbs sampling can be used to calculate the posterior expectation of the word distribution of each topic and the topic distribution of each document as the parameter estimation value. Using the trained LDA model, the theme of a new document is inferred and classified, and the topic distribution of the document is obtained by calculating the posterior probability of the document on each topic, and the topic with the highest probability is selected as the topic label of the document. Using the grid equipment account, topology structure, backup line, fault data and other documents, a document-word frequency matrix of 3000 devices, 5000 lines and 1000 topic words is obtained. Through the theme consistency evaluation and the perplexity curve, the optimal number of themes is determined to be 50. The theme-word distribution parameter β and the document-theme distribution parameter θ are initialized, and the prior parameters α and η are set to 1 and 0.1, respectively. The collapsed Gibbs sampling algorithm is used for LDA model training, and the iteration is performed for 1500 times, and each iteration takes 10 seconds. The sampling process is parallelized using OpenMP multi-threading to speed up the training time. The 50*1000 theme-word distribution matrix and the 3000*50 document-theme distribution matrix are obtained by training, and the Top10 coverage rate of the theme word is 85%. 500 new device account documents are selected for theme inference, and the average inference time is 20ms / document. The theme distribution features of each document are extracted, combined with the TF-IDF features, and a support vector machine classification model is trained, which achieves an accuracy of 90% in the device state classification task. The LDA model is applied to the theme clustering of distribution equipment, and 10 typical device combination modes are found, such as "oil-immersed transformer-vacuum circuit breaker-cable", etc., which match 90% of the device combinations in the expert knowledge base. The Word2Vec word embedding representation is introduced to map the theme-word distribution to a low-dimensional semantic space, and the t-SNE algorithm is used to visualize the theme distribution, which clearly presents the semantic structure of the distribution professional vocabulary, and finds the key theme clusters such as "fault warning" and "condition-based maintenance".

[0042] In S108, a feasibility report generation model is constructed based on the redundancy information, and the feasibility report generation model is used to analyze potential themes and patterns in the redundancy information to generate a power distribution network redundancy feasibility report.

[0043] Based on the redundancy information of each link of the power distribution network, the LDA topic model is trained to obtain the index distribution of each topic and the topic distribution of each device or node, and the potential semantic structure and association mode in the redundancy information are mined; the topic-index distribution is analyzed, each topic is explained and named, the redundancy mode and characteristics represented by the topic are summarized, the rationality and actual significance of each topic are evaluated, and representative redundancy topics are screened out; using the topic distribution of the device or node, the devices or nodes with similar redundancy characteristics are classified into the same category, the classification of the redundancy devices or nodes is formed, and the redundancy level and risk degree of each category are analyzed; the redundancy feasibility report generation model is constructed by comprehensively considering the redundancy mode mined by the LDA topic model and the device / node classification result, and the corresponding feasibility analysis report is generated according to the topic distribution characteristics of the device or node.

[0044] Specifically, the redundancy information of each link of the distribution network can also include the redundancy indicators of each device and node such as power supply, substation, line, switch, and load, the data sources can include power monitoring system, device management system, fault diagnosis system, etc., and the data formats can include CSV, JSON, Excel, etc. The redundancy information of each link of the distribution network is cleaned, standardized and integrated, and problems such as missing values, outliers and inconsistency are processed, a unified data format and representation method is constructed, a standardized redundancy dataset is formed, and data preparation is made for subsequent topic mining and pattern analysis. The standardized redundancy dataset is analyzed exploratorily, the data is read using the pandas library, the statistical characteristics of each redundancy indicator such as mean, variance and quantile are calculated using the describe() function, the grouped analysis is performed using the groupby() function, the index distribution histogram and box plot are drawn using the matplotlib library, the correlation heat map is drawn using the seaborn library, etc. Through visual analysis, outliers and outliers in the data are identified, the distribution characteristics and correlation patterns of the data are found, the abnormal data is checked and processed, and the data quality and consistency are ensured. The redundancy dataset is converted into a document-term frequency matrix or a TF-IDF matrix, each document corresponds to a device or node, the vocabulary is composed of redundancy indicators, and the elements in the matrix represent the values or weights of the indicators on the device or node. An input representation suitable for LDA model is constructed. On the basis of the converted redundancy data, the LDA topic model is trained. In order to optimize the performance of the model, various evaluation indicators such as perplexity, coherence score and topic distinctiveness are used to quantify the effect of the model. Through the implementation of grid search strategy, the key hyperparameters of LDA model are adjusted systematically, which include the number of topics, alpha (controls the smoothness of word distribution within a topic) and beta (controls the smoothness of topic distribution between documents). The goal is to find the combination of hyperparameters that achieve the best balance of evaluation indicators, so as to ensure that the model can accurately capture the topic structure in the data and exhibit good interpretability. The generalization performance of the model is evaluated using methods such as 5-fold cross-validation, the index distribution of each topic and the topic distribution of each device or node are obtained, and the potential semantic structure and correlation pattern in the redundancy information are mined. The topic-index distribution obtained by analyzing the LDA model is analyzed, each topic is explained and named, the keywords of each topic can be sorted using TF-IDF weighting to extract the keywords as topic labels, or the semantic similarity of topic words can be calculated using word embedding method to cluster and generate topic labels.In summary, the redundancy patterns and characteristics of each topic are summarized based on domain knowledge and expert experience, such as "high redundancy topic" and "low redundancy topic". The rationality and practical significance of each topic are evaluated, and important and representative redundancy topics are selected. The topic distribution of devices or nodes obtained by the LDA model is used to cluster and group devices or nodes, and devices or nodes with similar redundancy characteristics are classified into the same category to form a classification system of redundancy devices or nodes. The redundancy level and risk degree of each category are analyzed to identify key redundancy weaknesses and potential improvement opportunities. The LDA topic model is used to mine the redundancy patterns, device / node classification results, and domain expert knowledge to construct a redundancy feasibility report generation model. A template-based natural language generation method is used to predefine the framework and chapter structure of the feasibility report, such as "current situation evaluation", "problem analysis", and "improvement suggestions". According to the topic distribution characteristics of devices or nodes, the corresponding content templates are filled in to generate a complete feasibility report. Deep learning text generation models such as GPT and BERT can also be used to fine-tune on large-scale report corpus to generate more flexible and personalized feasibility reports. From the power distribution automation system and device management system, 10kV and 35kV distribution network redundancy data are extracted, including 110 main transformers, 326 feeder lines, 584 switches, and 1320 ring network cabinets. A total of 2GB, 15,000 records of redundancy data set are formed. The Pandas library in Python is used to clean the data, remove fields with more than 20% missing values, fill in the fields with less than 20% missing values using nearest neighbor interpolation, normalize the numerical indicators using max-min, and one-hot encode the enumeration indicators to form a standardized data representation. Matplotlib is used to draw the frequency distribution histogram of the redundancy indicators, and the outliers below the 5% quantile and above the 95% quantile are truncated. The TF-IDF algorithm is used to convert the redundancy data into a 1320x50 document-term frequency matrix, with each row representing a ring network cabinet and each column representing a TF-IDF weighted redundancy indicator. The Gensim library is used to train the LDA topic model, with 10 topics, α = 0.1, β = 0.01, and 1000 iterations. The CoherenceScore is used as the evaluation indicator, and the average CoherenceScore obtained by 5-fold cross-validation is 0.76. The top 5 keywords of each topic are extracted, such as "capacity margin", "power redundancy", "switch configuration", "line core number", and "connection mode". The topics are manually named as "high redundancy", "medium redundancy", "low redundancy", "N-1 redundancy", and "N-2 redundancy". The t-SNE algorithm is used for dimensionality reduction visualization of document topic distribution, and it is found that the ring network cabinets with "high redundancy" topic are clustered in 2-dimensional space, while the ring network cabinets with "low redundancy" topic are scattered.The K-Means algorithm is used to cluster the ring network cabinets, and the number of clusters is 5, and the number of ring network cabinets of each redundancy level is 205, 326, 418, 256, and 115 respectively. The Jinja2 template engine is used to automatically generate a feasibility analysis report, and the report content includes "overall redundancy level evaluation of distribution network: the average redundancy index of the distribution network is 0.72, which is at a medium level. Among them, the redundancy of the 35kV power grid is better than that of the 10kV power grid, and the redundancy of the rural power grid is lower than that of the urban power grid"; it can also include "weak link analysis: the ring network cabinets with numbers #254 and #380 have the lowest redundancy level, and there are weak links such as "single line" switch missing, and it is recommended to add standby lines and add tie switches"; it can also include "improvement measures and suggestions: for the 110 main transformers on the power supply side, increase the capacity margin and expand the standby capacity; for the 326 feeder lines on the line side, increase the line core number and add standby lines; for the 584 switches on the switch side, optimize the jumper mode and add bypass standby". The feasibility analysis report generated by the expert review has an accuracy of 95%, which can be used to guide the planning and construction of the distribution network and improve the power supply reliability of the distribution network.

[0045] The feasibility report generation model combines a local client or a database to generate a distribution network redundancy feasibility report, which includes redundancy calculation results, resource utilization analysis, and feasibility evaluation, and performs rule checking.

[0046] Read data including equipment account, topology structure, and redundancy calculation results of the distribution network; call the feasibility report generation model, fill in the pre-defined report template, and generate a preliminary draft of the distribution network redundancy analysis report; correlate and integrate the generated preliminary draft of the distribution network redundancy analysis report with the distribution network resource utilization analysis results, evaluate the balance between the distribution network redundancy level and resource utilization efficiency, and form a feasibility analysis conclusion; integrate the generated preliminary draft of the distribution network redundancy analysis report with the feasibility analysis conclusion; perform rule checking on the integrated distribution network redundancy feasibility analysis report; optimize and polish the distribution network redundancy feasibility analysis report after rule checking; digitally compile the optimized and polished distribution network redundancy feasibility analysis report; check and approve the compiled distribution network redundancy feasibility analysis report, and form the final distribution network redundancy feasibility analysis report.

[0047] Specifically, a data connection is established between the local client and the database, and data interaction and synchronization between the client and the database are realized through database connection technologies such as JDBC or ODBC. The database uses a relational database such as MySQL or Oracle, and device tables, line tables, node tables, etc. are designed to represent the topological connection relationship between devices through foreign key association. The client reads the basic data of the power distribution network from the database, such as device account, topological structure, and redundancy calculation results, to ensure data consistency and real-time performance. In the local client, based on the basic data of the power distribution network obtained from the database, the feasibility report generation model is called, and Java, Python, etc. are used to develop the feasibility report generation model based on document template engines such as Freemarker and Velocity. By reading the basic data of the power distribution network, the pre-defined report template is filled in to generate the first draft of the power distribution network redundancy analysis report. In the local client, the report generation model is called through API interface or command line. The report content includes overall assessment of power distribution network redundancy, weak link analysis, and potential improvement opportunities, forming the basic framework and content of the report. The first draft of the automatically generated power distribution network redundancy analysis report is associated and integrated with the power distribution network resource utilization analysis results. Data warehouse technology is used to load the redundancy analysis results and resource utilization analysis results into the same data mart or data set, and then SQL statements or OLAP tools are used for data association query and aggregation analysis to evaluate the balance between power distribution network redundancy level and resource utilization efficiency, and form a comprehensive feasibility analysis conclusion. The integrated power distribution network redundancy feasibility analysis report is subjected to comprehensive rule checking and quality checking. According to the power distribution network planning and design specifications and standards such as "Electric Power Planning and Design Technical Regulations" (DL / T5729-2016) and "20kV-110kV Power Distribution Network Planning and Design Technical Guidelines" (Q / GDW1738-2012), these specifications and standards are converted into executable checking rules and embedded into the feasibility report generation model to check the compliance of the redundancy indicators and resource utilization indicators in the report, identify errors, omissions, inconsistencies, etc. in the report, and prompt the user to correct and improve. Based on the rule checking, the power distribution network redundancy feasibility analysis report is further optimized and polished. Natural language processing tools such as StanfordCoreNLP and NLTK are used to process the report text, including word segmentation, part-of-speech tagging, and syntax analysis. Then, text summarization and text rewriting techniques are used to optimize the language expression of the report. Text generation models such as seq2seq and Transformer can be used in combination with power distribution network professional corpora to transform and polish the semantics of the report, improve the readability and professionalism of the report, and generate high-quality reports that meet industry standards and user needs.The optimized power distribution network redundancy feasibility analysis report is compiled and published in digital form. Java libraries such as iText and Apache POI are used to convert the generated HTML format report into PDF format. By embedding Microsoft Office, WPS Office, and LaTeX electronic document generation tools in the local client, the report is typeset and organized in the form of printable and interactive electronic documents. JavaScript chart libraries such as ECharts and Highcharts are used to embed interactive charts and data visualization components in the HTML report, enhancing the readability and interactivity of the report and achieving digital compilation and publication of the report. The compiled power distribution network redundancy feasibility analysis report is subjected to multi-level review and approval. The Activiti workflow engine is used to define the process model of report review and approval, including various approval nodes, approval roles, and approval conditions. The process model is deployed to the workflow system. In the local client, the report review and approval process is initiated by calling the REST API of the workflow engine, and the execution status and results of the process are tracked. The JIRA task management system is used to create task sheets for report review and approval, which are assigned to relevant experts and managers. The execution process and feedback of the task are recorded, and the key content, conclusions, and recommendations in the report are reviewed and confirmed to ensure the accuracy, reliability, and executability of the report. Finally, a formal power distribution network redundancy feasibility analysis report is formed, providing decision support for power distribution network planning, construction, and operation. MySQL relational database is used to design power distribution network resource data tables, including 10 data tables such as device table, line table, and node table, with a total of 120 fields and a data volume of 10 GB. JDBC connection pool technology is used to set the maximum number of connections to 50 and the connection timeout to 5 seconds, achieving efficient connection between the client and the database. Based on the Flask web framework of Python and the Jinja2 template engine, a power distribution network redundancy analysis report generation service is developed, with more than 10 report templates defined, covering overall assessment, regional assessment, and weak link analysis of power distribution networks. The Pandas data analysis library is used to perform correlation analysis on redundancy analysis results and resource utilization rate data, generating more than 20 pivot tables and cross tables, and using the Matplotlib visualization library to generate 10 different types of charts. The long short-term memory neural network (LSTM) model is applied to the 3 million historical operation data of the power distribution network as training corpus, and through sequence-to-sequence learning, a power distribution network operation analysis report is automatically generated at an average speed of 1000 words per second. Jenkins continuous integration environment is built, and the SonarQube code quality management platform is used to check the automatically generated report source code in terms of code specification, security, and reliability, and more than 30 code defects and security vulnerabilities are found and fixed.Using the deep learning library PyTorch, a Transformer text polishing model is built. With 5000 high-quality power distribution network analysis reports written by artificial as training corpus, the automatically generated reports are corrected in text, optimized in semantics and polished in language, so that the readability of the reports is improved by more than 25% on average. Using ApachePDFBox library, the generated HTML report is converted into PDF format, and using OpenCV computer vision library, the charts, formulas and tables in the report are identified and extracted, the report directory and index are automatically generated, and the interactive and searchable electronic report format is formed. On the Activiti workflow platform, based on the BPMN2.0 standard, the power distribution network redundancy analysis report approval process is defined, including 6 links such as expert review, countersignature, final review, etc. The average approval time is shortened to within 2 working days. Using Kafka distributed message flow platform, the real-time feedback and triggering of report approval opinions are realized, and the JIRA task management system is connected, realizing the closed-loop tracking management of problems and defects.

[0048] The above is only an example and description of the structure of the present application, and those skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present claims, which shall belong to the protection scope of the present application.< / circle>

Claims

1. A power distribution network feasibility study report automatic generation method based on an LDA probability model, characterized in that, The method comprises: obtaining power grid equipment account information and its electrical topology connection relationship, generating an account document containing device name, type, location, and connection relationship information; constructing a graph structure of the benchmark power grid based on the obtained account attributes and electrical topology relationship, and labeling the standby line and standby equipment data to generate a labeling document containing the benchmark power grid structure and labeling information; calculating the redundancy of each node in the distribution network based on the standby line and standby equipment data in the benchmark power grid structure, combining the number of alternative paths between different devices and lines, generating an electrical topology graph containing redundancy information and a redundancy information document containing redundancy calculation results; based on the redundancy index calculated by the node, analyze the fault recovery capability of the distribution network under different redundancy levels, evaluate the switching and recovery speed when a fault occurs, and generate a fault analysis document containing the fault recovery capability analysis results and the evaluation results of the switching and recovery speed when a fault occurs; evaluate the resource utilization of the distribution network according to the historical fault data of the distribution network, judge the idle situation of standby equipment and lines in the non-fault state, and generate an evaluation document containing historical fault data and resource utilization evaluation results; summarize the account document, labeling document, redundancy information document, fault analysis document and evaluation document of the power grid equipment, preprocess, construct a vocabulary table, represent each document as a word frequency vector in the vocabulary table, generate a document-word frequency matrix, the row of the matrix represents the document, the list represents the vocabulary, and the value in the matrix represents the frequency of the vocabulary in the document; train the LDA model, set the number of topics, and perform iterative training to extract the topic distribution of each document and the vocabulary distribution of each topic; based on the redundancy information, construct a feasibility report generation model, which is used to analyze the potential topics and patterns in the redundancy information, and generate a distribution network redundancy feasibility report.

2. The method of claim 1, wherein, The method comprises: obtaining power grid equipment account information and its electrical topology connection relationship, generating an account document containing device name, type, location, and connection relationship information; obtain and analyze the graph model file, extract the device symbol and text annotation information in the graph model file to obtain the attribute information of the name, type and location of the power grid equipment, and construct a device attribute table; analyze the graphical elements of the connection lines and arrows in the graph model file, combine the device location information in the device attribute table, determine the electrical connection relationship between devices, and construct a device topology connection relationship table; integrate the device attribute table and the device topology connection relationship table to form an account document; perform attribute analysis on the generated account document; according to the attribute analysis result, mine the implicit relationship and frequent pattern between the attributes, and find out the related attribute combination which can form context logical connection and thinking chain; use the found related attribute combination, combine the business characteristics and professional knowledge of the power grid equipment, and expand and / or refine the account document; perform logical checking and consistency checking on the expanded and / or refined account document.

3. The method of claim 1, wherein, The acquired account attributes and electrical topology relationship are constructed as a graph structure of the benchmark power grid, and the backup line and backup device data are labeled to generate a labeling document containing the benchmark power grid structure and labeling information, including: According to the extracted account attributes and electrical topology relationship, nodes and relationships are created, devices are taken as nodes, and topology connection relationships are taken as edges to construct a graph structure of the benchmark power grid, including labeling corresponding labels for backup lines and backup devices to distinguish main lines and backup lines and main devices and backup devices, and for backup lines, labeling the corresponding main line information to form a backup relationship mapping; The key nodes and key paths in the power grid topology structure are analyzed, the importance and influence range of the nodes are evaluated, and the network characteristic parameters of the power supply path and load distribution are generated according to the topology structure and device attributes; For the device nodes in the graph structure, the account attribute information is associated, and for the backup devices, the attributes of the operation condition and switching mode are associated to form a backup device attribute portrait; The graph structure and the geographical information of the device are integrated, the nodes are mapped to the map according to the geographical position attributes of the device to form a power grid geographical topology map to show the spatial distribution and geographical wiring relationship of the power grid; Based on the constructed graph structure of the benchmark power grid, a labeling document containing device nodes, topology connections and backup labeling information is generated, and the labeling document includes distinguishing different types of devices and connections through the visual attributes of the color and shape of the nodes.

4. The method of claim 1, wherein, According to the backup line and backup device data in the benchmark power grid structure, the redundancy of each node in the distribution network is calculated based on the number of alternative paths between different devices and lines, and an electrical topology graph containing redundancy information and a redundancy information document containing redundancy calculation results are generated, including: The connection relationships between nodes in the distribution network are analyzed to construct an adjacency matrix representing the network topology structure by analyzing the benchmark power grid structure; All reachable nodes of the adjacency matrix are traversed, and the paths from the starting node to each reachable node are recorded to obtain the alternative path set between all nodes in the distribution network, and each path in the alternative path set represents a potential power supply scheme; According to the attribute information of the backup line and the backup device, the alternative paths are screened and sorted, the paths containing the backup line and the backup device are selected, and the paths corresponding to the faulty devices and the faulty lines are excluded to obtain the effective alternative path set of each node; Combined with the effective alternative path set and the benchmark power grid structure, the redundancy index of each node is calculated, and the redundancy index represents the power supply reliability of the node; According to the calculation results of the redundancy index, the nodes of the distribution network are classified and sorted to generate a distribution network redundancy level division table; The calculation results of the node redundancy index are associated with the geographical topology information of the distribution network to label the nodes of different redundancy levels with different colors or patterns to generate an electrical topology graph integrating redundancy information to show the power supply reliability distribution of the distribution network. The data and results in the node redundancy analysis process are summarized, including the node adjacency matrix, the set of alternative paths, the set of effective alternative paths, the node redundancy index, the redundancy level division, and the power distribution network redundancy level division table, to generate an electrical topology map containing redundancy information and a redundancy information document of redundancy calculation results.

5. The method of claim 1, wherein, Based on the node-calculated redundancy index, the fault recovery capability of the power distribution network under different redundancy levels is analyzed, and the switching and recovery speed when a fault occurs is evaluated, to generate a fault analysis document containing the fault recovery capability analysis results and the evaluation results of the switching and recovery speed when a fault occurs, including: Based on the power distribution network redundancy level division table, fault simulation analysis is performed for each redundancy level, including calculating the power supply recovery time of the node under fault conditions according to the number of alternative paths and the power supply capacity of the node, to obtain fault recovery time data under different redundancy levels; The influence of automatic switching mode and manual switching mode on fault switching time is analyzed to obtain fault switching time data; The fault recovery capability of the power distribution network under different redundancy levels is evaluated using the fault recovery time data and the fault switching time data, including calculating the fault recovery capability index; The correlation between the fault recovery capability index and the redundancy index is analyzed, and the functional relationship between the redundancy index and the fault recovery capability index is fitted to obtain the influence law of the redundancy index on the fault recovery capability, forming a quantitative relationship model of the redundancy index and the fault recovery capability; The evaluation results of the fault recovery capability of the power distribution network under different redundancy levels and the quantitative relationship model of the redundancy index and the fault recovery capability are summarized to generate a fault analysis document containing the fault recovery capability analysis results and the evaluation results of the switching and recovery speed when a fault occurs.

6. The method of claim 1, wherein, Based on the historical fault data of the power distribution network, the resource utilization rate of the power distribution network is evaluated in combination with the redundancy index to judge the idle condition of the standby equipment and lines under non-fault state, to generate an evaluation document containing historical fault data and resource utilization rate evaluation results, including: Obtain the historical fault records of the power distribution network, including fault occurrence time, fault equipment, fault type, influence range, and repair time information; Preprocess the historical fault records of the power distribution network; Form a power distribution network fault data set by summarizing the preprocessed historical fault records of the power distribution network; According to the topological structure and equipment account information of the power distribution network, the spatial mapping of the power distribution network fault data set is performed to determine the specific location of each fault event; Associate the specific location with the attribute information of the fault equipment to form the attribute profile of the fault equipment; Analyze the fault characteristics and laws of different types of equipment using the attribute profile of the fault equipment; Based on the redundancy information document, the power distribution network fault data set is analyzed to calculate the fault rate, fault duration, and fault influence range index under each redundancy level to evaluate the influence of the redundancy level on the reliability of the power distribution network; Based on the evaluation results, a quantitative relationship between redundancy and reliability is established; Analyze the relationship between faults and types and redundancy to find out the causes and risk factors of faults; Evaluate the utilization efficiency of backup resources, identify the optimal balance point of redundancy and resource utilization rate; Construct a fault prediction model for predicting the probability and type of equipment failure; Summarize the analysis results of the failure characteristics and laws of different types of equipment, the quantitative relationship between redundancy and reliability, the causes and risk factors of failure, the evaluation results of backup resource utilization efficiency and the fault prediction results, and generate an evaluation report of the historical fault data analysis and resource utilization rate results of the power distribution network.

7. The method of claim 1, wherein, The account documents, labeling documents, redundancy information documents, fault analysis documents and evaluation documents of the power grid equipment are preprocessed, a vocabulary table is constructed, each document is represented as a word frequency vector in the vocabulary table, a document-word frequency matrix is generated, the rows of the matrix represent the documents, the columns represent the words, and the values in the matrix represent the frequency of the words in the documents, including: The account documents, labeling documents, redundancy information documents and fault analysis documents of the power grid equipment are summarized, the contents of the summarized documents are converted into plain text format, and non-text elements including pictures and tables are removed to obtain a pure text dataset; The pure text dataset is preprocessed, including word segmentation, stop word removal and part-of-speech tagging operations; The preprocessed pure text dataset is part-of-speech tagged to form a vocabulary table; Based on the vocabulary table, the frequency of each word is counted, and words with a frequency exceeding a threshold value are selected as keywords to form a keyword list; By traversing each document, the frequency of each word in the keyword list in the document is counted to form a word frequency vector of the document, the dimension of the word frequency vector is equal to the length of the keyword list, and each element of the word frequency vector represents the frequency of the corresponding keyword in the document; Based on the word frequency vectors of the documents, a document-word frequency matrix is constructed, the rows of the matrix represent the documents, the columns represent the words, and the values in the matrix represent the frequency of the words in the documents.

8. The method of claim 1, wherein, The LDA model is trained, the number of topics is set, and iterative training is performed to extract the topic distribution of each document and the word distribution of each topic, including: According to the document-word frequency matrix, the input of the LDA topic model is constructed, each document is represented as a frequency vector or weighted vector of words and is normalized to ensure that the sum of the vector elements is 1; The number of topics of the LDA model is set in combination with the document characteristics and domain knowledge; The parameters of the LDA model are initialized, including the topic-word distribution parameters and the document-topic distribution parameters; The LDA model is trained, and the optimal topic-word distribution and document-topic distribution are gradually converged; The posterior probability of each topic is calculated, and the topic-word distribution and document-topic distribution are updated; Based on the updated topic-word distribution and document-topic distribution, the word distribution of each topic and the topic distribution of each document are obtained.

9. The method of claim 1, wherein, The redundancy information is used to construct a feasibility report generation model, which is used to analyze the potential topics and patterns in the redundancy information and generate a redundancy feasibility report for the power distribution network, including: Based on the redundancy information of each link of the power distribution network, the LDA topic model is trained to obtain the index distribution of each topic and the topic distribution of each device or node, and the potential semantic structure and correlation pattern in the redundancy information are mined; The theme-index distribution is analyzed, each theme is explained and named, the redundancy pattern and characteristics represented by the theme are summarized, the rationality and actual significance of each theme are evaluated, and representative redundancy themes are selected; Using the topic distribution of the device or node, devices or nodes with similar redundancy characteristics are classified into the same category, forming the classification of redundancy devices or nodes, and the redundancy level and risk degree of each category are analyzed; Based on the redundancy patterns mined by the LDA topic model and the classification results of the devices or nodes, a redundancy feasibility report generation model is constructed, and according to the topic distribution characteristics of the devices or nodes, the corresponding feasibility analysis report is generated; The feasibility report generation model combines the local client or database to generate a power distribution network redundancy feasibility report, which includes the calculation results of redundancy, resource utilization analysis and feasibility evaluation, and performs rule checking.

10. The method of claim 9, wherein, The feasibility report generation model combines the local client or database to generate a power distribution network redundancy feasibility report, which includes the calculation results of redundancy, resource utilization analysis and feasibility evaluation, and performs rule checking, including: Reading data including device account, topology structure, redundancy calculation results of the power distribution network; Calling the feasibility report generation model, filling in the pre-defined report template, and generating the first draft of the power distribution network redundancy analysis report; Integrating the generated first draft of the power distribution network redundancy analysis report with the power distribution network resource utilization analysis results, evaluating the balance between the power distribution network redundancy level and resource utilization efficiency, and forming the feasibility analysis conclusion; Integrating the generated first draft of the power distribution network redundancy analysis report with the feasibility analysis conclusion; Performing rule checking on the integrated power distribution network redundancy feasibility analysis report; Optimizing and polishing the power distribution network redundancy feasibility analysis report after rule checking; Digitizing the optimized and polished power distribution network redundancy feasibility analysis report; Checking and approving the prepared power distribution network redundancy feasibility analysis report to form the final power distribution network redundancy feasibility analysis report.

Images