A knowledge graph-based power distribution network engineering whole-process management and control method and system

Abstract

This invention discloses a method and system for full-process management and control of agricultural power grid projects based on knowledge graphs. The method includes: acquiring agricultural power grid project data, performing preprocessing, and generating an agricultural power grid data set; constructing an engineering knowledge graph based on the entity set and relation set in the agricultural power grid data set; constructing a path reasoning state vector based on the selected starting node and target node in the engineering knowledge graph to obtain the initial input for path reasoning; inputting the initial input for path reasoning into a policy network; performing multi-hop path expansion and path validity determination on the updated path sequence; performing link type discrimination and anomaly detection; locating corresponding nodes in the entire project process; and updating node attributes and relation edges in the engineering knowledge graph. This invention uses knowledge graphs and multi-hop path reasoning to achieve anomaly identification and dynamic management and control throughout the entire process of agricultural power grid projects.

Description

Technical Field

[0001] This invention relates to the field of digital management and control technology for agricultural power grid projects, and in particular to a method and system for full-process management and control of agricultural power grid projects based on knowledge graphs. Background Technology

[0002] Throughout the planning, design, construction, acceptance, and settlement processes of rural power grid projects, a large amount of diverse engineering data is generated, including construction event records, material allocation information, process execution status, risk events, and cost settlement information. Existing information management methods mainly rely on manual data entry, process-driven management systems, or static rule engines. These methods suffer from weak correlation between data, failing to provide a unified representation of the relationship links between project stages and hindering the systematic analysis of complex engineering processes.

[0003] With new data constantly being generated at the engineering site, the existing system lacks the ability to automatically understand dynamic relationship changes and cannot update the correlation structure of the engineering process in a timely manner based on newly generated events.

[0004] Traditional methods for identifying engineering anomalies typically rely on fixed thresholds or single-point detection, such as making single-factor judgments based on abnormal material quantities, process delays, or frequent risk events. These methods often ignore the interconnected paths between different engineering entities, failing to identify construction risk chains, material declaration chains, process execution chains, and settlement cost chains at the overall chain level, making it difficult to achieve comprehensive anomaly analysis across stages and entities.

[0005] Meanwhile, traditional processes cannot provide the ability to locate key nodes, resulting in the inability to promptly identify risk nodes, material anomaly nodes, process breakpoint nodes, and settlement anomaly nodes; as the engineering process changes dynamically, the existing system also lacks a real-time update mechanism based on incremental data, making it impossible to form a continuously iterative full-process control result. Summary of the Invention

[0006] One objective of this invention is to propose a knowledge graph-based method and system for the full-process management and control of rural power distribution network projects. This method fully utilizes engineering knowledge graph modeling technology, reinforcement learning-driven path reasoning strategies, link-level anomaly identification models, and node-level localization models to perform structured representation, relational mining, and dynamic updating of data throughout the entire process of rural power distribution network projects, from planning, design, construction, acceptance to settlement. The invention details the algorithmic processes for engineering link identification, anomaly link detection, and anomaly node localization. By achieving continuous updating of the engineering knowledge graph under incremental data conditions, this invention possesses advantages such as high sensitivity to changes in the engineering process, high precision in anomaly identification, and strong dynamic management and control capabilities throughout the entire process.

[0007] A knowledge graph-based method for full-process management and control of agricultural power grid projects according to an embodiment of the present invention includes: Acquire multi-source heterogeneous data from the planning, design, construction, acceptance and settlement stages of the rural power grid project, perform preprocessing on the multi-source heterogeneous data, and generate a rural power grid data set; Based on the entity set and relation set in the agricultural distribution network dataset, an engineering knowledge graph containing the entity node set and relation edge set is constructed. In the engineering knowledge graph, the starting node and target node to be analyzed are determined, and the initial input for path reasoning is formed based on the current node representation, the target node representation, and the path history sequence. The current set of possible actions is generated based on the initial input of path reasoning. The action probability of the set of possible actions is calculated based on the policy network to determine the next hop action and update the path sequence. Based on the updated path sequence, action selection and path expansion are performed to form a set of inference paths. The validity of the inference path set is then determined to obtain a set of valid inference paths. Based on the set of valid reasoning paths, the construction risk links, material application links, process execution links and settlement cost links of the rural power grid project are identified, and the set of abnormal links is obtained. Based on the abnormal link set, the corresponding nodes in the entire process of the project are located, and early warning results and verification results are generated; Collect newly added project data during the implementation of the agricultural power grid project, update the node attributes and relationship edges in the project knowledge graph, and form a full-process control result.

[0008] Optionally, the preprocessing includes field parsing, time alignment, entity extraction, and relationship identification.

[0009] Optionally, the construction of an engineering knowledge graph containing a set of entity nodes and a set of relation edges based on the entity set and relation set in the agricultural distribution network data set specifically includes: Extract the entity set from the agricultural distribution network data set, and generate project nodes, process nodes, equipment nodes, material nodes, construction event nodes, risk event nodes, and settlement cost nodes respectively to form an initial set of nodes; Each node in the initial set of nodes is assigned node attribute values, including node identifier, node type, project stage information, and node association time information; Extract the relation set from the agricultural network dataset, and connect the nodes in the initial node set to form the initial set of relation edges; Each relation edge in the initial set of relation edges is assigned a relation attribute value, which includes relation type, relation direction, node identifier corresponding to the relation, and project stage identifier corresponding to the relation. The initial set of nodes and the initial set of relation edges are combined to form a knowledge graph for agricultural power grid projects that includes the set of entity nodes and the set of relation edges.

[0010] Optionally, the step of determining the starting node and target node to be analyzed in the engineering knowledge graph, and forming the initial input for path reasoning based on the current node representation, the target node representation, and the path history sequence specifically includes: In the entity node set of the engineering knowledge graph, the target analysis type is determined according to the control object of the rural power grid project, and the target node and the starting node are selected from the entity node set according to the target analysis type; The starting node is determined as the current node. The node attribute information of the current node and the relationship edge information directly connected to the current node are obtained. The node representation of the current node is generated. The node attribute information of the target node is obtained. The node representation of the target node is generated. Construct a path history sequence based on the node access order and relation edge connection order recorded in the engineering knowledge graph; The node representations of the current node, the target node, and the path history sequence are combined and arranged according to a fixed structure to generate a path reasoning state vector, which is then used as the initial input for path reasoning.

[0011] Optionally, the step of generating a current set of possible actions based on the initial input of path reasoning, calculating the action probability of the set of possible actions based on the policy network, determining the next hop action, and updating the path sequence specifically includes: The current node is determined based on the path reasoning state vector. The relationship edge directly connected to the current node is retrieved from the relationship edge set of the engineering knowledge graph to form the current candidate action set. Extract the relation type identifier, next node identifier, and current node representation from the candidate actions to generate an action feature sequence; The path reasoning state vector is concatenated with the action feature sequence to form the input data of the policy network. The input data is then fed into the policy network to obtain the action probability. Based on the action probability, select candidate actions whose action probabilities satisfy the action selection criteria from the current candidate action set, and set the next node contained in the candidate action as the updated current node. The relation edge identifiers and next node identifiers of candidate actions are appended to the path history sequence in sequence. The updated current node is combined with the updated path history sequence to generate a new path reasoning state vector.

[0012] Optionally, the step of selecting actions and expanding paths based on the updated path sequence to form a set of inference paths, and determining the validity of the set of inference paths to obtain a set of valid inference paths specifically includes: The new current node is determined based on the new path reasoning state vector, new candidate actions are obtained, and the new candidate actions are added to the candidate action set. Extract the relation type identifier, the next node identifier, and the current node representation from the candidate actions to generate a candidate action feature sequence; The candidate action feature sequence is concatenated with the corresponding path reasoning state vector to form the path extended input data corresponding to the candidate action; Based on the path expansion input data corresponding to the candidate actions, select candidate actions that meet the path expansion conditions from the candidate action set, and add the candidate actions to the path history sequence to form a reasoning path set; The validity of each reasoning path in the reasoning path set is determined to obtain the set of valid reasoning paths.

[0013] Optionally, the identification of construction risk links, material application links, process execution links, and settlement cost links of the rural power grid project based on the set of valid reasoning paths, to obtain the set of abnormal links, specifically includes: For each inference path in the set of valid inference paths, a corresponding link feature sequence is formed; Based on the link feature sequence, the path length, the number of entity nodes, and the number of relation edges of the corresponding inference path are counted to generate a link feature vector. Input the link feature vector corresponding to each reasoning path into the link type discrimination model to divide the construction risk link set, material declaration link set, process execution link set and settlement cost link set; The construction risk link anomaly score is calculated based on the link feature vector, and the risk path is determined based on the construction risk link anomaly score and the construction risk anomaly threshold. Calculate the abnormal score of the material declaration link based on the link feature vector to determine the abnormal path of the material; Calculate the process execution link anomaly score based on the link feature vector to determine the abnormal process path; Calculate the settlement fee link anomaly score based on the link feature vector to identify settlement inconsistency paths; The risk path, material anomaly path, process anomaly path, and settlement inconsistency path are merged to form an anomaly link set.

[0014] Optionally, the step of locating the corresponding nodes in the entire project process based on the abnormal link set and generating early warning results and verification results specifically includes: For each abnormal link in the abnormal link set, obtain the link feature vector based on the path history sequence; For each entity node in each abnormal link, the node type information, the project stage information, and the node position number are arranged in a fixed order to generate a node feature vector. The node feature vector is then concatenated with the corresponding link features to generate a node positioning input vector. Input the node localization input vector into the node localization model, and output the node category identifier and process location identifier; Based on the node category identifiers output by the node classification sublayer, the nodes are categorized to form candidate sets for each node; Calculate the node warning score for each node in each candidate set and compare it with the node category warning threshold to obtain risk nodes, material abnormal nodes, process breakpoint nodes and settlement abnormal nodes. Risk nodes, material anomaly nodes, process breakpoint nodes, and settlement anomaly nodes are grouped according to node category identifiers and process location identifiers to generate corresponding early warning results. The node encoding vector is then combined with the link feature vector of its corresponding anomaly link to generate corresponding verification results.

[0015] Optionally, the process of collecting newly added project data during the implementation of the rural power grid project, updating the node attributes and relationship edges in the project knowledge graph, and forming a full-process control result specifically includes: During the implementation of the rural power grid project, new construction events, material allocation records, risk event records, and settlement update records are collected to form a new project data set. Perform preprocessing on various types of data in the newly added project dataset to generate a new entity set and a new relation set; Based on the newly added entity set and the newly added relation set, the entity node set and relation edge set in the engineering knowledge graph are updated to form the updated engineering knowledge graph; In the updated engineering knowledge graph, new starting nodes and new target nodes are determined based on the current control objects of the agricultural power grid project, and new path reasoning state vectors are generated. Based on the new path reasoning state vector, a new set of candidate actions is generated in the updated engineering knowledge graph, resulting in a new set of effective reasoning paths. New link feature vectors are generated based on the new set of effective inference paths, resulting in new early warning results and new verification results. The new early warning results and new verification results obtained from the updated engineering knowledge graph are combined with the historical early warning results and historical verification results in chronological order to form the full-process control results.

[0016] A knowledge graph-based end-to-end management and control system for agricultural power grid projects, comprising: The data processing module is used to acquire multi-source heterogeneous data from the agricultural power grid project, perform preprocessing, and generate an agricultural power grid data set. The graph construction module is used to construct an engineering knowledge graph based on the entity set and relation set in the agricultural distribution network data set; The path input generation module is used to construct a path reasoning state vector based on the selected starting node and target node in the engineering knowledge graph, and obtain the initial input for path reasoning. The strategy action module is used to input the initial input of path reasoning into the strategy network, perform action probability calculation on the candidate action set, and obtain the next hop action and the updated path sequence. The multi-hop path generation module is used to perform multi-hop path expansion and path validity determination on the updated path sequence to obtain a set of valid inference paths. The link identification module is used to construct link feature vectors based on the set of valid inference paths, perform link type discrimination and anomaly detection on the link feature vectors, and obtain a set of abnormal links. The node location module is used to locate the corresponding nodes in the entire process of the project based on the abnormal link set, and generate early warning results and verification results. The dynamic graph update module is used to collect new project data during the implementation of the agricultural power grid project, update the node attributes and relationship edges in the project knowledge graph, and form the full-process control results.

[0017] In its implementation, this invention constructs an engineering knowledge graph covering the entire process of rural power grid planning, design, construction, acceptance, and settlement. This allows multi-source heterogeneous data to be expressed under a unified structure, solving the problems of scattered engineering data and missing relationships in existing technologies. By continuously updating entity nodes and relational edges, structural changes during the engineering process can be captured in a timely manner, providing a data foundation for subsequent path reasoning and anomaly identification.

[0018] This invention employs a multi-hop path reasoning method to dynamically generate interconnected links in the engineering process. This enables the system to analyze construction risk links, material application links, process execution links, and settlement cost links from an overall link perspective, overcoming the shortcomings of traditional methods that can only make anomaly judgments based on local features. The construction of link feature vectors and the introduction of a link type discrimination model allow complex link structures to be processed quantitatively, improving the accuracy and scalability of link identification.

[0019] This invention further identifies key nodes in the abnormal process chain through a node localization model, enabling precise location of risk nodes, material anomaly nodes, process breakpoint nodes, and settlement anomaly nodes, thus improving the precision of engineering problem localization. Simultaneously, based on a dynamic update mechanism using an engineering knowledge graph, new construction events, material allocation information, risk event records, and settlement update records can be integrated into the reasoning process in real time, achieving continuous iterative updates of abnormal processes and early warning results, enhancing the timeliness and coverage of full-process control. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the multi-hop path generation and effective inference path acquisition process in this invention; Figure 3 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figure 1-3 A knowledge graph-based method for full-process management and control of agricultural power grid projects, including: Acquire multi-source heterogeneous data from the planning, design, construction, acceptance and settlement stages of the rural power grid project, perform preprocessing on the multi-source heterogeneous data, and generate a rural power grid data set; Based on the entity set and relation set in the agricultural distribution network dataset, an engineering knowledge graph containing the entity node set and relation edge set is constructed. In the engineering knowledge graph, the starting node and target node to be analyzed are determined, and the initial input for path reasoning is formed based on the current node representation, the target node representation, and the path history sequence. The current set of possible actions is generated based on the initial input of path reasoning. The action probability of the set of possible actions is calculated based on the policy network to determine the next hop action and update the path sequence. Based on the updated path sequence, action selection and path expansion are performed to form a set of inference paths. The validity of the inference path set is then determined to obtain a set of valid inference paths. Based on the set of valid reasoning paths, the construction risk links, material application links, process execution links and settlement cost links of the rural power grid project are identified, and the set of abnormal links is obtained. Based on the abnormal link set, the corresponding nodes in the entire process of the project are located, and early warning results and verification results are generated; Collect newly added project data during the implementation of the agricultural power grid project, update the node attributes and relationship edges in the project knowledge graph, and form a full-process control result.

[0023] In this embodiment, the preprocessing includes field parsing, time alignment, entity extraction, and relationship identification; Collect multi-source heterogeneous raw data generated during the planning, design, construction, acceptance, and settlement phases of the rural power grid project. The multi-source heterogeneous raw data includes planning scheme-related data, design document-related data, construction process-related data, acceptance record-related data, and settlement document-related data. The multi-source heterogeneous raw data is subjected to field parsing processing. Fields in various types of raw data are extracted, standardized, and formatted according to the field set. The field set includes field name, field type, field format, and information on the project stage to which the field belongs. Time alignment processing is performed on the data after field parsing to map the time information recorded in various types of data into a unified time series, which includes time tags and corresponding project stage identifiers. Entity identification is performed on time-aligned data to generate an entity set, which includes project entities, process entities, equipment entities, material entities, construction event entities, risk event entities, and settlement fee entities. Relationships are identified among entities in an entity set to generate a relationship set, which includes process dependencies, material relationships, event relationships, and cost relationships. The field set, unified time series, entity set, and relation set are combined to form the agricultural distribution network data set.

[0024] In this embodiment, the construction of an engineering knowledge graph containing a set of entity nodes and a set of relation edges based on the entity set and relation set in the agricultural distribution network data set specifically includes: Extract the entity set from the agricultural distribution network data set. Based on the project entity, process entity, equipment entity, material entity, construction event entity, risk event entity, and settlement cost entity of the entity set, generate project nodes, process nodes, equipment nodes, material nodes, construction event nodes, risk event nodes, and settlement cost nodes respectively, forming an initial set of nodes. Each node in the initial set of nodes is assigned node attribute values ​​according to the project stage, project type and project object. The node attributes include node identifier, node type, project stage information and node association time information. Extract the relationship set from the agricultural distribution network data set. Based on the process dependency relationship, material association relationship, event association relationship and cost association relationship in the relationship set, connect the nodes in the initial node set to generate process dependency relationship edge, material consumption relationship edge, event trigger relationship edge, risk association relationship edge and cost mapping relationship edge, forming the initial relationship edge set; Each relation edge in the initial set of relation edges is assigned a relation attribute value, which includes relation type, relation direction, node identifier corresponding to the relation, and project stage identifier corresponding to the relation. The initial set of nodes and the initial set of relation edges are combined to form a knowledge graph of the agricultural power grid project that includes the set of entity nodes and the set of relation edges; The engineering knowledge graph is used to represent the topological information of entity nodes and their relationships in the entire process of rural power grid engineering.

[0025] In this embodiment, determining the starting node and target node to be analyzed in the engineering knowledge graph, and forming the initial input for path reasoning based on the current node representation, the target node representation, and the path history sequence specifically includes: In the entity node set of the engineering knowledge graph, the target analysis type is determined according to the control object of the agricultural power grid project. The target node corresponding to the object to be analyzed is selected from the entity node set according to the target analysis type, and the starting node corresponding to the starting state of the agricultural power grid project is selected from the entity node set. The starting node is determined as the current node. The node attribute information of the current node and the relationship edge information directly connected to the current node are obtained. The node type, node identifier, project stage identifier and node time attribute in the node attribute information are arranged in a fixed order. The relationship type identifier and next node identifier in the relationship edge information are arranged in a fixed order. The two types of information are concatenated to generate the node representation of the current node. Obtain the node attribute information of the target node, and arrange the node type, node identifier, project stage identifier and node time attribute in the node attribute information in a fixed order to generate the node representation of the target node; A path history sequence is constructed based on the node access order and relation edge connection order recorded in the engineering knowledge graph. The path history sequence includes a historical node identifier sequence arranged in access order and a historical relation edge identifier sequence arranged in connection order. The node representation of the current node, the node representation of the target node, and the path history sequence are combined and arranged according to a fixed structure to generate a path reasoning state vector, and the path reasoning state vector is used as the initial input for path reasoning.

[0026] In this embodiment, the step of generating a current set of possible actions based on the initial input of path reasoning, calculating the action probability of the set of possible actions based on the policy network, determining the next hop action, and updating the path sequence specifically includes: The current node is determined based on the path reasoning state vector. The relationship edge directly connected to the current node is retrieved from the relationship edge set of the engineering knowledge graph. Each relationship edge and its corresponding next node are combined into a candidate action to form the current candidate action set. For each candidate action in the current candidate action set, extract the relation type identifier, the next node identifier, and the node representation of the current node from the candidate action, and arrange them in the order of "relation type identifier - next node identifier - current node representation" to generate an action feature sequence; The path reasoning state vector and the action feature sequence are concatenated in the order of "path reasoning state vector first, action feature sequence second" to form the input data of the policy network. The input data is then input into the policy network to obtain the action probability corresponding to each candidate action in the current candidate action set. Based on the action probability, select candidate actions whose action probabilities satisfy the action selection criteria from the current candidate action set, and set the next node contained in the candidate action as the updated current node. The relation edge identifier and the next node identifier of the candidate action are appended to the path history sequence in sequence. The updated current node is combined with the updated path history sequence to generate a new path reasoning state vector, and the new path reasoning state vector is used for subsequent multi-hop path reasoning.

[0027] In this embodiment, the step of selecting actions and expanding paths based on the updated path sequence to form a set of inference paths, and then determining the validity of the inference path set to obtain a set of valid inference paths, specifically includes: The new current node is determined based on the new path reasoning state vector. The relation edges directly connected to the new current node are retrieved from the relation edge set of the engineering knowledge graph. Each relation edge connected to the new current node and its corresponding next node are combined to form a new candidate action, and the new candidate action is added to the candidate action set. For each candidate action in the candidate action set, extract the relation type identifier, the next node identifier, and the current node representation used to generate the candidate action. Arrange the three types of information in the order of "relation type identifier - next node identifier - current node representation" to generate a candidate action feature sequence. The candidate action feature sequence and the corresponding path reasoning state vector are concatenated in the order of "path reasoning state vector first, candidate action feature sequence second" to form the path extended input data corresponding to the candidate action. Based on the path expansion input data corresponding to the candidate actions, select the candidate actions that meet the path expansion conditions from the candidate action set, and add the candidate actions to the path history sequence to form a reasoning path set consisting of multiple path history sequences. For each reasoning path in the reasoning path set, the validity of the path is determined based on the order of the node identifier sequence and the relation edge identifier sequence in the path. Reasoning paths that meet the validity determination conditions are included in the set of valid reasoning paths.

[0028] In this embodiment, the identification of construction risk links, material application links, process execution links, and settlement cost links of the rural power grid project based on the effective reasoning path set, to obtain the abnormal link set, specifically includes: For each inference path in the set of valid inference paths, the node type information, project stage information, and relation type information of the entity nodes contained in the path are read sequentially according to the path history sequence. The node type information, project stage information, and relation type information are then connected to form a corresponding link feature sequence. Based on the link feature sequence, the path length, the number of entity nodes, and the number of relation edges contained in the corresponding inference path are counted. The link feature sequence is then combined with the path length, the number of entity nodes, and the number of relation edges to generate a link feature vector corresponding to each inference path. The link feature vector corresponding to each inference path is input into the link type discrimination model. The link type discrimination model outputs the link type identifier corresponding to the link feature vector. The link type identifier includes the construction risk link identifier, the material declaration link identifier, the process execution link identifier, and the settlement cost link identifier. Based on the link type identifier, the set of valid inference paths is divided into the construction risk link set, the material declaration link set, the process execution link set, and the settlement cost link set. For each construction risk link in the construction risk link set, an anomaly score is calculated based on the corresponding link feature vector. Based on the comparison between the anomaly score and the construction risk anomaly threshold, construction risk links with anomaly scores greater than the construction risk anomaly threshold are selected as risk paths from the construction risk link set. For each material declaration link in the material declaration link set, an anomaly score is calculated based on the corresponding link feature vector. Based on the comparison between the anomaly score and the material anomaly threshold, material declaration links with anomaly scores greater than the material anomaly threshold are selected from the material declaration link set as material anomaly paths. For each process execution link in the process execution link set, an anomaly score is calculated based on the corresponding link feature vector. Based on the comparison between the anomaly score and the process anomaly threshold, process execution links with anomaly scores greater than the process anomaly threshold are selected from the process execution link set as process anomaly paths. For each settlement cost link in the settlement cost link set, an anomaly score is calculated based on the corresponding link feature vector. Based on the comparison between the anomaly score and the settlement inconsistency threshold, settlement cost links with anomaly scores greater than the settlement inconsistency threshold are selected from the settlement cost link set as settlement inconsistency paths. The risk path, material anomaly path, process anomaly path, and settlement inconsistency path are merged to form an anomaly link set.

[0029] In this embodiment, the step of locating the corresponding nodes in the entire project process based on the abnormal link set and generating early warning results and verification results specifically includes: For each abnormal link in the abnormal link set, the node type information, project stage information, and position number of each entity node in the path are read sequentially according to the corresponding path history sequence, and the link feature vector corresponding to the abnormal link is obtained. For each entity node in each abnormal link, the node type information, the project stage information, and the node position number are arranged in a fixed order to generate a node feature vector. The node feature vector and the corresponding link feature vector are then concatenated in the order of "node feature vector first, link feature vector last" to generate a node positioning input vector. The node localization input vector is input into the node localization model, which includes a feature encoding sublayer and a node classification sublayer. The feature encoding sublayer performs nonlinear mapping processing on the node localization input vector to generate a node encoding vector. The node classification sublayer outputs the node category identifier and the process position identifier based on the node encoding vector. Based on the node category identifiers output by the node classification sub-layer, nodes with the node category identifier of risk node are classified to form a risk node candidate set; nodes with the node category identifier of material anomaly node are classified to form a material anomaly node candidate set; nodes with the node category identifier of process breakpoint node are classified to form a process breakpoint node candidate set; and nodes with the node category identifier of settlement anomaly node are classified to form a settlement anomaly node candidate set. For each node in the candidate sets of risk nodes, material anomaly nodes, process breakpoint nodes, and settlement anomaly nodes, a node warning score is calculated based on the corresponding node coding vector. The node warning score is then compared with the corresponding node category warning threshold. Risk nodes with warning scores greater than the corresponding node category warning threshold are recorded as risk nodes. Material anomaly nodes with warning scores greater than the corresponding node category warning threshold are recorded as material anomaly nodes. Process breakpoint nodes with warning scores greater than the corresponding node category warning threshold are recorded as process breakpoint nodes. Settlement anomaly nodes with warning scores greater than the corresponding node category warning threshold are recorded as settlement anomaly nodes. The risk nodes, material anomaly nodes, process breakpoint nodes, and settlement anomaly nodes are grouped according to node category identifiers and process location identifiers to generate corresponding early warning results. The node encoding vector is then combined with the link feature vector of its respective anomaly link to generate corresponding verification results.

[0030] In this embodiment, the process of collecting newly added project data during the implementation of the agricultural power grid project, updating the node attributes and relationship edges in the project knowledge graph, and forming a dynamic iterative full-process control result specifically includes: During the implementation of the rural power grid project, new construction events, material allocation records, risk event records, and settlement update records are collected and combined to form a new project data set. Perform field parsing and time alignment processing on various types of data in the newly added engineering data set to obtain relevant information such as engineering project information, process information, equipment information, material information, construction event information, risk event information, and settlement cost information. Perform entity recognition and relationship recognition on the information to generate a new entity set and a new relationship set. Based on the newly added entity set, the entity node set in the engineering knowledge graph is updated. For existing entity nodes corresponding to the newly added entity, the node attribute information of the entity node is updated, including node type information, node identification information, engineering stage information, and node time attribute information. For newly added entities that do not exist in the entity node set, new entity nodes are generated and added to the entity node set. Based on the newly added set of relations, the set of relation edges in the engineering knowledge graph is updated. For existing relation edges corresponding to the newly added relations, the relation attribute information of the relation edge is updated, including relation type information, relation direction information, entity node identification information associated with the relation, and engineering stage identification information corresponding to the relation. For newly added relations that do not exist in the set of relation edges, new relation edges are generated and added to the set of relation edges. After updating the entity node set and relation edge set, an updated engineering knowledge graph is formed. The updated entity node set and relation edge set are used as the engineering knowledge graph structure on which subsequent path reasoning processing is based. In the updated engineering knowledge graph, new starting nodes and new target nodes are determined based on the current management objects of the agricultural distribution network project. The node attribute information and node association information of the new starting node are obtained, as well as the node attribute information and project stage information of the new target node. A new path reasoning state vector is generated by combining the existing path history sequence. Based on the new path reasoning state vector, a new set of candidate actions is generated in the updated engineering knowledge graph. The candidate actions in the new set of candidate actions are selected and the path is expanded to form a new set of reasoning paths. The validity of the new set of reasoning paths is determined to obtain a new set of valid reasoning paths. A new link feature vector is generated based on the new set of valid inference paths. The new link feature vector is then input into the link type discrimination model and anomaly detection processing flow to obtain a new set of abnormal links. Based on the new set of abnormal links, new abnormal nodes are located, and new warning results and new verification results are generated. The new early warning results and new verification results obtained based on the updated engineering knowledge graph are combined with the historical early warning results and historical verification results in chronological order to form a dynamically iteratively updated full-process control result.

[0031] A knowledge graph-based end-to-end management and control system for agricultural power grid projects, comprising: The data processing module is used to acquire multi-source heterogeneous data from the agricultural power grid project, perform preprocessing, and generate an agricultural power grid data set. The graph construction module is used to construct an engineering knowledge graph based on the entity set and relation set in the agricultural distribution network data set; The path input generation module is used to construct a path reasoning state vector based on the selected starting node and target node in the engineering knowledge graph, and obtain the initial input for path reasoning. The strategy action module is used to input the initial input of path reasoning into the strategy network, perform action probability calculation on the candidate action set, and obtain the next hop action and the updated path sequence. The multi-hop path generation module is used to perform multi-hop path expansion and path validity determination on the updated path sequence to obtain a set of valid inference paths. The link identification module is used to construct link feature vectors based on the set of valid inference paths, perform link type discrimination and anomaly detection on the link feature vectors, and obtain a set of abnormal links. The node location module is used to locate the corresponding nodes in the entire process of the project based on the abnormal link set, and generate early warning results and verification results. The dynamic graph update module is used to collect new project data during the implementation of the agricultural power grid project, update the node attributes and relationship edges in the project knowledge graph, and form the full-process control results.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to the full-process management of a rural power grid construction project. This project comprises five phases: planning, design, construction, acceptance, and settlement. It has a long project cycle, involves numerous participating units, and has complex data sources. New material allocation records, construction event records, risk event records, and settlement updates are constantly generated at the construction site. Traditional systems lack the ability to structure and represent this dynamic data, and cannot perform unified analysis of cross-phase connections. This leads to inaccurate construction risk identification, difficulty in tracking material consumption, inability to identify process delays in advance, and settlement discrepancies often only becoming apparent after project completion, affecting the timeliness of decision-making.

[0033] In implementing the method of this invention, the project site first inputs planning documents, design drawings, construction logs, material allocation orders, acceptance records, and settlement ledger data into the system. The data processing module then performs field parsing and time alignment on this data. The processed data is converted into project entity nodes and relationship edges by the engineering knowledge graph construction module, so that the relationships between project entities, process entities, equipment entities, material entities, construction event entities, risk event entities, and settlement cost entities are fully expressed in the graph. When new hoisting events, material replenishment records, or safety hazard records occur at the construction site, the system will immediately collect this new data and update the engineering knowledge graph.

[0034] After applying the path reasoning module, the system automatically identifies risky links that may affect construction quality from the design phase to the construction phase. For example, during a foundation pouring operation, the system detected abnormal jumps between three consecutive construction event records and key process nodes in the design phase, marking these as high-risk links. Simultaneously, in the material application link at this stage, the system discovered through link feature vectors that the consumption of a certain type of steel increased rapidly in a short period, significantly deviating from previous consumption patterns, and forming abnormal correlations with multiple process nodes, indicating a risk of abnormal material usage.

[0035] During the node localization phase, the system identifies key risk nodes, material anomaly nodes, and process breakpoint nodes based on the characteristics of nodes in the abnormal chain. For example, in a project, the system identified multiple process breakpoint nodes related to concrete pouring and determined that they were strongly correlated with equipment failure events. Further verification results confirmed that the anomaly continuously affected six process nodes, prompting the construction team to immediately investigate the construction equipment.

[0036] Throughout the entire project cycle, this invention performed over fifty incremental updates to the dynamic knowledge graph. System records show that the dynamic link analysis mechanism implemented by this invention can identify potential process delay risks on average three to five days in advance, with an accuracy rate of over 90% in material anomaly identification and an improvement of settlement deviation correction efficiency by approximately 30%. After the engineering contractor adopted this system for full-process data analysis, the overall construction cycle was shortened by approximately 15%, the rework rate decreased by approximately 7%, and project quality and management efficiency were effectively improved.

[0037] Table 1. Statistics on Anomaly Analysis of Engineering Links As can be seen from the above "Statistical Table of Engineering Link Anomalies," this invention demonstrates significant performance advantages in engineering link identification, anomaly location, and dynamic monitoring. A total of 38 abnormal links were identified in the construction risk link, and 64 key nodes were located, indicating that this invention can accurately capture risk propagation paths in complex construction processes. 27 abnormal links were identified in the material declaration link, and 41 key nodes were located, providing timely alerts when material consumption deviates from normal patterns. 55 abnormal links were identified in the process execution link, and 72 key nodes were located, effectively detecting process jumps, process interruptions, and abnormal process sequences. 19 abnormal links were identified in the settlement cost link, and 26 key nodes were located, demonstrating that this invention can identify cost mapping deviations in structured links during the settlement stage.

[0038] In terms of early warning efficiency, the early warning capability of this invention is outstanding. The average early warning time for construction risk links is 3.2 days, for process execution links it is 3.8 days, and for material declaration links it is 2.6 days, all significantly better than traditional monitoring methods that rely on manual experience. The anomaly location accuracy for multiple types of links is at a high level, with 91.4% for construction risk links, 93.1% for material declaration links, 89.7% for process execution links, and 95.6% for settlement cost links. These results indicate that the comprehensive analysis framework built by the system based on path reasoning, link feature vectors, and node location models can provide highly reliable anomaly diagnosis results during the engineering process.

[0039] In terms of overall project management effectiveness, this invention completed 52 incremental map updates in the application scenario, achieving continuous monitoring of dynamic changes in the project. In this project, the construction cycle was shortened by approximately 15% and the rework rate decreased by approximately 7% after the system was applied. In summary, this invention can achieve real-time analysis and efficient management of the entire project process through structured modeling, link correlation mining, and dynamic reasoning of project process data, providing reliable data support and technical means for improving project quality and controlling costs.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based method for full-process management and control of agricultural power grid projects, characterized in that, include: Acquire multi-source heterogeneous data from the planning, design, construction, acceptance and settlement stages of the rural power grid project, perform preprocessing on the multi-source heterogeneous data, and generate a rural power grid data set; Based on the entity set and relation set in the agricultural distribution network dataset, an engineering knowledge graph containing the entity node set and relation edge set is constructed. In the engineering knowledge graph, the starting node and target node to be analyzed are determined, and the initial input for path reasoning is formed based on the current node representation, the target node representation, and the path history sequence. The current set of possible actions is generated based on the initial input of path reasoning. The action probability of the set of possible actions is calculated based on the policy network to determine the next hop action and update the path sequence. Based on the updated path sequence, action selection and path expansion are performed to form a set of inference paths. The validity of the inference path set is then determined to obtain a set of valid inference paths. Based on the set of valid reasoning paths, the construction risk links, material application links, process execution links and settlement cost links of the rural power grid project are identified, and the set of abnormal links is obtained. Based on the abnormal link set, the corresponding nodes in the entire process of the project are located, and early warning results and verification results are generated; Collect newly added project data during the implementation of the agricultural power grid project, update the node attributes and relationship edges in the project knowledge graph, and form a full-process control result.

2. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The preprocessing includes field parsing, time alignment, entity extraction, and relationship identification.

3. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The construction of an engineering knowledge graph containing a set of entity nodes and a set of relationship edges based on the entity set and relationship set in the agricultural distribution network dataset specifically includes: Extract the entity set from the agricultural distribution network data set, and generate project nodes, process nodes, equipment nodes, material nodes, construction event nodes, risk event nodes, and settlement cost nodes respectively to form an initial set of nodes; Each node in the initial set of nodes is assigned node attribute values, including node identifier, node type, project stage information, and node association time information; Extract the relation set from the agricultural network dataset, and connect the nodes in the initial node set to form the initial set of relation edges; Each relation edge in the initial set of relation edges is assigned a relation attribute value, which includes relation type, relation direction, node identifier corresponding to the relation, and project stage identifier corresponding to the relation. The initial set of nodes and the initial set of relation edges are combined to form a knowledge graph for agricultural power grid projects that includes the set of entity nodes and the set of relation edges.

4. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The process of determining the starting and target nodes to be analyzed in the engineering knowledge graph, and forming the initial input for path reasoning based on the current node representation, the target node representation, and the path history sequence, specifically includes: In the entity node set of the engineering knowledge graph, the target analysis type is determined according to the control object of the rural power grid project, and the target node and the starting node are selected from the entity node set according to the target analysis type; The starting node is determined as the current node. The node attribute information of the current node and the relationship edge information directly connected to the current node are obtained. The node representation of the current node is generated. The node attribute information of the target node is obtained. The node representation of the target node is generated. Construct a path history sequence based on the node access order and relation edge connection order recorded in the engineering knowledge graph; The node representations of the current node, the target node, and the path history sequence are combined and arranged according to a fixed structure to generate a path reasoning state vector, which is then used as the initial input for path reasoning.

5. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The process of generating a current set of selectable actions based on the initial input of path reasoning, calculating the action probability of the selectable action set based on the policy network, determining the next hop action, and updating the path sequence specifically includes: The current node is determined based on the path reasoning state vector. The relationship edge directly connected to the current node is retrieved from the relationship edge set of the engineering knowledge graph to form the current candidate action set. Extract the relation type identifier, next node identifier, and current node representation from the candidate actions to generate an action feature sequence; The path reasoning state vector is concatenated with the action feature sequence to form the input data of the policy network. The input data is then fed into the policy network to obtain the action probability. Based on the action probability, select candidate actions whose action probabilities satisfy the action selection criteria from the current candidate action set, and set the next node contained in the candidate action as the updated current node. The relation edge identifiers and next node identifiers of candidate actions are appended to the path history sequence in sequence. The updated current node is combined with the updated path history sequence to generate a new path reasoning state vector.

6. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The process of selecting actions and expanding paths based on the updated path sequence to form a set of inference paths, and then determining the validity of the inference path set to obtain a set of valid inference paths, specifically includes: The new current node is determined based on the new path reasoning state vector, new candidate actions are obtained, and the new candidate actions are added to the candidate action set. Extract the relation type identifier, the next node identifier, and the current node representation from the candidate actions to generate a candidate action feature sequence; The candidate action feature sequence is concatenated with the corresponding path reasoning state vector to form the path extended input data corresponding to the candidate action; Based on the path expansion input data corresponding to the candidate actions, select candidate actions that meet the path expansion conditions from the candidate action set, and add the candidate actions to the path history sequence to form a reasoning path set; The validity of each reasoning path in the reasoning path set is determined to obtain the set of valid reasoning paths.

7. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The method of identifying construction risk links, material application links, process execution links, and settlement cost links in rural power grid projects based on the set of valid reasoning paths yields a set of abnormal links, specifically including: For each inference path in the set of valid inference paths, a corresponding link feature sequence is formed; Based on the link feature sequence, the path length, the number of entity nodes, and the number of relation edges of the corresponding inference path are counted to generate a link feature vector. Input the link feature vector corresponding to each reasoning path into the link type discrimination model to divide the construction risk link set, material declaration link set, process execution link set and settlement cost link set; The construction risk link anomaly score is calculated based on the link feature vector, and the risk path is determined based on the construction risk link anomaly score and the construction risk anomaly threshold. Calculate the abnormal score of the material declaration link based on the link feature vector to determine the abnormal path of the material; Calculate the process execution link anomaly score based on the link feature vector to determine the abnormal process path; Calculate the settlement fee link anomaly score based on the link feature vector to identify settlement inconsistency paths; The risk path, material anomaly path, process anomaly path, and settlement inconsistency path are merged to form an anomaly link set.

8. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The process of locating corresponding nodes in the entire project process based on the abnormal link set and generating early warning and verification results specifically includes: For each abnormal link in the abnormal link set, obtain the link feature vector based on the path history sequence; For each entity node in each abnormal link, the node type information, the project stage information, and the node position number are arranged in a fixed order to generate a node feature vector. The node feature vector is then concatenated with the corresponding link features to generate a node positioning input vector. Input the node localization input vector into the node localization model, and output the node category identifier and process location identifier; Based on the node category identifiers output by the node classification sublayer, the nodes are categorized to form candidate sets for each node; Calculate the node warning score for each node in each candidate set and compare it with the node category warning threshold to obtain risk nodes, material abnormal nodes, process breakpoint nodes and settlement abnormal nodes. Risk nodes, material anomaly nodes, process breakpoint nodes, and settlement anomaly nodes are grouped according to node category identifiers and process location identifiers to generate corresponding early warning results. The node encoding vector is then combined with the link feature vector of its corresponding anomaly link to generate corresponding verification results.

9. The method for full-process management and control of agricultural power grid projects based on knowledge graphs according to claim 1, characterized in that, The process of collecting newly added project data during the implementation of the agricultural power grid project, updating the node attributes and relationship edges in the project knowledge graph, and forming the full-process control results specifically include: During the implementation of the rural power grid project, new construction events, material allocation records, risk event records, and settlement update records are collected to form a new project data set. Perform preprocessing on various types of data in the newly added project dataset to generate a new entity set and a new relation set; Based on the newly added entity set and the newly added relation set, the entity node set and relation edge set in the engineering knowledge graph are updated to form the updated engineering knowledge graph; In the updated engineering knowledge graph, new starting nodes and new target nodes are determined based on the current control objects of the agricultural power grid project, and new path reasoning state vectors are generated. Based on the new path reasoning state vector, a new set of candidate actions is generated in the updated engineering knowledge graph, resulting in a new set of effective reasoning paths. New link feature vectors are generated based on the new set of effective inference paths, resulting in new early warning results and new verification results. The new early warning results and new verification results obtained from the updated engineering knowledge graph are combined with the historical early warning results and historical verification results in chronological order to form the full-process control results.

10. A knowledge graph-based full-process management and control system for agricultural power grid projects, characterized in that, The method for full-process management and control of agricultural power grid projects based on knowledge graphs, as described in any one of claims 1-9, includes: The data processing module is used to acquire multi-source heterogeneous data from the agricultural power grid project, perform preprocessing, and generate an agricultural power grid data set. The graph construction module is used to construct an engineering knowledge graph based on the entity set and relation set in the agricultural distribution network data set; The path input generation module is used to construct a path reasoning state vector based on the selected starting node and target node in the engineering knowledge graph, and obtain the initial input for path reasoning. The strategy action module is used to input the initial input of path reasoning into the strategy network, perform action probability calculation on the candidate action set, and obtain the next hop action and the updated path sequence. The multi-hop path generation module is used to perform multi-hop path expansion and path validity determination on the updated path sequence to obtain a set of valid inference paths. The link identification module is used to construct link feature vectors based on the set of valid inference paths, perform link type discrimination and anomaly detection on the link feature vectors, and obtain a set of abnormal links. The node location module is used to locate the corresponding nodes in the entire process of the project based on the abnormal link set, and generate early warning results and verification results. The dynamic graph update module is used to collect new project data during the implementation of the agricultural power grid project, update the node attributes and relationship edges in the project knowledge graph, and form the full-process control results.

Images