A power grid policy intelligent matching method based on multi-source data and a knowledge graph

Abstract

This application provides a method for intelligent matching of power grid policies based on multi-source data and knowledge graphs, relating to the field of data processing technology. The method includes: acquiring multi-source power grid data; performing semantic alignment processing on the multi-source power grid data to obtain unified semantic data; constructing a dynamic knowledge graph based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; performing policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results; when user query data is received, inputting the user query data into the dynamic knowledge graph, generating an intelligent response based on the policy matching results, and performing closed-loop optimization of the dynamic knowledge graph and the matching model based on user feedback.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, specifically relating to a method for intelligent matching of power grid policies based on multi-source data and knowledge graphs. Background Technology

[0002] Against the backdrop of intelligent transformation in power grid operation and management, efficiently and accurately matching and utilizing various power grid-related policies issued by the state and local governments (such as renewable energy subsidies, energy efficiency incentives, and carbon trading policies) has become a core aspect for power grid companies to improve operational efficiency, ensure policy compliance, and maximize economic benefits. Traditional policy matching and management systems, typically built on keyword matching or static rule bases, have significant shortcomings in their processing flow: First, they struggle to effectively integrate multi-source heterogeneous data generated during power grid operation, including real-time changes in power load, renewable energy output, equipment status monitoring data, user electricity consumption behavior, and external meteorological information. This data is diverse in format, frequently updated, and deeply coupled with policy provisions, making it impossible for traditional systems to achieve real-time semantic alignment and integrated analysis. Second, existing knowledge base models are mostly statically constructed, lacking dynamic evolution capabilities and unable to adapt to frequent policy updates and instantaneous changes in power grid operating status, resulting in lagging policy matching results and inaccurate calculations. Finally, traditional systems suffer from rigid human-computer interaction, unable to understand users' natural language queries or perform deep reasoning based on data reports, and lack a closed-loop mechanism for self-optimization using user feedback, severely restricting the level of intelligence in policy services and the user experience. Summary of the Invention

[0003] This application provides a method for intelligent matching of power grid policies based on multi-source data and knowledge graphs to solve one of the aforementioned technical problems.

[0004] The technical solution adopted in this application is as follows: This application provides an embodiment of a method for intelligent matching of power grid policies based on multi-source data and knowledge graphs, including: Acquire multi-source power grid data, and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A dynamic knowledge graph is constructed based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; Based on a pre-trained graph neural network model, the dynamic knowledge graph is dynamically matched with policies and semantic reasoning is performed to obtain policy matching results. When user query data is received, the user query data is input into the dynamic knowledge graph, an intelligent response is generated based on the policy matching result, and the dynamic knowledge graph and matching model are optimized in a closed loop according to user feedback.

[0005] According to one embodiment of this application, the step of acquiring multi-source power grid data and performing semantic alignment processing on the multi-source power grid data to obtain unified semantic data includes: Real-time collection of power grid operation data, enterprise operation data, and policy data through API interfaces and IoT protocols; Natural language processing was used to extract policy entities from text-based policy data, time series analysis was used to extract key features from time-series power grid operation data, and computer vision technology was used to extract equipment relationships from image-based data. A multimodal feature fusion method is used to map the extracted features to the same vector space, thereby achieving cross-modal semantic alignment and obtaining unified semantic data.

[0006] According to one embodiment of this application, the step of constructing a dynamic knowledge graph based on the unified semantic data includes: An initial knowledge graph is constructed using a graph database, where nodes include policy entities, power grid equipment, operating parameters, and enterprise attributes, and edges represent semantic relationships between nodes. Design a streaming processing mechanism that automatically triggers incremental updates to the knowledge graph when new data flows in; The confidence fusion algorithm is used to detect and handle data conflicts during the update process to ensure the consistency of the knowledge graph. A timeliness-based weighting model is introduced to dynamically adjust node weights based on data source and timestamp.

[0007] According to one embodiment of this application, the step of performing policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results includes: The policy nodes and power grid data nodes in the dynamic knowledge graph are input into the graph neural network model, and the complex relationships between the nodes are learned through multi-layer graph convolution. Based on a multi-hop reasoning mechanism, the reasoning path related to user queries is traversed in the knowledge graph. By integrating reinforcement learning algorithms, the parameters of the matching model are dynamically adjusted based on user feedback to optimize matching accuracy; The policy calculation model is linked to the power grid operation status, and the weights of the calculation formula are dynamically adjusted through linear programming and time decay factors.

[0008] According to one embodiment of this application, the steps of inputting user query data into the dynamic knowledge graph when user query data is received, generating an intelligent response based on the policy matching result, and performing closed-loop optimization of the dynamic knowledge graph and the matching model based on user feedback include: It supports multimodal input including natural language, speech, and images, and uses a deep learning natural language understanding model to interpret user intent. Structured responses are generated based on the reasoning results of knowledge graphs, and the response content is dynamically optimized through user interaction feedback. Construct a knowledge subgraph of user profiles, integrating user historical behavior, job roles, and preference data to achieve personalized recommendations; User actions on the recommendation policy are fed back to the knowledge graph and matching model, forming a closed-loop optimization mechanism.

[0009] According to one embodiment of this application, the design of the streaming processing mechanism, which automatically triggers incremental updates of the knowledge graph when new data flows in, includes the following steps: Real-time monitoring of data source changes; automatic extraction of new data when new policies are released or power grid events occur. Each new node is checked against an existing node in the knowledge graph. If they are different, the node is added as a new node. Check whether the newly added edge is the same as an existing edge in the knowledge graph. If they are different, add it as a new edge. If they are the same, update the edge weight. Incremental maintenance is performed on the updated knowledge graph, including transaction mechanisms, dependency checks, and redundancy cleanup.

[0010] According to one embodiment of this application, the step of incrementally maintaining the updated knowledge graph includes: Each update operation is performed in a transactional manner to ensure the atomicity of the update; Check the dependencies between nodes to avoid conflicts between new rules and existing relationships; Remove invalid nodes and redundant edges to keep the knowledge graph structure compact.

[0011] A second aspect of this application provides a smart matching and optimization system for power grid policies based on multi-source data fusion and dynamic knowledge graphs, including: The data acquisition module is used to acquire multi-source power grid data and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A knowledge graph construction module is used to construct a dynamic knowledge graph based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; The matching and reasoning module is used to perform policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results. The interaction optimization module is used to input user query data into the dynamic knowledge graph when it receives user query data, generate an intelligent response based on the policy matching result, and perform closed-loop optimization of the dynamic knowledge graph and the matching model based on user feedback.

[0012] A third aspect of this application provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps described in the method.

[0013] A fourth aspect of this application provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method as described.

[0014] Due to the adoption of the above technical solution, the beneficial effects achieved by this application are as follows: This application acquires multi-source power grid data and performs semantic alignment and fusion processing, unifying the originally scattered and heterogeneous data into structured information with semantic relationships. The initial knowledge graph built upon this foundation is no longer a static database, but rather achieves automated and continuous evolution of the knowledge model through subsequent steps that dynamically update node relationships and policy weights based on user feedback data and real-time power grid data. This enables the system to respond in real-time to policy changes and changes in power grid operating conditions, fundamentally solving the problems of knowledge lag and rigidity in traditional systems.

[0015] By employing graph neural networks and multi-hop inference mechanisms for policy matching and semantic reasoning, the system can deeply uncover the complex and nonlinear relationships between policy provisions and power grid entities and operating parameters, achieving a deeper level of semantic understanding. In particular, by dynamically binding the policy calculation model with the real-time operating status of the power grid, the calculation of policy benefits is no longer an isolated, fixed formula application, but can adaptively adjust based on current grid load, renewable energy output, and other factors, thereby outputting more practically guiding optimization strategies and significantly improving matching accuracy and the practicality of the calculation results.

[0016] By generating intelligent responses in response to user interaction requests and collecting user feedback data, and then using this feedback data to dynamically optimize policy matching model parameters through reinforcement learning, a complete iterative process of "data collection - knowledge construction - matching reasoning - interactive feedback - closed-loop optimization" has been constructed. This design makes the system not only a query tool, but also an intelligent agent capable of learning from each interaction and continuously optimizing its core matching algorithm and knowledge structure. This greatly improves the system's intelligence level and the accuracy of long-term services, while providing users with increasingly personalized decision support, significantly improving user experience and decision-making efficiency. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for intelligent matching of power grid policies based on multi-source data and knowledge graphs, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0018] Figure label: 810, Processor; 820, Communication interface; 830, Memory; 840, Communication bus. Detailed Implementation

[0019] To more clearly illustrate the overall concept of this application, a detailed explanation is provided below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below. It should be noted that, unless otherwise specified, the embodiments of this application and the features thereof can be combined with each other.

[0021] In this application, unless otherwise expressly specified and limited, the "above" or "below" of the second feature can mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. In the description of this specification, references to terms such as "an embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0022] Example 1 like Figure 1 As shown, a smart matching method for power grid policies based on multi-source data and knowledge graphs includes: Acquire multi-source power grid data, and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data.

[0023] As mentioned above, "acquiring multi-source power grid data" refers to collecting heterogeneous data from multiple sources related to power grid operation in real time or periodically. This data includes, but is not limited to, power grid operation data (such as load curves and renewable energy output sequences provided by SCADA systems), equipment status data (such as transformer temperatures and line fault records), enterprise operation data (such as financial indicators and R&D investment records), policy text data (such as government-issued subsidy policies and energy efficiency standard documents), and external environmental data (such as meteorological information and user electricity consumption logs). These data typically have different formats (such as structured database records, unstructured text, time series, and images), update frequencies, and semantic expressions, making direct integration difficult. "Semantic alignment processing of the multi-source power grid data" refers to mapping the aforementioned heterogeneous data into a unified semantic framework through multimodal feature extraction and fusion techniques. Specifically, natural language processing techniques are used to extract entities (such as "policy name" and "applicable conditions") and relationships from policy texts; time series analysis methods are used to identify key features (such as "peak load" and "renewable energy share") from power grid operation data; and computer vision techniques are used to parse equipment relationships from image data (such as power grid topology diagrams). Subsequently, these features are uniformly represented as data elements with consistent semantic labels through semantic mapping models (such as ontology-based alignment or vector space embedding), eliminating terminological ambiguity and structural differences, thus forming "unified semantic data". This data not only preserves the core semantics of the original information, but also establishes cross-data source relationships, providing standardized input for the subsequent construction of dynamic knowledge graphs.

[0024] For example, the system acquires data from multiple sources: real-time photovoltaic power generation output data (time-series format) is collected from the SCADA system via API, policy text (PDF document) is downloaded from the government database, solar radiation intensity data (JSON format) is obtained from the meteorological service interface, and photovoltaic equipment files (structured tables) are extracted from the enterprise management system. In semantic alignment processing, the system uses natural language processing technology to parse the policy text, identifying entities such as "subsidy amount" and "power generation threshold"; simultaneously, it performs time-series analysis on the power generation output data to extract the "average daily power generation" feature; and it associates solar radiation data with power generation data, identifying "effective sunshine duration" as an influencing factor. Through semantic mapping, these elements are uniformly represented as "policy entity: photovoltaic subsidy," "equipment entity: photovoltaic array," "parameter entity: power generation," etc., and relationships such as "policy applicability conditions depend on power generation" are established. Finally, the unified semantic data is stored in a structured form (such as RDF triples) to ensure that all data is semantically consistent and interoperable, laying the foundation for subsequent knowledge graph construction and policy matching.

[0025] It should be noted that, in specific implementation scenarios, based on the above solutions, a streaming processing mechanism (such as a real-time data pipeline based on a message queue) can be introduced during the data acquisition phase to support the continuous collection of high-frequency updated data (such as smart meter flow data), ensuring that the system can respond to instantaneous changes in the power grid status. In semantic alignment processing, deep learning models (such as cross-modal Transformer networks) can be used for more refined semantic mapping, improving the accuracy of understanding complex policy provisions and power grid operation modes through pre-trained language models and temporal encoders. Furthermore, dynamic ontology management functions can be integrated to automatically adjust semantic alignment rules based on new policy releases or power grid topology changes. For example, when new terms appear in a policy, the system updates the semantic dictionary through online learning, avoiding manual intervention.

[0026] A dynamic knowledge graph is constructed based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes.

[0027] As described above, this step is the core process of transforming the aforementioned unified semantic data into a structured knowledge model, aiming to construct a dynamically evolving power grid policy knowledge graph. Specifically, this knowledge graph uses a graph structure to organize data, where nodes represent different categories of entities: policy entity nodes represent various policy elements, including attributes such as policy name, issuing agency, effective date, and applicable conditions; power grid equipment nodes describe physical or logical equipment units, such as transformers, photovoltaic arrays, and smart meters, with attributes such as equipment model, installation location, and rated parameters; and operating parameter nodes record dynamically changing power grid state indicators, such as time-series data like load curves, power generation, and voltage levels. These nodes are semantically linked through directed or undirected edges, forming a multi-dimensional relationship network of "policy-equipment-parameter," for example, "policy A applies to equipment B" and "equipment C is associated with operating parameter D." The dynamism of knowledge graphs is reflected in their support for streaming update mechanisms. When new unified semantic data is generated, the system can automatically identify new entities or relationships and integrate them into the existing graph in real time. At the same time, it maintains data consistency through conflict detection algorithms to ensure that the knowledge graph is always synchronized with the actual power grid status and policy environment.

[0028] For example, taking the construction of a knowledge graph related to "distributed photovoltaic subsidy policy" as an example, the system first creates policy entity nodes, including the "2025 photovoltaic electricity subsidy policy" and its attributes such as the subsidy standard of 0.5 yuan per kilowatt-hour and the applicable conditions of installed capacity exceeding 10 kilowatts; then, it establishes grid equipment nodes, such as "photovoltaic array A" and "smart meter B" in a certain community, recording their technical parameters and geographical location; finally, it generates operating parameter nodes, such as "daily power generation curve of photovoltaic array A" and "real-time load data of regional power grid". Subsequently, the system establishes semantic relationships between these nodes: connecting policy nodes and equipment nodes through "applicable objects" edges, associating equipment nodes and parameter nodes through "data generation" edges, and connecting different parameter nodes through "spatiotemporal association" edges. When the policy is updated or the equipment power generation data changes, the knowledge graph will dynamically adjust the corresponding node attributes and edge weights. For example, when the subsidy policy is adjusted, the policy node attributes are automatically updated and the association strength with related equipment is recalculated.

[0029] It should be noted that, in specific implementation scenarios, the above solutions can be further enhanced by: Firstly, introducing a timeliness-based weighting model to dynamically assign weights to nodes and edges based on factors such as policy release time, equipment operational years, and data freshness, allowing the knowledge graph to prioritize information with high timeliness during inference. Secondly, expanding node types to support more complex application scenarios, such as adding "user behavior nodes" to record electricity usage habits or "environmental factor nodes" to integrate meteorological data, thereby establishing a more comprehensive multi-dimensional network of "policy-equipment-parameter-environment-user". Furthermore, the knowledge graph can be deeply integrated with subsequent intelligent interaction modules, such as automatically optimizing the structure of relevant subgraphs based on user query hotspots or dynamically adjusting the strength of associations between nodes based on user feedback data.

[0030] Based on a pre-trained graph neural network model, the dynamic knowledge graph is used for policy dynamic matching and semantic reasoning to obtain policy matching results.

[0031] As described above, this step is the core process of using graph neural networks to perform deep analysis and reasoning on the constructed knowledge graph. Specifically, the pre-trained graph neural network model, through its unique graph convolution or graph attention mechanism, can simultaneously consider the feature information and topological structure information of nodes in the knowledge graph, achieving deep mining of complex policy-equipment-parameter relationships. In the dynamic policy matching process, the model first maps various nodes (policy entities, power grid equipment, operating parameters) in the knowledge graph into low-dimensional vector representations, preserving their semantic features and relationships; then, through a multi-layer information transmission mechanism, it aggregates the feature information of adjacent nodes, so that the vector representation of each node contains the semantics of its local network structure. Based on this, the model can perform multi-hop semantic reasoning, that is, logical deduction along the relational paths in the knowledge graph, identifying the potential matching relationship between policy clauses and specific power grid scenarios, and comprehensively considering the impact of the real-time state of the power grid on the applicability of the policy, finally outputting quantitative policy matching results, including policy applicability assessment, expected benefit analysis, and implementation plan suggestions.

[0032] For example, consider an industrial park applying for the "energy storage peak-valley pricing policy." The knowledge graph contains nodes for "energy storage subsidy policy," "energy storage device C," and operational parameters such as "day-ahead electricity price curve" and "load characteristics." The graph neural network model first learns the vector representations of these nodes, then discovers the path through multi-hop reasoning: "energy storage device C" → "participate in peak shaving" → "reduce peak-valley difference" → "meet subsidy conditions" → "energy storage subsidy policy." Simultaneously, the model comprehensively analyzes real-time operational parameters, such as whether the current grid peak-valley price difference reaches the policy threshold and whether the actual charging and discharging behavior of the energy storage device meets the requirements. Based on this comprehensive analysis, the model ultimately outputs a matching result: the industrial park is eligible for the energy storage subsidy policy, the estimated monthly subsidy amount range is provided, and suggestions for optimizing charging and discharging strategies to maximize benefits are given. When the grid operating status changes, such as when the peak-valley price difference is adjusted, the model dynamically updates the matching result to ensure the real-time effectiveness of the recommendations.

[0033] It should be noted that, in specific implementation scenarios, based on the above solutions, on the one hand, a reinforcement learning mechanism can be introduced. Feedback information such as policy application results and user adoption status can be used as reward signals to dynamically adjust the inference strategy and matching weights of the graph neural network, enabling the model to continuously learn and optimize from practical application effects. On the other hand, a multi-task joint learning framework can be extended, allowing the graph neural network to simultaneously complete multiple related tasks such as policy matching, anomaly detection, and benefit prediction. Knowledge sharing between tasks can improve the overall inference accuracy and efficiency. Furthermore, this matching inference module can form a closed-loop linkage with the dynamic update mechanism of the knowledge graph. When the model identifies a long-term mismatch between certain policies and the power grid's operating status, it can automatically trigger weight adjustments or label updates for the corresponding node relationships in the knowledge graph.

[0034] When user query data is received, the user query data is input into the dynamic knowledge graph, an intelligent response is generated based on the policy matching result, and the dynamic knowledge graph and matching model are optimized in a closed loop according to user feedback.

[0035] As mentioned above, this step is a key link in realizing intelligent interaction and self-optimization of the system, and it constructs a complete "query-response-feedback-optimization" closed loop. Specifically, when the system receives a query request from a user in the form of natural language, structured query, or mixed modality, it first parses the query intent through the natural language understanding module and establishes semantic associations with nodes in the dynamic knowledge graph. Then, based on the policy matching results obtained in the early stage and combined with the multi-dimensional associations in the graph, the system generates a structured intelligent response that includes policy applicability analysis, benefit evaluation results, and specific operational suggestions. After outputting the response to the user, the system continuously collects data on user interaction behavior, including explicit feedback such as the adoption of recommended policies, the success or failure of policy applications, and user active evaluations, as well as implicit feedback such as the user's dwell time and query depth. This feedback data is systematically used to optimize two core components: on the one hand, the association weights between policy nodes and power grid entities in the knowledge graph are verified and adjusted based on the actual policy adoption results of the user, correcting inaccurate semantic relationships; on the other hand, the feedback signals are used as reward indicators for reinforcement learning to dynamically adjust the parameters of the graph neural network matching model, so that the model gradually adapts to changes in user preferences and the power grid environment, thereby achieving continuous self-improvement of the system.

[0036] For example, suppose an operator from a power grid company enters the query "How to maximize the benefits of the current distributed photovoltaic (PV) policy." The system first analyzes the key intents in the query: "distributed PV policy" and "maximize utilization." Then, it locates relevant policies and the company's existing PV equipment nodes in the knowledge graph. Using a graph neural network model, it calculates the matching degree of different policy combinations and comprehensively considers the company's electricity load characteristics and the regional power grid's absorption capacity. This generates an intelligent response containing multiple suggestions, including "apply for the 2025 PV electricity subsidy policy," "configure a smart energy storage system to improve self-consumption rate," and "participate in green electricity trading." When the user adopts some of the suggestions and successfully applies for subsidies, the system collects this positive feedback and accordingly strengthens the association weight between "PV equipment" and "electricity subsidy policy" in the knowledge graph. Simultaneously, for the suggestion "participate in green electricity trading," which the user did not adopt, the system analyzes the user's subsequent behavior and finds that the trading threshold is too high. Therefore, it lowers the recommendation priority of similar high-threshold policies in the graph neural network matching model and adds an attribute label of "high applicable threshold" to the policy node in the knowledge graph, thus providing more accurate recommendations in subsequent queries.

[0037] It should be noted that, in specific implementation scenarios, based on the above solutions, on the one hand, a multi-dimensional user profile model can be established. By continuously collecting data such as user role attributes, historical query preferences, and decision-making characteristics, personalized user subgraphs can be constructed in the knowledge graph, enabling the system to provide more targeted and differentiated services, such as providing detailed technical parameter analysis for technical decision-makers and comprehensive benefit assessment for management decision-makers. On the other hand, a progressive learning mechanism can be introduced. When the system detects that user feedback on certain policy nodes in the knowledge graph continuously deviates from expectations, a special knowledge update process is automatically triggered. By increasing the collection frequency and verification intensity of policy-related data, the convergence and optimization of the model in that domain can be accelerated. In addition, the system can also expand the multimodal feedback interface to support users in semantic annotation and relationship correction of recommendation results. For example, users can be allowed to directly point out that "this policy is not applicable because the device capacity is insufficient." The system transforms this explicit feedback into a structured update of the knowledge graph, significantly improving optimization efficiency.

[0038] According to one embodiment of this application, the step of acquiring multi-source power grid data and performing semantic alignment processing on the multi-source power grid data to obtain unified semantic data includes: Real-time collection of power grid operation data, enterprise operation data, and policy data through API interfaces and IoT protocols; Natural language processing was used to extract policy entities from text-based policy data, time series analysis was used to extract key features from time-series power grid operation data, and computer vision technology was used to extract equipment relationships from image-based data. A multimodal feature fusion method is used to map the extracted features to the same vector space, thereby achieving cross-modal semantic alignment and obtaining unified semantic data.

[0039] As mentioned above, grid-related information from various data sources is collected in real time through pre-configured API interfaces and IoT protocols. This data includes grid operation data, enterprise operation data, and policy data. Specifically, grid operation data originates from supervisory control and data acquisition systems, smart meters, and other monitoring devices; enterprise operation data includes financial records and project management information; and policy data is obtained by crawling government public information platforms and policy databases.

[0040] Feature extraction was performed on the collected multimodal data. For text-based policy data, natural language processing techniques were used for entity recognition and relation extraction to extract key elements such as policy name, applicable conditions, and subsidy standards. For time-series power grid operation data, time series analysis methods were used to identify its fluctuation patterns and key feature points, including statistical characteristics such as load extremes and changing trends. For image-based data containing power grid topology, computer vision technology was used to identify the equipment graphics and their connection relationships.

[0041] A multimodal feature fusion method is employed to map features from different sources and in different formats to a unified vector space. This process is achieved through a semantic embedding model, which transforms textual concepts, numerical features, and visual elements into vector representations with the same dimension, and establishes cross-modal semantic associations within this vector space, ultimately forming unified semantic data with consistent semantic expression.

[0042] According to one embodiment of this application, the step of constructing a dynamic knowledge graph based on the unified semantic data includes: An initial knowledge graph is constructed using a graph database, where nodes include policy entities, power grid equipment, operating parameters, and enterprise attributes, and edges represent semantic relationships between nodes. Design a streaming processing mechanism that automatically triggers incremental updates to the knowledge graph when new data flows in; The confidence fusion algorithm is used to detect and handle data conflicts during the update process to ensure the consistency of the knowledge graph. A timeliness-based weighting model is introduced to dynamically adjust node weights based on data source and timestamp.

[0043] As mentioned above, firstly, an initial knowledge graph is constructed using a graph database to transform unified semantic data into a graph structure representation. The nodes include four core entities: policy entity nodes represent policy provisions and their attributes; power grid equipment nodes represent physical equipment and their technical parameters; operating parameter nodes record power grid status data; and enterprise attribute nodes describe enterprise characteristic information. Nodes are connected by edges, which represent various semantic relationships, including applicability, subordination, and influence relationships.

[0044] A streaming processing mechanism is designed to support the dynamic updating of the knowledge graph. This mechanism continuously monitors data input, automatically identifying new entities and relationships when new unified semantic data flows in, and triggering an incremental update process for the knowledge graph. This mechanism ensures that the knowledge graph can reflect changes in power grid status and policy updates in real time.

[0045] Data conflicts during the update process are handled using a confidence-based fusion algorithm. This algorithm comprehensively considers the reliability of data sources, data collection time, and the quality indicators of the data itself, and performs weighted fusion of conflicting information from different sources. Through this mechanism, the consistency of internal data is ensured while maintaining the real-time nature of the knowledge graph.

[0046] A timeliness-based weighting model is introduced to optimize the knowledge graph. This model dynamically assigns weights to each node based on the authority of the data source and the timestamp of the data generation. More recent data and data from authoritative sources receive higher weights and have a stronger influence in subsequent queries and inferences. This design enables the knowledge graph to adaptively adjust the importance distribution of its internal structure.

[0047] According to one embodiment of this application, the step of performing policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results includes: The policy nodes and power grid data nodes in the dynamic knowledge graph are input into the graph neural network model, and the complex relationships between the nodes are learned through multi-layer graph convolution. Based on a multi-hop reasoning mechanism, the reasoning path related to user queries is traversed in the knowledge graph. By integrating reinforcement learning algorithms, the parameters of the matching model are dynamically adjusted based on user feedback to optimize matching accuracy; The policy calculation model is linked to the power grid operation status, and the weights of the calculation formula are dynamically adjusted through linear programming and time decay factors.

[0048] As described above, policy nodes and power grid data nodes from the dynamic knowledge graph are first input into a graph neural network model. This model then aggregates the feature information of adjacent nodes layer by layer through multi-layer graph convolution operations, thereby learning the complex nonlinear relationships between nodes. This process can capture the deep correlation between policy provisions and the power grid's operational status.

[0049] Next, based on a multi-hop reasoning mechanism, the reasoning path related to the user query is traversed in the knowledge graph. By making multiple jumps along node edges, potential connections between policy entities and power grid equipment and operating parameters are discovered, and a complete logical reasoning chain is constructed.

[0050] Then, reinforcement learning algorithms are integrated, using user feedback on the matching results as reward signals to dynamically adjust the parameters of the graph neural network model. Through continuous optimization, the model gradually adapts to the actual power grid operating environment and user needs, improving the accuracy of policy matching.

[0051] Finally, the policy calculation model is linked to the real-time operating status of the power grid, and the optimal matching scheme is solved using linear programming. A time decay factor is introduced to dynamically adjust the weights of each parameter in the calculation formula. This mechanism ensures that the policy matching results not only conform to the current power grid operating conditions but also adapt to the time-varying characteristics of policy effectiveness.

[0052] According to one embodiment of this application, the steps of inputting user query data into the dynamic knowledge graph when user query data is received, generating an intelligent response based on the policy matching result, and performing closed-loop optimization of the dynamic knowledge graph and the matching model based on user feedback include: It supports multimodal input including natural language, speech, and images, and uses a deep learning natural language understanding model to interpret user intent. Structured responses are generated based on the reasoning results of knowledge graphs, and the response content is dynamically optimized through user interaction feedback. Construct a knowledge subgraph of user profiles, integrating user historical behavior, job roles, and preference data to achieve personalized recommendations; User actions on the recommendation policy are fed back to the knowledge graph and matching model, forming a closed-loop optimization mechanism.

[0053] As mentioned above, it first supports multimodal input methods such as natural language, voice, and image. It uses a deep learning natural language understanding model to parse the semantic intent of user queries and transform them into structured queries that can be recognized by the knowledge graph.

[0054] The system generates structured responses based on reasoning results from knowledge graphs. These responses include policy matching details, applicability analysis, and operational recommendations. The system continuously collects user feedback to dynamically optimize the accuracy and completeness of the response content.

[0055] Construct a user profile knowledge subgraph, which integrates user historical query records, job responsibility characteristics, and personal preference data to form a personalized user model, and on this basis, realize differentiated policy recommendations.

[0056] User actions related to the recommendation policy, including adoption, ignoring, or modification, are transmitted in real time to the knowledge graph and matching model. The knowledge graph adjusts the node association weights based on the feedback, and the matching model updates its parameters using reinforcement learning, thus forming a closed-loop mechanism for continuous optimization.

[0057] According to one embodiment of this application, the design of the streaming processing mechanism, which automatically triggers incremental updates of the knowledge graph when new data flows in, includes the following steps: Real-time monitoring of data source changes; automatic extraction of new data when new policies are released or power grid events occur. Each new node is checked against an existing node in the knowledge graph. If they are different, the node is added as a new node. Check whether the newly added edge is the same as an existing edge in the knowledge graph. If they are different, add it as a new edge. If they are the same, update the edge weight. Incremental maintenance is performed on the updated knowledge graph, including transaction mechanisms, dependency checks, and redundancy cleanup.

[0058] As described above, the designed streaming processing mechanism automatically triggers incremental updates to the knowledge graph when new data flows in. Specifically, this includes the following process: The system continuously tracks changes in the data source status through a real-time monitoring module. When a new policy is released or a power grid event occurs, the system automatically triggers a data extraction process to obtain new data from the relevant source. Subsequently, the system parses the new data into nodes and edges of the knowledge graph and compares them with existing nodes. It determines whether a node is duplicated based on a unique identifier or key attribute; if a node does not exist, it is added as a new node. Similarly, a similar check is performed on new edges. If an edge does not exist, it is added as a new edge; if an edge already exists, its weight is updated based on data freshness or association strength. Finally, the system performs incremental maintenance on the updated knowledge graph, ensuring the atomicity and consistency of the update operation through a transaction mechanism, performing dependency checks to verify the logical rationality of relationships between nodes, and implementing redundancy cleanup to remove invalid or duplicate data elements, ensuring the optimized structure and efficient operation of the knowledge graph.

[0059] According to one embodiment of this application, the step of incrementally maintaining the updated knowledge graph includes: Each update operation is performed in a transactional manner to ensure the atomicity of the update; Check the dependencies between nodes to avoid conflicts between new rules and existing relationships; Remove invalid nodes and redundant edges to keep the knowledge graph structure compact.

[0060] As mentioned above, a transaction mechanism is first used to execute each update operation, ensuring the atomicity of knowledge graph updates. This means that all steps in each update process either execute successfully or are rolled back to the state before the update, thus avoiding data inconsistency issues caused by partial updates.

[0061] Next, a dependency check is performed to verify the logical consistency between the newly added nodes and edges and the existing graph structure. This process involves traversing the associated paths to detect whether the newly added policy rules conflict with existing device associations or operational parameter constraints, and then marking and handling any detected conflicts.

[0062] Finally, a structural optimization operation is performed. Based on preset failure determination rules, expired policy nodes and decommissioned equipment nodes are identified and deleted, while redundant edges generated by relationship updates are cleaned up. This process ensures the efficiency of subsequent queries and reasoning by maintaining the structural compactness of the knowledge graph.

[0063] A second aspect of this application provides a smart matching and optimization system for power grid policies based on multi-source data fusion and dynamic knowledge graphs, including: The data acquisition module is used to acquire multi-source power grid data and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A knowledge graph construction module is used to construct a dynamic knowledge graph based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; The matching and reasoning module is used to perform policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results. The interaction optimization module is used to input user query data into the dynamic knowledge graph when it receives user query data, generate an intelligent response based on the policy matching result, and perform closed-loop optimization of the dynamic knowledge graph and the matching model based on user feedback.

[0064] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any of the first aspects above.

[0065] Figure 2 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 2 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call logical instructions in the memory 830 to execute the method in any of the embodiments of the first aspect described above, the method including: Acquire multi-source power grid data, and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A dynamic knowledge graph is constructed based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; Based on a pre-trained graph neural network model, the dynamic knowledge graph is dynamically matched with policies and semantic reasoning is performed to obtain policy matching results. When user query data is received, the user query data is input into the dynamic knowledge graph, an intelligent response is generated based on the policy matching result, and the dynamic knowledge graph and matching model are optimized in a closed loop according to user feedback.

[0066] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0067] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer being able to perform the methods provided by the above methods, the method comprising: Acquire multi-source power grid data, and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A dynamic knowledge graph is constructed based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; Based on a pre-trained graph neural network model, the dynamic knowledge graph is dynamically matched with policies and semantic reasoning is performed to obtain policy matching results. When user query data is received, the user query data is input into the dynamic knowledge graph, an intelligent response is generated based on the policy matching result, and the dynamic knowledge graph and matching model are optimized in a closed loop according to user feedback.

[0068] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided by the above methods, the method comprising: Acquire multi-source power grid data, and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A dynamic knowledge graph is constructed based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; Based on a pre-trained graph neural network model, the dynamic knowledge graph is dynamically matched with policies and semantic reasoning is performed to obtain policy matching results. When user query data is received, the user query data is input into the dynamic knowledge graph, an intelligent response is generated based on the policy matching result, and the dynamic knowledge graph and matching model are optimized in a closed loop according to user feedback.

[0069] Example 2 1. System Configuration and Scenario Setting System Environment: This system is deployed on a cloud computing platform and adopts a microservice architecture, including data acquisition services, knowledge graph services, matching and inference engines, and an interactive front-end. Hardware resources include multiple servers for handling high-concurrency data streams and graphics computing.

[0070] Data source: Power grid operation data: Real-time power load curves (sampling frequency 1 minute) and photovoltaic power station output power (collected via IoT protocol) are obtained from the SCADA system.

[0071] Policy data: The latest policy texts are periodically retrieved from the government energy policy database API.

[0072] Enterprise data: Obtain financial data (such as R&D investment amount) and equipment ledgers from the enterprise's internal ERP system.

[0073] Meteorological data: Obtain data such as solar radiation intensity and temperature from the meteorological bureau's API.

[0074] User scenario: An operations staff member of a power grid company enters a query using natural language: "How do I apply for photovoltaic subsidies? My current photovoltaic power generation is 500kW, and the peak load is 800kW." 2. Real-time fusion of multi-source data and construction of dynamic knowledge graphs Data Acquisition and Semantic Alignment: The system collects multi-source data in real time via API and IoT protocols. For example, photovoltaic power generation data is input in time-series format, and policy text is parsed in JSON format.

[0075] Multimodal feature extraction: Entities are extracted from the policy text using an NLP model (based on a variant of BERT), including the policy name (e.g., "photovoltaic subsidy policy"), applicable conditions (e.g., "power generation must exceed 300kW"), and validity period.

[0076] Time series analysis is used on power grid operation data to extract peak and trough values ​​from the load curve, as well as the daily average and volatility of photovoltaic power generation. For example, peak load is calculated using a sliding window.

[0077] This formula is used to calculate the power grid over a time interval. The peak load within a given interval. Here, P(t) represents the load power value at time t, and the max operation extracts the maximum value within this interval. This helps identify peak periods in grid operation for policy matching (such as peak electricity subsidies).

[0078] The computer vision model (based on CNN) is used to extract the device relationships from the uploaded power grid topology map (such as PNG format), for example, to identify the connection relationship between photovoltaic inverters and transformers.

[0079] Semantic alignment: Multi-source data is mapped to the same vector space using a unified semantic representation model (based on cross-modal embedding). For example, the correlation between "peak electricity subsidy" in policy and "peak hours" in load curves is calculated using cosine similarity.

[0080] This formula calculates the policy text embedding vector. With power grid data feature vector Cosine similarity between them. Dot product. The consistency of vector direction is measured, and the denominator is normalized. The closer the result is to 1, the higher the semantic alignment, which is used to associate policy provisions with power grid operating parameters.

[0081] Dynamic knowledge graph construction and incremental updates: The knowledge graph is built on the Neo4j graph database, and the node types include: Policy entities: such as policy ID, name, and clauses.

[0082] Power grid equipment: such as photovoltaic panels, transformers, and smart meters.

[0083] Operating parameters: such as power generation, peak load, and carbon emissions.

[0084] Enterprise attributes: such as user roles and historical declaration records.

[0085] The relationships between the parties include: "applicable equipment under the policy", "impact on operating indicators", and "user preferences".

[0086] Streaming and Incremental Updates: When new policies are released or power generation data is updated, streaming engines (such as Apache Kafka) trigger knowledge graph updates.

[0087] Conflict detection algorithm: Based on confidence fusion, for example, using a weighted average for multiple data sources on the same policy.

[0088] When multiple data sources provide the weight of the same node, this formula calculates the final weight through a weighted average. Wherein, It represents the confidence level of the i-th data source (e.g., official policy sources have high confidence levels). These are the corresponding weights. This ensures the reliability and consistency of node weights in the knowledge graph.

[0089] Time-sensitivity weighting model: Node weights decay over time. For example, the weight of a policy node is calculated as follows:

[0090] This model describes the process of policy node weights decaying over time. λ is the initial weight, and λ is the decay factor (controlling the decay rate). This refers to the policy release time. Exponential decay ensures that new policies have a higher weight and older policies have a lower weight, maintaining the real-time nature of the knowledge graph.

[0091] 3. Dynamic matching and semantic reasoning of power grid policies Multi-dimensional matching degree evaluation model: Policy-Grid Correlation Model Based on Graph Neural Network (GNN): Input: Subgraphs in the knowledge graph, including policy nodes, photovoltaic equipment nodes, power generation nodes, etc.

[0092] Graph convolutional layers learn node embeddings and aggregate neighbor information through multi-layer propagation. The matching score is calculated as follows:

[0093] The formula uses a graph neural network (GNN) to calculate the degree of matching between policy and power grid data. and These are the embedding vectors for policy nodes and data nodes, respectively. This represents vector concatenation, where W and b are model parameters, and σ is the Sigmoid function that maps the output to between 0 and 1, representing the matching probability.

[0094] Multi-hop reasoning mechanism: For a user query "How to apply for photovoltaic subsidies?", the system traverses the path in the knowledge graph: Example path: Photovoltaic equipment → Power generation (500kW) → Subsidy policy (applicable conditions: power generation > 300kW) → Application conditions (power generation certificate required).

[0095] The inference path score is calculated based on a combination of edge weights and node similarity.

[0096] Reinforcement learning optimization: Update the matching model parameters based on the user's application result (success / failure) as a reward.

[0097] This formula is used for policy gradient updates in reinforcement learning to optimize the matching of model parameters. θ is the model parameter, α is the learning rate, and r is the actual reward (e.g., successful policy application). It is the baseline reward. This is the gradient of the policy function. The model is dynamically adjusted based on user feedback to improve matching accuracy.

[0098] Dynamic calculation formula optimization: The policy calculation model is linked to the grid operation status: for example, the photovoltaic subsidy amount is calculated as follows:

[0099] This formula calculates the amount of photovoltaic subsidies and dynamically adjusts it based on multi-source data. This is the average daily power generation. It is the basic subsidy rate. It is the grid absorption capacity adjustment factor (based on real-time curtailment rate). (Optimized through linear programming). Ensure that subsidy calculations are linked to the grid's operational status.

[0100] Time decay factor: Policy calculation priority is related to the power grid operating cycle.

[0101] This formula determines the priority of policy calculations, taking into account time decay and seasonality. β is the decay coefficient. It is the effective date of the policy. These are seasonal indicators (such as quarterly load trends). Priority decreases with policy obsolescence but is adjusted for seasonality.

[0102] 4. Intelligent Interaction and Closed-Loop Optimization Multimodal intelligent question answering engine: User input parsing: Natural language queries are parsed into structured intents using an NLU model (based on Transformer), such as identifying the entity "photovoltaic subsidy" and the condition "power generation 500kW".

[0103] Answer generation: Based on the reasoning results from the knowledge graph, a structured answer is generated, including: Matching policy list: such as policy name, estimated subsidy amount, and application steps.

[0104] Calculation details: Display The calculation process includes , Parameters such as these.

[0105] Dynamic optimization: If a user asks, "How can I increase the subsidy amount?", the system will combine real-time data to recommend adjustments to the power generation strategy (such as optimizing the tilt angle of the photovoltaic panels) and update the answer.

[0106] Personalized recommendations and strategy optimization: User profile construction: Generate a knowledge subgraph based on historical behavior (e.g., user role is an operations staff member), with nodes including user preferences (e.g., attention to subsidy policies) and job responsibilities.

[0107] Recommendation strategy: Calculate user-policy match using collaborative filtering and knowledge graph embedding.

[0108] This formula combines user profiles with policy relevance to generate personalized recommendation scores. It is the similarity between user embedding vectors and policy embedding vectors. λ is the base match score, and λ is the balancing factor. The higher the score, the higher the recommendation priority.

[0109] Feedback loop: User actions on recommended policies (such as successful application) are fed back to the knowledge graph, updating the policy node weights and user preference edges. For example, the policy weight increases for successful applications.

[0110] The formula updates the policy node weights based on user feedback. γ is the current weight, γ is the learning rate, and Δw is the feedback increment (Δw > 0 if the application is successful). This forms a closed-loop optimization, enabling the system to adapt to policy changes and user preferences.

[0111] 5. System Effects and Output Real-time response: The system returns results within 2 seconds of a user's query, including: Policy matching: Three applicable photovoltaic subsidy policies were identified.

[0112] Personalized recommendation: Based on the current power generation and load, we recommend applying for the "high penetration rate photovoltaic subsidy", with an estimated subsidy amount of 12,000 yuan / month.

[0113] Optimization tip: It is recommended to increase the use of energy storage equipment during periods of low load to improve eligibility for subsidies.

[0114] For any parts not mentioned in this application, existing technologies may be used or referenced.

[0115] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0116] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for intelligent matching of power grid policies based on multi-source data and knowledge graph, characterized in that, include: Acquire multi-source power grid data, and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A dynamic knowledge graph is constructed based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; Based on a pre-trained graph neural network model, the dynamic knowledge graph is dynamically matched with policies and semantic reasoning is performed to obtain policy matching results. When user query data is received, the user query data is input into the dynamic knowledge graph, an intelligent response is generated based on the policy matching result, and the dynamic knowledge graph and matching model are optimized in a closed loop according to user feedback.

2. The power grid policy intelligent matching and optimization method based on multi-source data fusion and dynamic knowledge graph according to claim 1, characterized in that, The steps of acquiring multi-source power grid data and performing semantic alignment processing on the multi-source power grid data to obtain unified semantic data include: Real-time collection of power grid operation data, enterprise operation data, and policy data through API interfaces and IoT protocols; Natural language processing was used to extract policy entities from text-based policy data, time series analysis was used to extract key features from time-series power grid operation data, and computer vision technology was used to extract equipment relationships from image-based data. A multimodal feature fusion method is used to map the extracted features to the same vector space, thereby achieving cross-modal semantic alignment and obtaining unified semantic data. 3.The power grid policy intelligent matching and optimization method based on multi-source data fusion and dynamic knowledge graph according to claim 1, characterized in that, The steps for constructing a dynamic knowledge graph based on the unified semantic data include: An initial knowledge graph is constructed using a graph database, where nodes include policy entities, power grid equipment, operating parameters, and enterprise attributes, and edges represent semantic relationships between nodes. Design a streaming processing mechanism that automatically triggers incremental updates to the knowledge graph when new data flows in; The confidence fusion algorithm is used to detect and handle data conflicts during the update process to ensure the consistency of the knowledge graph. A timeliness-based weighting model is introduced to dynamically adjust node weights based on data source and timestamp.

4. The method for intelligent matching and optimization of power grid policies based on multi-source data fusion and dynamic knowledge graph as described in claim 1, characterized in that, The step of performing policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results includes: The policy nodes and power grid data nodes in the dynamic knowledge graph are input into the graph neural network model, and the complex relationships between the nodes are learned through multi-layer graph convolution. Based on a multi-hop reasoning mechanism, the reasoning path related to user queries is traversed in the knowledge graph. By integrating reinforcement learning algorithms, the parameters of the matching model are dynamically adjusted based on user feedback to optimize matching accuracy; The policy calculation model is linked to the power grid operation status, and the weights of the calculation formula are dynamically adjusted through linear programming and time decay factors.

5. The method for intelligent matching and optimization of power grid policies based on multi-source data fusion and dynamic knowledge graph as described in claim 1, characterized in that, The steps of receiving user query data, inputting the user query data into the dynamic knowledge graph, generating an intelligent response based on the policy matching result, and performing closed-loop optimization of the dynamic knowledge graph and matching model based on user feedback include: It supports multimodal input including natural language, speech, and images, and uses a deep learning natural language understanding model to interpret user intent. Structured responses are generated based on the reasoning results of knowledge graphs, and the response content is dynamically optimized through user interaction feedback. Construct a knowledge subgraph of user profiles, integrating user historical behavior, job roles, and preference data to achieve personalized recommendations; User actions on the recommendation policy are fed back to the knowledge graph and matching model, forming a closed-loop optimization mechanism.

6. The method for intelligent matching and optimization of power grid policies based on multi-source data fusion and dynamic knowledge graph as described in claim 3, characterized in that, The described streaming processing mechanism automatically triggers incremental updates to the knowledge graph when new data flows in, including: Real-time monitoring of data source changes; automatic extraction of new data when new policies are released or power grid events occur. Each new node is checked against an existing node in the knowledge graph. If they are different, the node is added as a new node. Check whether the newly added edge is the same as an existing edge in the knowledge graph. If they are different, add it as a new edge. If they are the same, update the edge weight. Incremental maintenance is performed on the updated knowledge graph, including transaction mechanisms, dependency checks, and redundancy cleanup.

7. The method for intelligent matching and optimization of power grid policies based on multi-source data fusion and dynamic knowledge graph as described in claim 6, characterized in that, The steps for incrementally maintaining the updated knowledge graph include: Each update operation is performed in a transactional manner to ensure the atomicity of the update; Check the dependencies between nodes to avoid conflicts between new rules and existing relationships; Remove invalid nodes and redundant edges to keep the knowledge graph structure compact.

8. A smart matching system for power grid policies based on multi-source data and knowledge graphs, characterized in that, include: The data acquisition module is used to acquire multi-source power grid data and perform semantic alignment processing on the multi-source power grid data to obtain unified semantic data; A knowledge graph construction module is used to construct a dynamic knowledge graph based on the unified semantic data, wherein the dynamic knowledge graph includes policy entity nodes, power grid equipment nodes, and operating parameter nodes; The matching and reasoning module is used to perform policy dynamic matching and semantic reasoning on the dynamic knowledge graph based on a pre-trained graph neural network model to obtain policy matching results. The interaction optimization module is used to input user query data into the dynamic knowledge graph when it receives user query data, generate an intelligent response based on the policy matching result, and perform closed-loop optimization of the dynamic knowledge graph and the matching model based on user feedback.

9. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the method as described in any one of claims 1-7.

Images