A power grid architecture data processing method and system

Abstract

The application discloses a kind of power grid architecture data processing method and system, belong to data processing field, this method includes: obtaining current project information, and current query set is constructed based on current project information;Obtain historical knowledge graph, extract structure information and construct historical query set;According to semantic query set obtained by large language model;Current query set, historical query set and semantic query set are confidence score, and initial decision group is obtained and fitness function is constructed;Obtain initial evaluation particle swarm, based on fitness function and compression factor iterative calculation, when reaching maximum iteration number or the average fitness of initial evaluation particle swarm meets preset fitness requirement, final evaluation particle swarm is obtained, to obtain final decision group, to complete power grid architecture data processing process based on final decision group.The power grid architecture data processing method disclosed in the application can optimize the data processing architecture of power grid, and improve data processing efficiency.

Description

Technical Field

[0001] This invention belongs to the field of data processing, and in particular relates to a method for processing power grid architecture data. Background Technology

[0002] With the deepening of the digital transformation of power grid companies, architectural reviews of a massive number of construction projects are required annually. The core of this review involves precisely matching and verifying the consistency of project functionalities against the architectural artifacts regularly updated by the architecture department. These artifacts cover business architecture, data architecture, application architecture, and related tables, including several fields. Because architectural artifacts contain numerous complex one-to-many reference relationships, and the mapping results depend on manual judgment based on the specific business context of the functionalities, traditional manual verification methods are not only inefficient and prone to errors, but also difficult to adapt to the frequent iterations of architectural standards.

[0003] To improve processing efficiency, some existing technologies have attempted to incorporate natural language processing and large-scale modeling techniques to assist in power grid data governance. For example, some existing solutions construct knowledge graphs and data lineages to train dedicated large-scale models that automatically convert business rules described in natural language into SQL statements, thereby achieving automated data governance. However, the core intelligence of these existing technologies remains at the level of grammatical translation, lacking the ability to deeply analyze the semantics of complex business logic. When faced with numerous fuzzy mappings and one-to-many matching scenarios in architecture reviews, existing technologies cannot combine specific business contexts to evaluate the semantic fit of multiple candidate items and intelligently select the best option. Furthermore, their model architecture is relatively rigid, lacking a feedback loop that transforms manual review experience into system knowledge, resulting in the system's inability to continuously evolve with the changes in business rules and failing to meet the business needs of automated power grid architecture filling and verification. Therefore, existing large-scale power grid data processing lacks the ability to accurately analyze business processes due to its decision-making architecture, resulting in low data processing efficiency for power grid data. Summary of the Invention

[0004] The present invention aims to provide a power grid architecture data processing method and system to solve the above-mentioned technical problems, optimize the data processing architecture, and improve the processing efficiency of power grid data.

[0005] To address the aforementioned technical problems, this invention provides a power grid architecture data processing method, comprising the following steps: Obtain current project information and construct the current query set based on the current project information; Obtain a historical knowledge graph, extract structural information related to the current project information from the historical knowledge graph, and construct a historical query set based on the structural information and the current project information; The current project information is expanded and enhanced based on a preset large language model to obtain a semantic query set; Confidence scores are applied to the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and a fitness function is constructed based on the initial decision group. An initial evaluation particle swarm is obtained, and the initial evaluation particle swarm is iteratively calculated based on the fitness function and the pre-acquired compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. The weight values ​​of the initial decision group are determined based on the final evaluation particle swarm, and the final decision group is obtained to complete the power grid architecture data processing process based on the final decision group.

[0006] In the above scheme, a current query set, a historical query set, and a semantic query set are constructed based on current project information, historical knowledge graph, and a large language model. Confidence scores are then applied to these sets to obtain an initial decision group and construct a fitness function. This transforms multi-level query results, including direct semantic information from the current project, historical knowledge graph information, and generalized semantic information, into quantifiable indicators. This improves the ability to perform deep semantic analysis of complex business logic and optimizes the power grid data processing architecture. Furthermore, the initial evaluation particle swarm is iteratively calculated based on the fitness function and compression factor to obtain the final evaluation particle swarm. This determines the weights of the initial decision group and leads to the final decision group. By calculating the average fitness of the evaluation particle swarm, the scheme overcomes the low accuracy of traditional particle optimization methods, ensuring that the weight confirmation process can quickly and stably obtain the optimal solution. This solves the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a multi-source information dynamic fusion mechanism in the existing power grid processing architecture. This optimizes the power grid data processing architecture and improves data processing efficiency.

[0007] Further, confidence scores are applied to the current query set, the historical query set, and the semantic query set to obtain an initial decision group. A fitness function is then constructed based on the initial decision group, including: performing item association analysis on the current query set, the historical query set, and the semantic query set respectively to obtain a current confidence score, a historical confidence score, and a semantic confidence score; calculating a comprehensive confidence weight set based on the current confidence score, the historical confidence score, and the semantic confidence score; reviewing the comprehensive confidence weight set according to preset review requirements to obtain a review opinion set; merging the current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set to obtain the initial decision group; and constructing a fitness function based on the initial decision group.

[0008] In the above scheme, by performing item association analysis on the current query set, historical query set, and semantic query set respectively, the three-layer query results can be transformed into quantifiable confidence score indicators, realizing a mechanism from semantic matching to numerical evaluation. By calculating a comprehensive confidence weight set from the current confidence score, historical confidence score, and semantic confidence score, multi-source query information can be structurally integrated, avoiding decision-making bias caused by insufficient coverage of a single query path. By reviewing the comprehensive confidence weight set according to preset review requirements and obtaining a review opinion set, a controllable collaborative interface can be established between automated decision-making and human expert experience, ensuring efficient automatic processing of high-confidence scenarios and accurate submission for review in low-confidence scenarios, balancing efficiency and reliability. By combining the current query set, historical query set, semantic query set, comprehensive confidence weight set, and review opinion set to obtain an initial decision group, and constructing a fitness function based on the initial decision group, the multi-source query set, quantitative confidence index, and review feedback can be uniformly encapsulated into an iterative optimization objective. This ensures that the weight adaptive adjustment process always revolves around business data-driven decision-making. This method addresses the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic multi-source information fusion mechanism in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0009] Further, after determining the weight values ​​of the initial decision group based on the final evaluation particle swarm to obtain the final decision group, and completing the power grid architecture data processing process based on the final decision group, the process includes: determining the comprehensive confidence weight set of the initial decision group based on the final evaluation particle swarm to obtain the final confidence weight set; merging the current query set, the historical query set, the semantic query set, the final confidence weight set, and the review opinion set to obtain the final decision group; introducing a bypass matrix and training the bypass matrix based on the final decision group; obtaining a reliable bypass matrix when the bypass matrix meets preset matrix requirements; and updating the large language model based on the reliable bypass matrix to obtain the final large language model, thereby completing the iterative update process of the large language model.

[0010] Further, the step of obtaining the initial evaluation particle swarm, which iteratively calculates the initial evaluation particle swarm based on the fitness function and a pre-acquired compression factor, and obtains the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement, includes: obtaining the initial evaluation particle swarm, and obtaining the initial particle swarm position and initial particle swarm velocity based on the initial evaluation particle swarm; obtaining the optimal position of the particle swarm based on the initial particle swarm position; iteratively updating the initial particle swarm position and the initial particle swarm velocity based on the pre-acquired compression factor and the optimal position of the particle swarm, and calculating the average fitness of the initial evaluation particle swarm based on the fitness function in each iteration; and obtaining the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement.

[0011] In the above scheme, by acquiring an initial evaluation particle swarm and obtaining its initial particle swarm position and velocity, and then determining its optimal position, the optimal search direction under the current fitness function can be quickly established, providing a clear reference benchmark for subsequent iterative calculations. The initial particle swarm position and velocity are iteratively updated using a pre-acquired compression factor and the optimal particle swarm position. In each iteration, the average fitness of the initial evaluation particle swarm is calculated based on the fitness function. This allows for dynamic balancing of the global exploration and local development search step size using the compression factor, and real-time evaluation of the convergence state by combining the trend of average fitness changes. This effectively overcomes the shortcomings of traditional particle optimization methods, such as premature convergence and low accuracy, ensuring that the fitness function can stably and efficiently approach the global optimum. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirements, the final evaluation particle swarm is obtained. Iterative calculations ensure that the output weight values ​​accurately adapt to the actual business distribution of the current power grid architecture data. This method solves the technical problem that existing power grid processing architectures are difficult to update with business rules, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0012] Furthermore, the step of obtaining current project information and constructing a current query set based on the current project information includes: obtaining current project information and extracting features from the current project information to obtain a current project query term set; constructing the large language model based on the pre-acquired project knowledge base, and querying the current project query term set according to the large language model to obtain the current query set.

[0013] In the above solution, by acquiring current project information and extracting features from it to obtain a current project query term set, lengthy project data can be transformed into semantic elements, ensuring that subsequent queries directly focus on business data. By constructing a large language model based on a pre-acquired project knowledge base and querying the current project query term set according to this model, a current query set can be obtained. This allows for semantic alignment queries through the large language model, enabling the understanding of the business connotations of power grid terminology and directly obtaining information highly relevant to the current project. This results in highly reliable query results, solving the technical problem of low matching accuracy caused by the lack of a multi-source information dynamic fusion mechanism in the existing power grid processing architecture. This optimizes the power grid's data processing architecture and improves data processing efficiency.

[0014] Furthermore, the step of acquiring a historical knowledge graph, extracting structural information related to the current project information from the historical knowledge graph, and constructing a historical query set based on the structural information and the current project information includes: constructing a historical knowledge graph based on pre-acquired historical power grid data; performing graph traversal in the historical knowledge graph based on the current project query term set to extract structural information; and querying the structural information and the current project query term set according to the large language model to obtain the historical query set.

[0015] It should be noted that the structural information may include entities, neighboring entities, relationship paths, and attributes related to the current project query term set.

[0016] In the above scheme, a historical knowledge graph is constructed based on pre-acquired historical power grid data. This transforms discrete historical power grid data into a graph-structured knowledge graph. By extracting structural information through graph traversal based on the current project query term set, the dependency logic of the current project can be mined along the node relationship paths of the structural information, supplementing the associated context not covered by direct text queries. By querying the structural information and the current project query term set according to the large language model, a historical query set is obtained. This allows for semantic alignment and reorganization of the graph data, generating query results that match the current project query term set. Through this method, a structured query dimension based on historical evolution patterns and business data association is constructed, solving the technical problem of the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture. This optimizes the power grid data processing architecture and improves data processing efficiency.

[0017] Furthermore, the step of expanding and enhancing the current project information according to the preset large language model to obtain a semantic query set includes: expanding the relevance of the current project query term set based on the large language model to obtain an expanded query term set; and performing enhanced queries on the expanded query term set according to the large language model to obtain a semantic query set.

[0018] In the above scheme, by expanding the current project query term set based on the large language model to obtain the extended query term set, it is possible to overcome the limitations of semantic expression in querying and generate generalized query elements covering synonyms and related terms, i.e., a semantic query set, providing a more comprehensive semantic coverage foundation. By enhancing the query of the extended query term set through the large language model to obtain the semantic query set, the generalized semantic query set can be used to recall contextual data that is potentially related to the current project information in business logic but has significant literal differences, forming a supplementary query dimension. This method solves the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0019] The present invention also provides a power grid architecture data processing system, comprising: The query set acquisition module is used to acquire current project information and construct a current query set based on the current project information; acquire a historical knowledge graph, extract structural information related to the current project information from the historical knowledge graph, and construct a historical query set based on the structural information and the current project information; and expand and enhance the current project information according to a preset large language model to obtain a semantic query set. The confidence scoring module is used to perform confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and to construct a fitness function based on the initial decision group; The iterative calculation module is used to obtain an initial evaluation particle swarm, and to iteratively calculate the initial evaluation particle swarm based on the fitness function and the pre-obtained compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. The decision group confirmation module is used to determine the weight values ​​of the initial decision group based on the final evaluation particle swarm, and obtain the final decision group, so as to complete the power grid architecture data processing process based on the final decision group.

[0020] The system architecture provided by the above solution is logically clear, with well-defined functional divisions and close collaboration among modules, forming a complete power grid architecture data processing flow. Each module performs logical synchronization control over the entire process of data acquisition, confidence assessment, weight optimization, and decision output, ensuring data consistency at each stage and avoiding additional errors caused by mapping deviations. This system achieves power grid architecture data processing through the collaborative work of its modules. The query set acquisition module constructs multi-source query sets through a three-layer retrieval strategy, providing a comprehensive and structured information input foundation for subsequent decision-making. The confidence scoring module establishes a controllable collaborative interface between automated decision-making and human expert experience through multi-source confidence quantification and review feedback mechanisms. The iterative calculation module overcomes the shortcomings of traditional optimization methods, such as premature convergence and low convergence accuracy, by introducing an improved particle swarm optimization algorithm with compression factors and adaptive inertia strategies. The decision group confirmation module outputs a highly robust comprehensive decision confidence level through dynamic weight allocation, ensuring automatic locking of the optimal architecture entry in complex scenarios of one-to-many mapping across tables. The logical synchronization of each module ensures consistent data sources at each stage, thereby ensuring the stable operation of the entire system. This collaborative mechanism addresses the technical issues of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0021] Further, the confidence scoring module is used to perform confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and to construct a fitness function based on the initial decision group, including: performing item association analysis on the current query set, the historical query set, and the semantic query set respectively to obtain a current confidence score, a historical confidence score, and a semantic confidence score; calculating a comprehensive confidence weight set based on the current confidence score, the historical confidence score, and the semantic confidence score; reviewing the comprehensive confidence weight set according to preset review requirements to obtain a review opinion set; merging the current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set to obtain the initial decision group; and constructing a fitness function based on the initial decision group.

[0022] Further, the iterative calculation module is used to obtain an initial evaluation particle swarm, iteratively calculate the initial evaluation particle swarm based on the fitness function and a pre-acquired compression factor, and obtain a final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement. This includes: obtaining the initial evaluation particle swarm, and obtaining the initial particle swarm position and initial particle swarm velocity based on the initial evaluation particle swarm; obtaining the optimal position of the particle swarm based on the initial particle swarm position; iteratively updating the initial particle swarm position and initial particle swarm velocity based on the pre-acquired compression factor and the optimal position of the particle swarm; calculating the average fitness of the initial evaluation particle swarm based on the fitness function in each iteration; and obtaining the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement.

[0023] Furthermore, the query set acquisition module is used to acquire current project information and construct a current query set based on the current project information, including: acquiring current project information and extracting features from the current project information to obtain a current project query term set; constructing the large language model based on the pre-acquired project knowledge base, and querying the current project query term set according to the large language model to obtain the current query set.

[0024] In the above scheme, the query set acquisition module forms a query set by acquiring data, and the input confidence scoring module performs item association analysis on the current query set, the historical query set, and the semantic query set respectively to obtain the current confidence score, the historical confidence score, and the semantic confidence score, transforming the multi-level search results into quantifiable indicators; by calculating the comprehensive confidence weight set and reviewing it according to preset review requirements, a review opinion set is obtained, establishing a collaborative interface between automated decision-making and human experience; by merging multi-source data to obtain an initial decision group and constructing a fitness function, a mechanism leap from semantic matching to numerical evaluation is achieved. Furthermore, the iterative calculation module acquires the initial evaluation particle swarm and determines its position and velocity. Based on the fitness function and the pre-acquired compression factor, it iteratively updates the initial particle swarm position and velocity. Combining the average fitness of each iteration with real-time evaluation of the convergence state, it dynamically balances the optimization step size between global exploration and local development. This overcomes the shortcomings of traditional optimization methods, such as premature convergence and low accuracy, ensuring that the weight value confirmation process quickly and stably approaches the optimal solution, outputting a final evaluation particle swarm that accurately adapts to the actual business distribution. Through the synergistic linkage of the above modules, a complete processing chain from feature extraction and multi-source confidence quantization to adaptive weight optimization is constructed. This solves the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency. Attached Figure Description

[0025] Figure 1 This is a schematic flowchart of a power grid architecture data processing method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a power grid architecture data processing system according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 This embodiment provides a power grid architecture data processing method, including the following steps: Step S1: Obtain the current project information and construct the current query set based on the current project information; Step S2: Obtain a historical knowledge graph, extract structural information related to the current project information from the historical knowledge graph, and construct a historical query set based on the structural information and the current project information; Step S3: Expand and enhance the current project information according to the preset large language model to obtain a semantic query set; Step S4: Calculate confidence scores for the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and construct a fitness function based on the initial decision group; Step S5: Obtain the initial evaluation particle swarm, and iteratively calculate the initial evaluation particle swarm based on the fitness function and the pre-acquired compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. Step S6: Determine the weight values ​​of the initial decision group based on the final evaluation particle swarm to obtain the final decision group, and complete the power grid architecture data processing process based on the final decision group.

[0028] In this embodiment, a current query set, a historical query set, and a semantic query set are constructed based on current project information, historical knowledge graph, and a large language model. Confidence scores are then applied to these sets to obtain an initial decision group and construct a fitness function. This transforms multi-level query results, including direct semantic information of the current project, historical knowledge graph information, and generalized semantic information, into quantifiable indicators, thereby improving the ability to perform deep semantic analysis of complex business logic and optimizing the power grid data processing architecture. Furthermore, the initial evaluation particle swarm is iteratively calculated based on the fitness function and compression factor to obtain the final evaluation particle swarm. This determines the weight values ​​of the initial decision group and leads to the final decision group. By calculating the average fitness of the evaluation particle swarm, the low accuracy of traditional particle optimization methods is overcome, ensuring that the weight value confirmation process can quickly and stably obtain the optimal solution. This solves the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a multi-source information dynamic fusion mechanism in the existing power grid processing architecture, thereby optimizing the power grid data processing architecture and improving data processing efficiency.

[0029] It should be noted that when the power grid company conducts an architecture review for each project, it needs to match the functional points of each construction project with the corresponding content of the architecture artifacts provided by the architecture department in the previous year. Because the power grid company's internal applications are constantly iterating, the architecture artifacts are also updated. These architecture artifacts include business architecture tables, data architecture tables, application architecture tables, etc. These artifacts have complex correspondences, spanning dozens of tables across business architecture, data architecture, and application architecture, and contain many one-to-many relationships that require manual filtering based on the specific content of the functional points.

[0030] Furthermore, the step of obtaining current project information and constructing a current query set based on the current project information includes: obtaining current project information and extracting features from the current project information to obtain a current project query term set; constructing the large language model based on the pre-acquired project knowledge base, and querying the current project query term set according to the large language model to obtain the current query set.

[0031] It should be noted that the current project information may be architectural artifacts uploaded from the power grid system, including business architecture tables, data architecture tables, and application architecture tables.

[0032] In one embodiment, obtaining the current project information and extracting features from it to obtain the current project query term set involves directly using the cell content of the current project information to be processed as the current project query term set Q. Then, a large language model is constructed using a pre-acquired project knowledge base, and semantic vector queries are performed on the current project query term set Q using this large language model. This method most closely matches the original query intent and yields highly relevant content, denoted as the current query set R. h .

[0033] In this embodiment, by acquiring current project information and extracting features from it to obtain a current project query term set, lengthy project data can be transformed into semantic elements, ensuring that subsequent queries directly focus on business data. By constructing a large language model based on a pre-acquired project knowledge base and querying the current project query term set according to this model, a current query set can be obtained. This allows for semantic alignment queries using the large language model, enabling the understanding of the business connotations of power grid terminology and directly obtaining information highly relevant to the current project. This results in highly reliable query results, solving the technical problem of low matching accuracy caused by the lack of a multi-source information dynamic fusion mechanism in existing power grid processing architectures. This optimizes the power grid's data processing architecture and improves data processing efficiency.

[0034] Furthermore, the step of acquiring a historical knowledge graph, extracting structural information related to the current project information from the historical knowledge graph, and constructing a historical query set based on the structural information and the current project information includes: constructing a historical knowledge graph based on pre-acquired historical power grid data; performing graph traversal in the historical knowledge graph based on the current project query term set to extract structural information; and querying the structural information and the current project query term set according to the large language model to obtain the historical query set.

[0035] It should be noted that the structural information may include entities, neighboring entities, relationship paths, and attributes related to the current project query term set.

[0036] In this embodiment, the pre-acquired historical power grid data includes historical documents such as old version architecture materials. A historical knowledge graph is constructed using information extraction technology. Starting with the current project query term set Q, structural information related to the current project query term set Q is extracted from the historical knowledge graph, including entities, neighboring entities, relationship paths, and attributes. A large language model is used to combine this structural information with the current project query term set for querying. This captures logically related information not directly reflected in the query, defining it as a highly relevant semantic set, denoted as the historical query set R. m .

[0037] In this embodiment, a historical knowledge graph is constructed based on pre-acquired historical power grid data. This transforms discrete historical power grid data into a graph-structured knowledge graph. By extracting structural information through graph traversal based on the current project query term set within this historical knowledge graph, the dependency logic of the current project can be mined along the node relationship paths of the structural information, supplementing the associated context not covered by direct text queries. By querying the structural information and the current project query term set according to the large language model, a historical query set is obtained. This allows for semantic alignment and reorganization of the graph data, generating query results that match the current project query term set. Through this method, a structured query dimension based on historical evolution patterns and business data association is constructed, solving the technical problem of the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture. This optimizes the power grid's data processing architecture and improves data processing efficiency.

[0038] Furthermore, the step of expanding and enhancing the current project information according to the preset large language model to obtain a semantic query set includes: expanding the relevance of the current project query term set based on the large language model to obtain an expanded query term set; and performing enhanced queries on the expanded query term set according to the large language model to obtain a semantic query set.

[0039] In this embodiment, the original current project query term set Q is expanded with synonyms and related terms using a large language model to generate an expanded query term set Q′. Then, based on the large language model, paragraphs matching any keyword in Q′ are queried from the project knowledge base to obtain a semantic query set R. l This set has relatively weak semantic relevance and is only used as a supplementary query dimension.

[0040] In this embodiment, by expanding the current project query term set based on the large language model to obtain the extended query term set, the limitations of semantic expression in querying can be overcome, generating generalized query elements covering synonyms and related terms, i.e., a semantic query set, providing a more comprehensive semantic coverage foundation. The semantic query set is obtained by enhancing the query on the extended query term set using the large language model. This generalized semantic query set can be used to recall contextual data that is potentially related to the current project information in business logic but has significant literal differences, forming a supplementary query dimension. This method solves the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0041] Further, confidence scores are applied to the current query set, the historical query set, and the semantic query set to obtain an initial decision group. A fitness function is then constructed based on the initial decision group, including: performing item association analysis on the current query set, the historical query set, and the semantic query set respectively to obtain a current confidence score, a historical confidence score, and a semantic confidence score; calculating a comprehensive confidence weight set based on the current confidence score, the historical confidence score, and the semantic confidence score; reviewing the comprehensive confidence weight set according to preset review requirements to obtain a review opinion set; merging the current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set to obtain the initial decision group; and constructing a fitness function based on the initial decision group.

[0042] In the above embodiments, for candidate output link t i , will the current query set R h Historical query set R m and semantic query set R l They were submitted together to the large language scoring model, which performed three independent semantic association analyses to evaluate t. i The functional fit between the text and the business descriptions described at different levels of information. This process can be formalized into three mapping functions: ; ; ; In the formula, These represent the current query set R respectively. h Historical query set R m and semantic query set R l Candidate links t obtained from the analysis iThe confidence scores are the current confidence score, the historical confidence score, and the semantic confidence score. Based on these, a comprehensive confidence weight set is then calculated, specifically: ; In the formula, , and These correspond to the weighting coefficients of the three confidence scores, and the sum of these three weighting coefficients is 1. The initial values ​​of the weighting coefficients can be set as ξ=0.6, β=0.3, and γ=0.1.

[0043] If there exists a candidate link t k Its confidence weight S k The candidate link is significantly higher than all other links, and is marked as a recommended item for manual review. It is then submitted for manual review to obtain a set of review opinions based on preset review requirements. The current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set are then combined to obtain the initial decision group D=(I,C,O). Here, I represents the set of queries including the current query set R. h The historical query set R m The semantic query set R l The set of ; C is the confidence weight including all candidate output links, i.e., the comprehensive confidence weight set Si, and O is the review opinion set.

[0044] Using a historical sample set of decision groups D = (D1, D2, ..., D...) n Using D as the validation set, construct the fitness function F(w). For each sample D K =(I k C k O k Given weight w, the process of constructing the fitness function is as follows: ; Where N is the size of the decision group sample set, O k For the k-th review opinion set, The set that matches the review opinion set; 1(·) is the indicator function, which is 1 when there is a perfect match; Sim(·) is the partial similarity, and Jaccard similarity is calculated when the predicted candidate set overlaps with the expert review results. This is a balancing factor used to weigh the importance of exact and approximate matching, typically ranging from 0.7 to 0.9. This fitness function encourages precise decision-making while preserving gradient information from partial matches, which helps guide the particle search. Jaccard similarity is a statistical metric used to compare the similarity between two sets, measuring their similarity by calculating the ratio of the intersection size to the union size.

[0045] In this embodiment, by performing item association analysis on the current query set, historical query set, and semantic query set respectively, the three-layer query results can be transformed into quantifiable confidence scores, realizing a mechanism from semantic matching to numerical evaluation and providing an objective basis for subsequent weight allocation. By calculating a comprehensive confidence weight set from the current confidence score, historical confidence score, and semantic confidence score, multi-source query information can be structurally integrated, avoiding decision-making bias caused by insufficient coverage of a single query path. By reviewing the comprehensive confidence weight set according to preset review requirements and obtaining a review opinion set, a controllable collaborative interface can be established between automated decision-making and human expert experience, ensuring efficient automatic processing of high-confidence scenarios and accurate submission for review in low-confidence scenarios, balancing efficiency and reliability. By combining the current query set, historical query set, semantic query set, comprehensive confidence weight set, and review opinion set to obtain an initial decision group, and constructing a fitness function based on the initial decision group, multi-source query evidence, quantitative confidence indicators, and review feedback can be uniformly encapsulated into an iterative optimization objective. This ensures that the weight adaptive adjustment process always revolves around business data-driven decision-making. This method addresses the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic multi-source information fusion mechanism in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0046] Further, after determining the weight values ​​of the initial decision group based on the final evaluation particle swarm to obtain the final decision group, and completing the power grid architecture data processing process based on the final decision group, the process includes: determining the comprehensive confidence weight set of the initial decision group based on the final evaluation particle swarm to obtain the final confidence weight set; merging the current query set, the historical query set, the semantic query set, the final confidence weight set, and the review opinion set to obtain the final decision group; introducing a bypass matrix and training the bypass matrix based on the final decision group; obtaining a reliable bypass matrix when the bypass matrix meets preset matrix requirements; and updating the large language model based on the reliable bypass matrix to obtain the final large language model, thereby completing the iterative update process of the large language model.

[0047] In this embodiment, after each data processing step based on the final decision-making group and subsequent manual review, the system automatically captures the current project information, the final decision-making group, and its corresponding final confidence weight set. These, along with the manual review confirmation result, are used to construct a decision triplet sample D=(I,C,O). Here, I encompasses the current project's functional point information and the current query set, historical query set, and semantic query set; C contains all candidate output links and the comprehensive confidence score calculated based on the final confidence weight set; and O represents the correct architecture mapping link confirmed by the manual review or a rejection flag. When the accumulated number of decision triplet samples reaches a preset trigger threshold, the system automatically initiates the incremental fine-tuning and rule base expansion process of the large language model. The first step is to construct the instruction fine-tuning dataset. The system traverses the sample library, transforming each decision triple into an instruction format that the model can train. Using a preset prompt word template, I and C are concatenated to form the input prompt word, and O is used as the standard output answer, forming a supervised training pair corresponding to the input and output. This data construction process enables the model to explicitly learn, under specific multi-dimensional contextual information and candidate confidence distributions, to make decision judgments consistent with those of domain experts.

[0048] The second step involves performing incremental fine-tuning based on the side-path matrix using low-rank adaptation (LoRA). The side-path matrix is ​​a trainable low-rank parameter matrix appended to the weight matrix of a specific layer in the large language model. During fine-tuning, the system freezes all original parameters of the large language model base and performs supervised training only on the side-path matrix with an extremely low learning rate. The training loop specifically involves: feeding the input prompts from the instruction dataset into the model for forward propagation to generate the model's predicted output; calculating the cross-entropy loss between the predicted output and the standard triplet answer O; calculating the gradient based on the loss value through backpropagation and updating the parameter weights of the side-path matrix; and iterating through multiple rounds until the accuracy of the decision results generated by the side-path matrix on the validation set converges to the preset matrix requirements, at which point a reliable side-path matrix incorporating expert decision-making wisdom is obtained.

[0049] The third step is parameter smoothing replacement and model update deployment. After training, the number of parameters in the reliable bypass matrix is ​​only on the order of one-thousandth of the original model. The system smoothly merges its low-rank weights with the frozen base model parameters or independently mounts them to complete the online update of the large language model, resulting in the final large language model. After the update, the system uses a reserved validation sample set to evaluate the effect of the new model. When the decision accuracy exceeds the preset standard, it is officially activated, ensuring that the model retains its general semantic understanding ability while continuously absorbing business experience and avoiding forgetting.

[0050] In this embodiment, the system performs high-frequency pattern mining and explicit rule distillation on the accumulated decision triplet sample library. The system extracts decision paths and context mapping relationships that meet preset frequency conditions using a sequence pattern mining algorithm, transforms them into candidate business rules, and automatically adds them to the project knowledge base and rule base used to construct the current query set and historical query set. Thus, the system forms a dynamic adaptive iterative closed loop, enabling the power grid architecture data processing architecture to continuously evolve with the business rules, achieving automatic update and optimization, and completely solving the technical problems of low matching accuracy and difficulty in updating with rules in traditional architecture review.

[0051] Further, the step of obtaining the initial evaluation particle swarm, which iteratively calculates the initial evaluation particle swarm based on the fitness function and a pre-acquired compression factor, and obtains the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement, includes: obtaining the initial evaluation particle swarm, and obtaining the initial particle swarm position and initial particle swarm velocity based on the initial evaluation particle swarm; obtaining the optimal position of the particle swarm based on the initial particle swarm position; iteratively updating the initial particle swarm position and the initial particle swarm velocity based on the pre-acquired compression factor and the optimal position of the particle swarm, and calculating the average fitness of the initial evaluation particle swarm based on the fitness function in each iteration; and obtaining the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement.

[0052] In the above embodiment, based on the initial decision group D=(I,C,O) and the fitness function F(w), M particles are randomly generated to obtain the initial evaluation particle swarm, where each particle represents a set of candidate weights. Each particle's position is randomly initialized within the constrained space, and its velocity v i Random initialization in [−v max ,v max Within a preset velocity range, the initial particle swarm position and initial particle swarm velocity are obtained. The current fitness value of each particle is calculated according to the aforementioned fitness function, and the optimal position of the particle swarm, including the individual's historical optimal position p, is recorded. best and the global optimal position g best A rate update formula with a compression factor χ is introduced to balance local and global search capabilities: ; ; In the formula, d=1,2,3 represent dimensions, which correspond to the weight coefficients of the three confidence scores mentioned above. , and t is the number of iterations; c1 and c2 are inertial weights; c1 and c2 are learning factors, typically taking the value 2; r1 and r2 are uniformly random numbers in the range [0,1]; χ is the compression factor, calculated from the learning factors. ; In the formula The learning factors c1 and c2 are sums and need to be greater than 4, typically 4.1; the compression factor χ is usually 0.729, which ensures the convergence of the algorithm. The inertia weights in the above embodiments... The adjustment is made dynamically according to the number of iterations, specifically as follows: ; In the formula, A minimum value is preset for the inertia weight. The inertia weight is preset to a maximum value; T is the maximum number of iterations; This is the attenuation factor, which can take a value of 3; Let be the average fitness of the initial evaluated particle swarm at iteration number t. The iterative calculation process described above ensures that when the swarm fitness clusters, the inertia weight automatically decreases, promoting fine-grained local search; conversely, when the swarm disperses, the inertia weight increases, maintaining global exploration capability. This process continues until the maximum iteration number T or the global optimum is reached. The fitness of the particle swarm is compared with the average fitness of the initial evaluation particle swarm. When the difference between the two values ​​is less than a preset threshold ϵ, the iteration stops and the final position is output. This process yields the final evaluated particle swarm. The weight values ​​of the initial decision group are then determined. The value is used to obtain the final decision group and complete the power grid architecture data processing process.

[0053] In this embodiment, by acquiring an initial evaluation particle swarm and obtaining its initial particle swarm position and velocity, and then determining its optimal position, the optimal optimization direction under the current fitness function can be quickly established, providing a clear reference benchmark for subsequent iterative calculations. The initial particle swarm position and velocity are iteratively updated using a pre-acquired compression factor and the optimal particle swarm position. In each iteration, the average fitness of the initial evaluation particle swarm is calculated based on the fitness function. This allows for dynamic balancing of the optimization step size between global exploration and local development using the compression factor, and real-time evaluation of the convergence state by combining the trend of average fitness changes. This effectively overcomes the shortcomings of traditional particle optimization methods, such as premature convergence and low accuracy, ensuring that the fitness function can stably and efficiently approach the global optimum. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. Iterative calculations ensure that the output weight values ​​accurately adapt to the actual business distribution of the current power grid architecture data. This method solves the technical problem that existing power grid processing architectures are difficult to update with business rules, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0054] like Figure 2 As shown, this embodiment also provides a power grid architecture data processing system, including: The query set acquisition module is used to acquire current project information and construct a current query set based on the current project information; acquire a historical knowledge graph, extract structural information related to the current project information from the historical knowledge graph, and construct a historical query set based on the structural information and the current project information; and expand and enhance the current project information according to a preset large language model to obtain a semantic query set. The confidence scoring module is used to perform confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and to construct a fitness function based on the initial decision group; The iterative calculation module is used to obtain an initial evaluation particle swarm, and to iteratively calculate the initial evaluation particle swarm based on the fitness function and the pre-obtained compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. The decision group confirmation module is used to determine the weight values ​​of the initial decision group based on the final evaluation particle swarm, and obtain the final decision group, so as to complete the power grid architecture data processing process based on the final decision group.

[0055] The system architecture provided in the above embodiments is logically clear, with well-defined functional divisions and close collaboration among modules, forming a complete power grid architecture data processing flow. Each module performs logical synchronization control over the entire process of data acquisition, confidence assessment, weight optimization, and decision output, ensuring data consistency at each stage and avoiding additional errors caused by mapping deviations. This system achieves power grid architecture data processing through the collaborative work of its various modules. The query set acquisition module constructs a multi-source query set using a three-layer retrieval strategy, providing a comprehensive and structured information input foundation for subsequent decision-making. The confidence scoring module establishes a controllable collaborative interface between automated decision-making and human expert experience through multi-source confidence quantification and review feedback mechanisms. The iterative calculation module overcomes the shortcomings of traditional optimization methods, such as premature convergence and low convergence accuracy, by introducing an improved particle swarm optimization algorithm with compression factors and adaptive inertia strategies. The decision group confirmation module outputs a highly robust comprehensive decision confidence level through dynamic weight allocation, ensuring automatic locking of the optimal architecture entry in complex scenarios of one-to-many mapping across tables. The logical synchronization of each module ensures consistent data sources at each stage, thereby ensuring stable operation of the entire system. This collaborative mechanism addresses the technical issues of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0056] Further, the confidence scoring module is used to perform confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and to construct a fitness function based on the initial decision group, including: performing item association analysis on the current query set, the historical query set, and the semantic query set respectively to obtain a current confidence score, a historical confidence score, and a semantic confidence score; calculating a comprehensive confidence weight set based on the current confidence score, the historical confidence score, and the semantic confidence score; reviewing the comprehensive confidence weight set according to preset review requirements to obtain a review opinion set; merging the current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set to obtain the initial decision group; and constructing a fitness function based on the initial decision group.

[0057] Further, the iterative calculation module is used to obtain an initial evaluation particle swarm, iteratively calculate the initial evaluation particle swarm based on the fitness function and a pre-acquired compression factor, and obtain a final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement. This includes: obtaining the initial evaluation particle swarm, and obtaining the initial particle swarm position and initial particle swarm velocity based on the initial evaluation particle swarm; obtaining the optimal position of the particle swarm based on the initial particle swarm position; iteratively updating the initial particle swarm position and initial particle swarm velocity based on the pre-acquired compression factor and the optimal position of the particle swarm; calculating the average fitness of the initial evaluation particle swarm based on the fitness function in each iteration; and obtaining the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement.

[0058] Furthermore, the query set acquisition module is used to acquire current project information and construct a current query set based on the current project information, including: acquiring current project information and extracting features from the current project information to obtain a current project query term set; constructing the large language model based on the pre-acquired project knowledge base, and querying the current project query term set according to the large language model to obtain the current query set.

[0059] In this embodiment, the query set acquisition module forms a query set by acquiring data. The input confidence scoring module performs item association analysis on the current query set, the historical query set, and the semantic query set respectively to obtain the current confidence score, the historical confidence score, and the semantic confidence score, transforming the multi-level retrieval results into quantifiable indicators. A comprehensive confidence weight set is calculated and reviewed according to preset review requirements to obtain a review opinion set, establishing a collaborative interface between automated decision-making and human experience. Multi-source data is merged to obtain an initial decision group and a fitness function is constructed, realizing a mechanism leap from semantic matching to numerical evaluation. Furthermore, the iterative calculation module acquires the initial evaluation particle swarm and determines its position and velocity. Based on the fitness function and the pre-acquired compression factor, it iteratively updates the initial particle swarm position and velocity. Combining the average fitness of each iteration with real-time evaluation of the convergence state, it dynamically balances the optimization step size between global exploration and local development. This overcomes the shortcomings of traditional optimization methods, such as premature convergence and low accuracy, ensuring that the weight value confirmation process quickly and stably approaches the optimal solution, outputting a final evaluation particle swarm that accurately adapts to the actual business distribution. Through the synergistic linkage of the above modules, a complete processing chain from feature extraction and multi-source confidence quantization to adaptive weight optimization is constructed. This solves the technical problems of low matching accuracy and difficulty in updating with business rules caused by the lack of a dynamic fusion mechanism for multi-source information in the existing power grid processing architecture, thereby optimizing the power grid's data processing architecture and improving data processing efficiency.

[0060] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for processing power grid architecture data, characterized in that, include: Obtain current project information and construct the current query set based on the current project information; Obtain a historical knowledge graph, extract structural information related to the current project information from the historical knowledge graph, and construct a historical query set based on the structural information and the current project information; The current project information is expanded and enhanced based on a preset large language model to obtain a semantic query set; Confidence scores are applied to the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and a fitness function is constructed based on the initial decision group. An initial evaluation particle swarm is obtained, and the initial evaluation particle swarm is iteratively calculated based on the fitness function and the pre-acquired compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. The weight values ​​of the initial decision group are determined based on the final evaluation particle swarm, and the final decision group is obtained to complete the power grid architecture data processing process based on the final decision group.

2. The power grid architecture data processing method according to claim 1, characterized in that, The step of performing confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and constructing a fitness function based on the initial decision group, includes: Item association analysis is performed on the current query set, the historical query set, and the semantic query set respectively to obtain the current confidence score, the historical confidence score, and the semantic confidence score; A comprehensive confidence weight set is calculated based on the current confidence score, the historical confidence score, and the semantic confidence score; The comprehensive confidence weight set is reviewed according to the preset review requirements to obtain a review opinion set; The current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set are combined to obtain the initial decision group; A fitness function is constructed based on the initial decision set.

3. The power grid architecture data processing method according to claim 2, characterized in that, After determining the weight values ​​of the initial decision group based on the final evaluation particle swarm optimization to obtain the final decision group, and completing the power grid architecture data processing process based on the final decision group, the process includes: The comprehensive confidence weight set of the initial decision group is determined based on the final evaluation particle swarm, and the final confidence weight set is obtained. The current query set, the historical query set, the semantic query set, the final confidence weight set, and the review opinion set are combined to obtain the final decision group; A bypass matrix is ​​introduced and trained based on the final decision group. When the bypass matrix meets the preset matrix requirements, a reliable bypass matrix is ​​obtained. The large language model is updated based on the reliable bypass matrix to obtain the final large language model, thereby completing the iterative update process of the large language model.

4. The power grid architecture data processing method according to claim 1, characterized in that, The process of obtaining the initial evaluation particle swarm involves iteratively calculating the initial evaluation particle swarm based on the fitness function and a pre-acquired compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets a preset fitness requirement, the final evaluation particle swarm is obtained, including: Obtain an initial evaluation particle swarm, and based on the initial evaluation particle swarm, obtain the initial particle swarm position and initial particle swarm velocity; The optimal position of the particle swarm is obtained based on the initial particle swarm position; Based on the pre-acquired compression factor and the optimal position of the particle swarm, the initial particle swarm position and the initial particle swarm velocity are iteratively updated, and the average fitness of the initial evaluated particle swarm is calculated based on the fitness function in each iteration. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained.

5. The power grid architecture data processing method according to claim 1, characterized in that, The step of obtaining current project information and constructing a current query set based on the current project information includes: Obtain current project information and perform feature extraction on the current project information to obtain the current project query term set; The large language model is constructed based on the pre-acquired project knowledge base, and the current project query term set is queried according to the large language model to obtain the current query set.

6. The power grid architecture data processing method according to claim 5, characterized in that, The step of acquiring a historical knowledge graph, extracting structural information related to the current project information from the historical knowledge graph, and constructing a historical query set based on the structural information and the current project information includes: A historical knowledge graph is constructed based on pre-acquired historical power grid data; Based on the current project query term set, a graph traversal is performed in the historical knowledge graph to extract structural information; Based on the large language model, the structural information and the current item query term set are queried to obtain the historical query set.

7. The power grid architecture data processing method according to claim 5, characterized in that, The step of expanding and enhancing the current project information based on a preset large language model to obtain a semantic query set includes: Based on the large language model, the current project query term set is expanded to obtain an extended query term set. Enhanced queries are performed on the extended query term set based on the large language model to obtain a semantic query set.

8. A power grid architecture data processing system, characterized in that, include: The query set acquisition module is used to acquire current project information and construct the current query set based on the current project information; Obtain a historical knowledge graph, extract structural information related to the current project information from the historical knowledge graph, construct a historical query set based on the structural information and the current project information, and expand and enhance the current project information according to a preset large language model to obtain a semantic query set; The confidence scoring module is used to perform confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and to construct a fitness function based on the initial decision group; The iterative calculation module is used to obtain an initial evaluation particle swarm, and to iteratively calculate the initial evaluation particle swarm based on the fitness function and the pre-obtained compression factor. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained. The decision group confirmation module is used to determine the weight values ​​of the initial decision group based on the final evaluation particle swarm, and obtain the final decision group, so as to complete the power grid architecture data processing process based on the final decision group.

9. A power grid architecture data processing system according to claim 8, characterized in that, The confidence scoring module is used to perform confidence scoring on the current query set, the historical query set, and the semantic query set to obtain an initial decision group, and to construct a fitness function based on the initial decision group, including: Item association analysis is performed on the current query set, the historical query set, and the semantic query set respectively to obtain the current confidence score, the historical confidence score, and the semantic confidence score; A comprehensive confidence weight set is calculated based on the current confidence score, the historical confidence score, and the semantic confidence score; The comprehensive confidence weight set is reviewed according to the preset review requirements to obtain a review opinion set; The current query set, the historical query set, the semantic query set, the comprehensive confidence weight set, and the review opinion set are combined to obtain the initial decision group; A fitness function is constructed based on the initial decision set.

10. A power grid architecture data processing system according to claim 8, characterized in that, The iterative calculation module is used to obtain an initial evaluation particle swarm, iteratively calculate the initial evaluation particle swarm based on the fitness function and the pre-obtained compression factor, and obtain the final evaluation particle swarm when the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, including: Obtain an initial evaluation particle swarm, and based on the initial evaluation particle swarm, obtain the initial particle swarm position and initial particle swarm velocity; The optimal position of the particle swarm is obtained based on the initial particle swarm position; Based on the pre-acquired compression factor and the optimal position of the particle swarm, the initial particle swarm position and the initial particle swarm velocity are iteratively updated, and the average fitness of the initial evaluated particle swarm is calculated based on the fitness function in each iteration. When the maximum number of iterations is reached or the average fitness of the initial evaluation particle swarm meets the preset fitness requirement, the final evaluation particle swarm is obtained.

Images