Power distribution network engineering project document review system and method

Abstract

This application provides a system and method for reviewing project documents for power distribution network engineering projects. The review system includes: a multi-source parsing module configured to receive project documents in various formats for power distribution network engineering, perform optical character recognition, perform structural segmentation based on the project documents, and extract key fields and entity information; a rule base management module configured to establish a review rule base for power distribution network engineering projects; a knowledge graph engine configured to construct a knowledge graph for power distribution network engineering projects; a Guangming Power Big Model configured to perform text context-aware understanding and reasoning; and a review and feedback module configured to perform text checking and verification based on the multi-source parsing module, rule base management module, knowledge graph engine, and Guangming Power Big Model, and generate a structured review report. The power distribution network engineering project document review system and method provided in this application are simple and convenient, and can improve review efficiency, consistency, accuracy, and professionalism.

Description

Technical Field

[0001] This application relates to the field of power grid technology, and in particular to a system and method for reviewing project documents for power distribution networks. Background Technology

[0002] Distribution network engineering is a complex systems engineering project, encompassing not only the installation of various electrical equipment such as transformers, switchgear, and lines, but also the entire process from design and construction to commissioning and acceptance. These interconnected stages constitute the core content of distribution network engineering. Feasibility study review for distribution network projects is a routine task in the development and construction of power grids. Traditional review methods mainly rely on manual extraction of key indicators and item-by-item comparison with review rules. When faced with complex long text documents (such as Word, PDF, and Excel formats), these methods suffer from drawbacks such as redundant information interference, insufficient dynamic adaptation of professional rules, and low efficiency in compiling review opinions. Some technologies only partially achieve digital review, making it difficult to accurately extract cross-professional indicators, resulting in inconsistent review quality and failing to meet the needs of efficient and intelligent transformation of power grid engineering. Therefore, a more efficient review system and method are urgently needed. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a document review system and method for power distribution network engineering projects to solve the above-mentioned technical problems.

[0004] The first aspect of this application provides a power distribution network engineering project document review system, comprising: a multi-source parsing module configured to receive project documents in various formats for power distribution network engineering, perform optical character recognition, perform structural segmentation based on the project documents, and extract key fields and entity information; a rule base management module configured to establish a review rule base for power distribution network engineering, supporting rule creation, editing, version management, failure management, and retrieval; a knowledge graph engine configured to construct a knowledge graph for power distribution network engineering, supporting graph-based knowledge reasoning and querying; a Guangming Power Big Model configured to perform text context-aware understanding and reasoning, supporting knowledge-enhanced question answering and attribution analysis; and a review and feedback module configured to perform text checking and verification based on the multi-source parsing module, the rule base management module, the knowledge graph engine, and the Guangming Power Big Model, generating a structured review report.

[0005] Furthermore, the power distribution network project document review system also includes: a continuous learning and iteration module, configured to obtain review result correction information to optimize the review and feedback module, obtain new rules to update the rule base management module, obtain new knowledge information to update the knowledge graph engine, and adaptively evolve and iterate the project document review system through sandbox verification and gray release mechanisms; and an issue disclosure and approval interface, configured to distribute the review report and assign tasks, supporting issue tracking and rectification verification.

[0006] The second aspect of this application provides a method for reviewing project documents for power distribution networks. This method utilizes the power distribution network project document review system described in the first aspect above. The reviewer includes: using a multi-source parsing module to receive project documents in various formats for power distribution network engineering, performing optical character recognition (OCR) to extract key fields; matching the key fields with rules in a rule base management module and establishing a connection with a knowledge graph in a knowledge graph engine to generate text; using the Guangming Power Big Data Model for text understanding and reasoning; and using a review and feedback module to perform multi-dimensional checks on the text to identify problems and suggestions. After logical verification, a structured review report is generated.

[0007] Furthermore, the logical verification includes semantic consistency verification, logical rule verification, knowledge graph reasoning verification, and two-way verification of questions and suggestions.

[0008] Furthermore, the semantic consistency verification includes: encoding the questions and suggestions in the text into high-dimensional vectors based on the Guangming Power big data model, calculating the semantic relevance of the high-dimensional vectors using cosine similarity, and passing the semantic consistency verification when the semantic relevance is greater than or equal to a preset relevance.

[0009] Furthermore, the preset correlation degree is 0.75.

[0010] Furthermore, the logical rule verification includes: constructing causal relationship templates, conditional relationship templates, and parallel relationship templates based on the rule base management module; when the text suggestion matches the relationship template, it passes the logical rule verification.

[0011] Furthermore, the knowledge graph reasoning verification includes: constructing a triple association of questions, specifications, and suggestions based on the knowledge graph engine, capturing the association between nodes and edges in the knowledge graph, verifying whether the specifications and suggestions in the text match based on the association, and if they match, passing the knowledge graph reasoning verification.

[0012] Furthermore, the two-way verification of questions and suggestions includes: determining the number of matching suggestions from the questions in the text, determining the number of matching questions from the suggestions in the text, and calculating the traceability integrity index = (number of matching questions + number of matching suggestions) / (total number of questions + total number of suggestions). When the traceability integrity index is greater than or equal to a preset value, the two-way verification of questions and suggestions is passed.

[0013] Furthermore, the templates for the review report include a standardized document structure template, a standardized data field template, a problem description template, and a rectification suggestion template.

[0014] As can be seen from the above, this application provides a system and method for reviewing project documents of power distribution network engineering projects. The review system includes: a multi-source parsing module, configured to receive project documents of various formats of power distribution network engineering projects, perform optical character recognition, perform structural segmentation based on the project documents, and extract key fields and entity information; a rule base management module, configured to establish a review rule base for power distribution network engineering projects, supporting rule creation, editing, version management, failure management, and retrieval; a knowledge graph engine, configured to construct a knowledge graph for power distribution network engineering projects, supporting knowledge reasoning and querying based on the graph; a Guangming Power Big Model, configured to perform text context-aware understanding and reasoning, supporting knowledge-enhanced question answering and attribution analysis; and a review and feedback module, configured to perform text checking and verification based on the multi-source parsing module, rule base management module, knowledge graph engine, and Guangming Power Big Model, and generate a structured review report. This power distribution network project document review system and method improves review efficiency by identifying documents of various formats through a multi-source parsing module, enhances review accuracy and professionalism by using a rule base management module and a knowledge graph engine, increases text perception capabilities through the Guangming Power large-scale model, and improves review accuracy through inspection and verification by an examination and feedback module. Finally, it generates a structured review report, enhancing standardization. This system and method is simple and convenient, improving review efficiency, consistency, accuracy, and professionalism. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of a module of a power distribution network engineering project document review system in an embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the hierarchical interaction of a power distribution network engineering project document review system in an embodiment of this application.

[0018] Figure 3 This is a flowchart illustrating a method for reviewing project documents in a power distribution network project, as described in this application.

[0019] Figure 4 This is a schematic diagram illustrating the logical relationship of a document review method for power distribution network engineering projects in this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0021] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0022] In recent years, with the initial application of artificial intelligence technology in the power sector, some review assistance systems based on rule engines or simple natural language processing have emerged. These systems can automatically extract some structured data and perform simple rule matching, such as text comparison tools based on keyword retrieval and indicator reporting systems based on templates. Existing similar technical solutions mostly adopt a shallow digital review model of "document parsing + rule matching," typically extracting limited fields (such as project name, project time, etc.) based on fixed templates or regular expressions, and then performing simple comparisons with pre-set review items. These systems can only handle well-formatted tables or text paragraphs, lacking effective parsing capabilities for unstructured text, cross-page tables, and information embedded in images. Furthermore, the review rules are static and fixed, unable to adapt to the dynamic changes in different project types and regional regulations, and even more difficult to achieve in-depth semantic compliance review and intelligent opinion generation.

[0023] Specifically, traditional methods for reviewing feasibility studies of digital power distribution network projects have the following main problems:

[0024] First, the system has weak multi-format document parsing capabilities. Due to the lack of integration of OCR (Optical Character Recognition), multimodal parsing, and semantic enhancement technologies, the system struggles to accurately extract key indicators (such as construction scale and necessity arguments) from Word, PDF, Excel, and images, resulting in incomplete information extraction and large errors.

[0025] Second, the review rules are rigid and incomplete. Due to the reliance on a pre-built rule base and the lack of knowledge graph support, the system cannot dynamically adapt to various engineering review scenarios, making it difficult to achieve cross-professional and multi-dimensional compliance reviews (such as multiple sources of evidence, including technical standards, policy documents, and historical cases), resulting in many blind spots and insufficient professionalism in the review process.

[0026] Third, the review process is not highly intelligent. Due to the lack of large-scale model-driven semantic understanding and reasoning capabilities, the system can only perform surface keyword matching and cannot deeply understand the semantics, logical relationships, and implicit compliance issues of the text, resulting in limited review depth and a high rate of missed or incorrect reviews.

[0027] Fourth, the generation of opinions relies on templates and lacks flexibility. Due to the lack of NLP (Natural Language Processing) generation models and reinforcement learning mechanisms, review opinions are mostly based on fixed sentence concatenation, lacking specificity and logical coherence. They cannot adapt to different engineering characteristics and problem types, resulting in low practicality of opinions and a large amount of manual modification work.

[0028] Fifth, the system lacks continuous evolution capabilities. Due to the absence of a feedback loop and incremental learning mechanism, the system cannot learn and optimize from historical review cases, resulting in rigid model capabilities and difficulty in adapting to updates in review standards and expansion of project types.

[0029] To address the aforementioned shortcomings, this application aims to provide an intelligent review method for power distribution network engineering feasibility studies based on the fusion of large-scale models and knowledge graphs, achieving the following objectives: First, by constructing a multi-format parsing engine and an OCR enhancement module, it enables unified extraction of multi-modal data such as text, tables, and images, and structuring of key indicators, thereby improving the parsing accuracy and completeness of multi-source heterogeneous documents. Second, by integrating the semantic understanding capabilities of power industry knowledge graphs and large-scale models, it supports multi-level, cross-professional dynamic rule matching and deep semantic review. Third, based on NLP and reinforcement learning, it constructs an opinion generation model to automatically generate standardized, coherent, and targeted initial drafts of review opinions, and supports manual proofreading and optimization. Fourth, through incremental learning and feedback mechanisms, it enables continuous iteration of the model and knowledge base, improving system adaptability and review quality.

[0030] The following describes specific embodiments in conjunction with... Figures 1 to 4 The technical solution of this application will be described in detail below.

[0031] Some embodiments of this application provide a power distribution network engineering project document review system, such as... Figure 1 and Figure 2As shown, the system includes: a multi-source parsing module 11, configured to receive project documents of various formats for power distribution network engineering, perform optical character recognition, perform structural segmentation based on the project documents, and extract key fields and entity information; a rule base management module 12, configured to establish a review rule base for power distribution network engineering, supporting rule creation, editing, version management, failure management, and retrieval; a knowledge graph engine 13, configured to construct a knowledge graph for power distribution network engineering, supporting graph-based knowledge reasoning and querying; a Guangming Power Big Model 14, configured to perform text context-aware understanding and reasoning, supporting knowledge-enhanced question answering and attribution analysis; and a review and feedback module 15, configured to perform text checking and verification based on the multi-source parsing module, the rule base management module, the knowledge graph engine, and the Guangming Power Big Model, generating a structured review report.

[0032] The multi-source parsing module receives project document files in various formats, performs OCR recognition on scanned documents, realizes structured parsing and block processing of documents, and extracts key fields and entity information. The rule base management module establishes and maintains a multi-level rule base system for the feasibility study review of power distribution network projects, supports the creation, editing, version management and failure management of rules, realizes the correlation analysis and conflict detection between rules, and provides rule retrieval and visualization functions. The knowledge graph engine constructs a professional knowledge graph of power grid development, maps the concepts and relationships in the field of feasibility study review of power distribution network projects, realizes the semantic expression and association of rules, supports knowledge reasoning and query based on the graph, and provides knowledge visualization and interactive functions. The Guangming Power Big Model provides deep semantic understanding capabilities, realizes context-aware text understanding and reasoning, provides knowledge-enhanced intelligent question answering and attribution analysis, and supports domain-adaptive fine-tuning and continuous learning. The automatic review and feedback module realizes automatic inspection of feasibility study reports, accurately locates problems and related evidence, verifies them, generates structured review reports, and supports human-machine collaborative review processes.

[0033] The multi-source parsing module is capable of receiving project document files in various formats (PDF, Word, Excel, images, etc.). It can perform OCR recognition on scanned documents, realize structured parsing and block processing of documents, and extract key fields and entity information.

[0034] The multi-source parsing module's specific operations include: 1. Document parsing: Based on Apache Tika / POI technology, it supports parsing multiple document formats and converting them into a unified text format. It accurately identifies elements such as titles, body text, tables, and images based on document layout analysis technology, enabling structured extraction and parsing of tables. 2. Intelligent segmentation: It uses a combination of rules and learning to segment documents, employing a hierarchical segmentation strategy, first segmenting at the chapter level and then at the paragraph level. It also supports custom segmentation templates to adapt to the differentiated segmentation needs of power distribution network engineering feasibility study reports. 3. Field extraction: It uses named entity recognition technology to identify key entities in the report, such as project names, voltage levels, and line lengths, achieving relationship extraction and capturing the dependencies between entities. Based on a pre-trained extraction model, it accurately identifies power industry terminology and indicators. 4. Data standardization: It performs unified standardization processing on extracted fields, achieving unit conversion and standardization of numerical fields (ten thousand yuan, km, kVA, etc.), and supports the standardization of special fields such as time, location, and amount.

[0035] The rule base management module can establish and maintain a multi-level rule base system for the feasibility study review of power distribution network projects, support the creation, editing, version management and failure management of rules, realize the correlation analysis and conflict detection between rules, and provide rule retrieval and visualization display functions.

[0036] The rule base management module includes the following specific operations: 1. Rule storage: Utilizing a distributed database to store rule information, enabling categorized and hierarchical storage of rules, supporting multi-dimensional tag management (such as project attributes, engineering types, regional standards, etc.), and establishing rule indexes to effectively improve rule retrieval efficiency. 2. Rule entry and editing: Providing a visual rule editing interface for easy rule creation and modification, supporting automatic conversion from natural language rules to machine-parseable rules, and implementing rule template functionality to improve rule creation efficiency. 3. Rule version management: Implementing rule version control, recording rule change history in detail, supporting rule timeliness management, automatically handling expired rules, and providing rule change impact analysis functionality to help users understand the impact of rule changes on the system. 4. Rule conflict detection: Automatically detecting conflicts and contradictions in the rule set, implementing rule dependency analysis, and providing rule optimization suggestions to ensure the accuracy and effectiveness of the rule base.

[0037] Knowledge graph engines can construct professional knowledge graphs for power grid development, map concepts and relationships in the field of distribution network engineering feasibility study review, realize semantic expression and association of rules, support graph-based knowledge reasoning and querying, and provide knowledge visualization and interactive functions.

[0038] The specific operations of the knowledge graph engine can include: 1. Graph construction: A semi-automatic approach is used to build domain knowledge graphs, supporting automatic extraction of entities and relationships from rule bases and document libraries. This enables multi-source knowledge fusion and integrates internal and external knowledge resources such as the "Distribution Network Planning and Design Guidelines," technical standards, policy documents, and historical review cases. 2. Graph storage and indexing: Knowledge graphs are stored using a graph database (Neo4j), achieving an efficient graph indexing mechanism. This supports fast retrieval and querying, incremental updates, and historical version management, ensuring the timeliness and traceability of the knowledge graph. 3. Knowledge reasoning: A reasoning engine based on a combination of rule-based and statistical reasoning supports various reasoning methods such as transitive reasoning and inductive reasoning. It enables reasoning on uncertainties, handles fuzzy and incomplete knowledge, and enhances the intelligent reasoning capabilities of the knowledge graph. 4. Graph visualization: Multi-dimensional visualization of the knowledge graph is provided, supporting interactive graph exploration and querying. Knowledge path analysis and display are enabled, allowing users to intuitively understand the structure and relationships of the knowledge graph.

[0039] Guangming Power's large-scale model integrates multimodal data analysis capabilities, including text, images, and videos, with a parameter scale reaching hundreds of billions. It possesses functions such as power knowledge understanding, logical reasoning, numerical calculation, and content-assisted generation. It can provide deep semantic understanding capabilities, achieve context-aware text understanding and reasoning, provide knowledge-enhanced intelligent question answering and attribution analysis, and support domain-adaptive fine-tuning and continuous learning.

[0040] The specific operations of the Guangming Power large-scale model can include: 1. Model integration: Integrating multiple large language models, standardizing model interfaces, supporting seamless model switching, and adopting a model routing mechanism to select the most suitable model based on task characteristics, thereby improving the system's processing efficiency and accuracy. 2. Domain adaptation: Fine-tuning the model based on power industry data and review corpora, achieving efficient parameter fine-tuning (PEFT), reducing computational resource requirements, supporting continuous learning, and continuously optimizing model performance. 3. Knowledge enhancement: Achieving deep integration between the large-scale model and knowledge graphs, using Retrieval Enhanced Generation (RAG) technology to improve answer accuracy, supporting real-time retrieval and integration of external knowledge bases, and enhancing the system's knowledge reserves and application capabilities. 4. Reasoning explanation: Providing interpretable displays of the model's reasoning paths and evidence, achieving automatic generation and display of reasoning evidence chains, supporting reasoning correction mechanisms based on domain knowledge, and improving the transparency and reliability of model decisions.

[0041] The review and feedback module can automatically check the feasibility study report of power distribution network project from multiple dimensions, identify problems and suggestions, perform logical verification, generate a structured review report, and support a human-machine collaborative review process.

[0042] The review and feedback module can include the following specific operations: 1. Consistency check: This compares the consistency of fields and data within the report, supporting cross-document, cross-chapter, and cross-table data consistency verification. It also detects inconsistencies in expression based on semantic similarity analysis. 2. Compliance check: This uses a combination of rule matching and large-scale model reasoning for compliance judgment. It supports hierarchical checks of multi-level compliance standards and automatically locates and references compliance clauses. 3. Reasonableness check: This uses statistical analysis and knowledge reasoning for anomaly detection, verifying logical relationships, detecting the reasonableness of causal inferences, supporting industry benchmark comparison analysis, and identifying data deviating from the normal range. 4. Problem location and attribution: This accurately locates the position and scope of problems, automatically associates violating content with corresponding rule clauses, and generates a problem severity rating and priority suggestions. 5. Logical verification: This employs a "three-layer verification + two-way traceability" mechanism to ensure the completeness and accuracy of the logical connections between problems and suggestions in the review comments. 6. Review report generation: This automatically generates structured review reports, supports multiple output formats (Word, PDF, HTML, etc.), provides visual problem displays and statistical analysis, and facilitates user viewing and understanding of the review results.

[0043] This power distribution network project document review system is simple and convenient, improving review efficiency, consistency, accuracy, and professionalism. It enhances review efficiency by using a multi-source parsing module to identify documents of various formats, improves accuracy and professionalism by using a rule base management module and a knowledge graph engine to increase review criteria, enhances text perception capabilities by leveraging the Guangming Power big data model, and improves accuracy through inspection and verification by the review and feedback module. Finally, it generates structured review reports, improving standardization.

[0044] In some embodiments, such as Figure 1 As shown, the power distribution network project document review system further includes: a continuous learning and iteration module 16, configured to obtain review result correction information to optimize the review and feedback module, obtain new rules to update the rule base management module, obtain new knowledge information to update the knowledge graph engine, and adaptively evolve and iterate the project document review system through sandbox verification and gray release mechanisms; and an issue disclosure and approval interface 17, configured to distribute the review report and assign tasks, supporting issue tracking and rectification verification.

[0045] The continuous learning and iteration module is used to build a "human-machine feedback closed-loop" learning mechanism. It can collect data on expert corrections, confirmations, and new rule operations for review results, dynamically update entities and relationships in the knowledge graph using incremental learning technology, optimize large model strategies using parameter efficient fine-tuning (PEFT) and reinforcement learning human feedback (RLHF) techniques, and achieve adaptive evolution and version iteration of system capabilities through sandbox verification and canary release mechanisms. The problem disclosure and approval interface is used to connect with the internal workflow system of the power grid enterprise, support the automatic distribution of review results and task assignment, realize problem tracking and rectification verification, and provide API interfaces to support third-party system integration.

[0046] Among them, the continuous learning and iteration module can build a continuous learning mechanism of "human-machine feedback closed loop". By using the correction, confirmation and addition of rules by review experts on the system output results, it can automatically feed back to the large model and knowledge graph, realize the dynamic evolution and adaptive optimization of system capabilities, and solve the problems of frequent updates of power standards and insufficient coverage of long-tail scenarios.

[0047] The continuous learning and iteration module can include the following specific operations: 1. Multi-source feedback data collection: Automatically capture expert "adoption," "modification," and "rejection" records on the review interface regarding system-generated opinions; record manually added rule entries, revised rectification suggestions, and manually annotated error cases; track user dwell time in review reports, repeatedly reviewed clauses, and the degree of difference between the final draft and the initial draft, construct user preference vectors, and identify potential misjudgment patterns in the system. 2. Incremental knowledge graph updates: Based on newly added terms or newly emerging violation patterns corrected by experts, automatically extract new entities and relationships using few-sample learning technology. After confidence assessment (Confidence Score > 0.85), automatically integrate them into the existing knowledge graph without requiring a full reconstruction; when newly entered rules logically conflict with existing rules, automatically activate the conflict detection algorithm, combine historical execution success rates and expert weights, recommend retaining the optimal rule or generating a fusion rule, and mark it for manual review; establish timeline version control for the knowledge graph, supporting the retrospective of rule status at the time of project completion, ensuring the traceability and consistency of historical project review results. 3. Adaptive fine-tuning of the large model domain: The expert-corrected "problem-basis-suggestion" triplet is transformed into a high-quality instruction fine-tuning dataset, with weighted sampling specifically for long-tail complex cases and error-prone scenarios; Low-rank adapter (LoRA) technology is adopted to train only a small number of adaptation layer parameters on the basis of freezing the backbone parameters of the Guangming Power large model, so as to achieve low-cost and high-frequency model iteration and avoid catastrophic forgetting; A reward model is constructed, defining the expert's "adoption" behavior as a positive reward and "significant modification" as a negative reward, and optimizing the policy network of the large model through the PPO algorithm to gradually align its output style with the expert's habits. 4. Sandbox Validation and Canary Release: Before a model or rule goes live, regression tests are automatically run in a sandbox environment containing a set of typical historical errors to ensure that the new iteration does not reduce the accuracy of the original scenario (Accuracy Drop < 1%). Small-scale canary releases are supported by region, project type, or user group. The approval rate and manual intervention rate of the new version are monitored in real time, and full rollout is automatically implemented after the indicators meet the standards. The changes in key indicators before and after system iteration are visualized (e.g., automatic adoption rate improvement curve, false positive rate decline trend, new standard response time), quantifying the actual effectiveness of continuous learning.

[0048] The problem disclosure and approval interface can be connected to the internal workflow system (OA system) of power grid enterprises, supporting the automatic distribution of review results and task assignment, realizing problem tracking and rectification verification, and providing API interface to support third-party system integration.

[0049] The specific operations of the issue disclosure and approval interface can include: 1. Workflow integration: Supporting integration with mainstream workflow platforms (such as OA systems), enabling automatic triggering and status synchronization of approval processes, and providing customizable workflow configuration functions to meet the business process needs of different units. 2. Task distribution: Automatically assigning tasks based on issue type and responsibility division, supporting task priority settings and deadline management, and implementing task dependency management to ensure a reasonable rectification order. 3. Rectification tracking: Automatically tracking rectification progress and status changes, providing rectification evidence collection and verification functions, supporting automated verification of rectification effects, and ensuring that issues are effectively resolved. 4. Providing a standardized REST API interface, supporting security authentication mechanisms such as JWT, enabling interface version management and backward compatibility, and facilitating seamless integration with other systems.

[0050] In some embodiments, the system's data flow and processing workflow adopts a highly efficient pipelined processing architecture. The main processing flow includes: 1. Data acquisition and preprocessing stage: Receiving project documents in multiple formats, performing OCR recognition and document parsing, completing preliminary structuring and cleaning, and converting the raw report data into a format that the system can process. 2. Segmentation and field extraction stage: Performing intelligent segmentation based on the document structure, extracting key fields and entity information, achieving data standardization processing, and preparing for subsequent review work. 3. Rule matching and knowledge association stage: Matching the extracted fields with the rule base, establishing the association between fields and knowledge graphs, generating text, and providing basic data for multi-dimensional inspection. 4. Multi-dimensional inspection stage: Performing consistency checks, compliance checks, and reasonableness checks, summarizing the comprehensive inspection results, and comprehensively and deeply reviewing the report content. 5. Problem identification and location stage: Determining the problem type and severity, accurately locating the problem location, associating rule basis and knowledge support, and providing a clear direction for problem handling. 6. Logical verification: Adopting a "three-layer verification + two-way traceability" mechanism to ensure the completeness and accuracy of the logical association between problems and suggestions in the text. 7. Report Generation and Process Triggering Stage: Generate a structured review report, trigger the corresponding approval process, distribute rectification tasks and notices, and ensure that the review results are effectively processed and fed back.

[0051] In some embodiments, the system adopts a microservice architecture design, supporting flexible deployment and horizontal scaling: 1. Technology Stack Selection: The front-end uses the Vue.js framework, based on component-based design, to provide a good user interaction experience; the back-end uses the Spring Boot microservice framework to ensure system stability and scalability; the database uses a combination of relational database (MySQL) and graph database (Neo4j) to meet different data storage needs; the message queue uses RocketMQ to implement asynchronous processing, improving system processing efficiency; the search engine uses Elasticsearch to provide full-text search capabilities. 2. Deployment Method: Supports containerized deployment, based on Kubernetes orchestration, to achieve efficient resource management and scheduling; provides dual-mode deployment solutions for private cloud and public cloud to meet the security and cost needs of different enterprises; implements multi-tenant isolation design, supports SaaS service mode, and improves system resource utilization and service flexibility. 3. Extensibility Design: Based on interface design principles, it supports dynamic expansion of functional modules, implements a plug-in architecture, supports the integration of third-party functional components, provides open APIs, supports seamless integration with other systems, and facilitates continuous system upgrades and expansions. 4. Security considerations: Data transmission encryption technology is adopted to ensure the security of information during transmission; role-based access control (RBAC) is used to achieve fine-grained permission management; a complete audit log is established to record all operation behaviors, which facilitates security auditing and problem tracing; data is anonymized to protect sensitive information from being leaked.

[0052] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0053] Furthermore, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the apparatus may be shown in block diagram form. This is to prevent the embodiments of this application from being difficult to understand, and it also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully within the understanding of those skilled in the art). In setting forth specific details to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0054] A second aspect of this application provides a method for reviewing project documents for power distribution networks, using the power distribution network project document review system described in any of the above embodiments, such as... Figure 3 and Figure 4 As shown, the reviewers for the power distribution network project documents include: S1. Use the multi-source parsing module to receive project documents of various formats in the power distribution network project, perform optical character recognition, and extract key fields.

[0055] S2. Match the key fields with the rules in the rule base management module and establish a connection with the graph in the knowledge graph engine to generate text.

[0056] S3. Use the Guangming Power large model to understand and reason about the text, and use the review and feedback module to conduct multi-dimensional checks on the text to identify problems and suggestions. After logical verification, generate a structured review report.

[0057] Data acquisition and preprocessing: Receives project documents in multiple formats, performs OCR recognition and document parsing, and completes preliminary structuring and cleaning; Blocking and field extraction: Intelligently blocks documents based on their structure, extracting key fields and entity information to achieve standardized data processing; Rule matching and knowledge association: Matches extracted fields with a rule base, establishing associations between fields and the knowledge graph, and generating text; Problem identification and location: Determines problem type and severity, accurately locates problem positions, and associates rule bases and knowledge support; Multi-dimensional inspection: Generates structured review reports; Triggers corresponding approval processes, distributing rectification tasks and notifications. Continuous learning and system iteration: Builds a high-quality instruction set based on human-machine feedback data, driving incremental updates of the knowledge graph and adaptive fine-tuning of the large model. After sandbox validation and gray-scale release, completes system version iteration, forming a positive evolutionary closed loop of "data-model-application".

[0058] The complete process of this power distribution network project document review method involves, for example, acquiring and identifying PDF and Word format project documents of a power station project, extracting key fields such as "line length," and then matching them with relevant rules and diagrams, such as matching "power grid specifications and rules for a certain region" and "power distribution network planning and design diagrams" to form text. The Guangming Power large model is used to assist in understanding and reasoning about the text. The review and feedback module is used to conduct multi-dimensional checks on the text, checking compliance, rationality, etc., identifying problems in the text and generating corresponding suggestions. For example, if the problem is "the line length is too long," the suggestion is "shorten the line length at a certain point," and the specification is "power grid specifications and rules for a certain region." Then, logical verification is performed on the problems and suggestions, and finally, a review report is generated based on a preset template.

[0059] This document review method for power distribution network projects significantly improves review efficiency. Compared to traditional manual review, it increases automation by over 85% and reduces review time by 75%, greatly enhancing the efficiency of power distribution network feasibility study reviews and shortening the review cycle. It also enhances accuracy, achieving an overall accuracy rate of over 90%, significantly higher than pure rule-based or pure model-based solutions, effectively avoiding compliance risks caused by inaccurate reviews. Furthermore, it reduces labor costs, decreasing manual review workload by over 55%, substantially lowering enterprise operating costs and allowing enterprises to invest human resources in more valuable work. It also enhances compliance, increasing the problem detection rate by 35%, enabling more comprehensive identification of compliance issues in feasibility study reports, significantly reducing compliance risks and ensuring the legal and compliant operation of enterprises. Finally, it improves interpretability, increasing model decision transparency by 70% through knowledge graph enhancement, making review results more persuasive and easier for users to understand and accept review conclusions. Finally, it supports rapid adaptation to new regulations and standard changes, shortening the rule update cycle by 90%, ensuring timely updates to regulations and industry standards and maintaining the system's effectiveness and usability. It can perform multi-dimensional checks on consistency, compliance, and reasonableness, enabling a comprehensive and in-depth review of report content and the identification of different types of issues. Logical consistency checks cover over 95% of review comments, ensuring a complete and accurate logical connection between issues and recommendations.

[0060] In some embodiments, the logical verification includes semantic consistency verification, logical rule verification, knowledge graph reasoning verification, and bidirectional verification of questions and suggestions.

[0061] The logical verification adopts a "three-layer verification + two-way traceability" mechanism to ensure that the logical connection between the questions and suggestions in the text is complete and accurate.

[0062] The semantic consistency check includes: encoding the questions and suggestions in the text into high-dimensional vectors based on the Guangming Power big data model, calculating the semantic relevance of the high-dimensional vectors using cosine similarity, and passing the semantic consistency check when the semantic relevance is greater than or equal to a preset relevance.

[0063] Based on the semantic vector embedding technology of Guangming Power's large model, the problem description and suggested measures in the text are encoded into high-dimensional vectors respectively; the semantic correlation between the problem and the suggestion is calculated by using cosine similarity, and a preset correlation θ=0.75 is set. If the correlation is lower than the preset correlation, a consistency alarm is triggered; a problem-suggestion mapping matrix M(n×m) is established, where n is the number of problems, m is the number of suggestions, and the matrix elements represent the correlation strength.

[0064] The preset relevance level is 0.65-0.85. This application conducted tests with different preset relevance levels. When the preset relevance level is 0.65, the matching rate between questions and suggestions is approximately 73%; when the preset relevance level is 0.7, the matching rate is approximately 86%; when the preset relevance level is 0.75, the matching rate is approximately 95%; when the preset relevance level is 0.8, the matching rate is approximately 84%; and when the preset relevance level is 0.85, the matching rate is approximately 69%. The overall change shows a parabolic trend. The preferred preset relevance level is 0.75; both excessively high and low preset relevance levels will reduce the matching rate between questions and suggestions.

[0065] Logical rule verification includes: constructing causal relationship templates, conditional relationship templates, and parallel relationship templates based on the rule base management module. When the suggestion in the text matches the relationship template, it passes the logical rule verification.

[0066] A logical rule base for the feasibility study review of power distribution network projects is constructed, containing three types of logical relationship templates: "causal," "conditional," and "parallel." Causal relationship templates indicate that if a certain result suggestion exists, a corresponding cause must also exist. For example, if "the cable cross-section selection is based on short-circuit thermal stability verification calculation," then the suggestion "short-circuit current calculation process and results" must be included. Conditional relationship templates indicate that if a certain condition suggestion exists, a corresponding comparison process must also exist. For example, if "the length of the new line is >10km," then the content of "route scheme comparison" must be included. Parallel relationship templates indicate that if a certain specified suggestion exists, another related suggestion must also exist. For example, if the review opinion includes "replacing old distribution transformers," then both "low-voltage distribution area wire diameter verification" and "reactive power compensation configuration optimization" must be included (these are necessary parallel items, neither can be omitted). A rule engine is used to perform logical rule validation on the generated text.

[0067] The knowledge graph reasoning verification includes: constructing a triple association of questions, specifications, and suggestions based on the knowledge graph engine, capturing the association between nodes and edges in the knowledge graph, verifying whether the specifications and suggestions in the text match based on the association, and if they match, passing the knowledge graph reasoning verification.

[0068] Based on the professional knowledge graph of power grid development, a triplet association of "problem type - regulatory clause - rectification measures" is established; multi-hop reasoning is carried out using graph neural network (GNN) to capture the deep association between nodes (such as power compliance indicators and feasibility study review points) and edges (such as indicator association and compliance constraint relationship) in the knowledge graph, and to explore hidden cross-dimensional compliance logic and verify the matching degree between the recommended measures and regulatory clauses; the reasoning path visualization is supported and the interpretability proof of the review basis is provided.

[0069] The two-way verification of issues and suggestions includes: determining the number of matching suggestions from the issues in the text, determining the number of matching issues from the suggestions in the text, and calculating the traceability integrity index = (number of matching issues + number of matching suggestions) / (total number of issues + total number of suggestions). When the traceability integrity index is greater than or equal to the preset value, the two-way verification of issues and suggestions is passed.

[0070] Forward tracing starts from the problem and verifies whether each problem has a corresponding rectification suggestion; backward tracing starts from the suggestion and verifies whether each suggestion has a corresponding problem basis; establish a traceability integrity index CI = (number of matching problems + number of matching suggestions) / (total number of problems + total number of suggestions), requiring CI ≥ 0.95.

[0071] In some embodiments, the template for the review report includes a standardized document structure template, a standardized data field template, a problem description template, and a rectification suggestion template.

[0072] To address the unique characteristics of feasibility study reports in the power industry, a multi-level standardized template system was developed to ensure consistent and standardized output of review reports.

[0073] Standardized document structure template: Defines a four-level structure specification of "chapter-section-article-clause" that conforms to the DL / T 5438-2020 power industry standard; specifies format parameters such as title font size (small two / black), body font size (small four / Song), and line spacing (1.5 times); and unifies the format of 12 types of standard tables such as investment estimation tables, equipment list tables, and technical specification tables.

[0074] Standardized templates for data fields: Numerical fields use a unified unit system (ten thousand yuan, km, kVA, etc.) and support automatic unit conversion and normalization; Time fields adopt the "YYYY-MM-DD" standard format and support the standardization of fields such as construction period and commissioning time; Enumerated fields establish standard enumerated values ​​for project attributes (heavy overload / optimized grid structure / high line loss management, etc.) and project types (line / transformer area / business expansion).

[0075] The problem description template adopts the format of "[Problem Type] + [Specific Location] + [Violation Content] + [Standard Basis]", for example: "Investment Estimation Problem - Chapter 3, Section 2 - Total Investment Exceeds Budget by 15% - Violation of Article 5.3 of the 'Guidelines for Distribution Network Planning and Design'".

[0076] The rectification suggestion template adopts the format of "[Suggestion Type] + [Rectification Measures] + [Completion Deadline] + [Responsible Entity]", for example: "Economic Supplementary Suggestion - Supplementary Multi-Solution Comparison Analysis - Within 5 Working Days After Review - Design Unit"; supports output in multiple formats such as Word, PDF, and HTML, and maintains format consistency.

[0077] The review and feedback module automatically loads the corresponding template set based on the project type (new line construction / line renovation / business expansion support, etc.); it supports adaptation to regional standard differences (such as distribution network construction standards in different provinces); it provides a visual template editing interface and supports user-defined template extensions. The document structure template is standardized, conforming to the DL / T 5438-2020 power industry standard; the data field template is standardized, unifying the formats of numerical, time, and enumeration fields; the review opinion output template includes problem description templates and rectification suggestion templates; and a dynamic template adaptation mechanism supports adaptation to differences in project type and regional standards.

[0078] It is understood that before using the technical solutions of the various embodiments in this disclosure, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.

[0079] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media performing the operations of this disclosed technical solution.

[0080] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0081] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0082] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description.

[0083] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A power distribution grid engineering project document review apparatus, characterized by, include: The multi-source parsing module is configured to receive project documents of various formats of power distribution network projects, perform optical character recognition, perform structural segmentation based on the project documents, and extract key fields and entity information; The rule base management module is configured to establish a review rule base for power distribution network projects, supporting the creation, editing, version management, expiration management, and retrieval of rules; The knowledge graph engine is configured to build a knowledge graph for power distribution network engineering, supporting graph-based knowledge reasoning and querying. The Guangming Power large model is configured to perform context-aware understanding and reasoning of text, and support knowledge-enhanced question answering and attribution analysis; The review and feedback module is configured to perform text checks and verifications based on the multi-source parsing module, the rule base management module, the knowledge graph engine, and the Guangming Power big data model, and generate a structured review report.

2. The electric power distribution network engineering project documentation review apparatus of claim 1, wherein, Also includes: The continuous learning and iteration module is configured to obtain review results and correct information to optimize the review and feedback module, obtain new rules to update the rule base management module, obtain new knowledge information to update the knowledge graph engine, and adaptively evolve and iterate the project document review system through sandbox verification and canary release mechanisms. The problem disclosure and approval interface is configured to distribute the review report and assign tasks, and supports problem tracking and rectification verification.

3. A power distribution network engineering project document review method, characterized by, Using the power distribution network project document review system according to claim 1 or 2, the power distribution network project document reviewer includes: The multi-source parsing module is used to receive project documents of various formats from power distribution network projects, perform optical character recognition, and extract key fields. The key fields are matched with the rules in the rule base management module and associated with the graph in the knowledge graph engine to generate text. The Guangming Power large model is used for text understanding and reasoning, and the review and feedback module is used to conduct multi-dimensional checks on the text to identify problems and suggestions. After logical verification, a structured review report is generated.

4. The electrical distribution network engineering project documentation review method of claim 3, wherein, The logical verification includes semantic consistency verification, logical rule verification, knowledge graph reasoning verification, and two-way verification of questions and suggestions.

5. The electrical distribution network engineering project documentation review method of claim 4, wherein, The semantic consistency verification includes: encoding the questions and suggestions in the text into high-dimensional vectors based on the Guangming Power big data model, calculating the semantic relevance of the high-dimensional vectors using cosine similarity, and passing the semantic consistency verification when the semantic relevance is greater than or equal to a preset relevance.

6. The electrical distribution network engineering project documentation review method of claim 5, wherein, The preset correlation degree is 0.

75.

7. The electrical distribution network engineering project documentation review method of claim 4, wherein, The logical rule verification includes: constructing causal relationship templates, conditional relationship templates, and parallel relationship templates based on the rule base management module; when the text suggestion matches the relationship template, it passes the logical rule verification.

8. The electrical distribution network engineering project documentation review method of claim 4, wherein, The knowledge graph reasoning verification includes: constructing a triple association of questions, specifications, and suggestions based on the knowledge graph engine, capturing the association between nodes and edges in the knowledge graph, verifying whether the specifications and suggestions in the text match based on the association, and if they match, passing the knowledge graph reasoning verification.

9. The method for reviewing project documents for power distribution networks according to claim 4, characterized in that, The two-way verification of questions and suggestions includes: determining the number of matching suggestions from the questions in the text, determining the number of matching questions from the suggestions in the text, and calculating the traceability integrity index = (number of matching questions + number of matching suggestions) / (total number of questions + total number of suggestions). When the traceability integrity index is greater than or equal to a preset value, the two-way verification of questions and suggestions is passed.

10. The method for reviewing project documents for power distribution networks according to claim 3, characterized in that, The templates for the review reports include standardized document structure templates, standardized data field templates, problem description templates, and rectification suggestion templates.

Images