Plant operation and maintenance information decision generation method and system, and storage medium

Abstract

The present application relates to the technical field of factory operation and maintenance management, and particularly relates to a factory operation and maintenance information decision generation method and system. The method comprises the following steps: constructing a factory configuration file library comprising a factory facility configuration file database; obtaining change information of a change management pre-process flow and preprocessing the change information to generate change element descriptions; training an intelligent recognition inference model to obtain a required intelligent recognition inference model, and setting the required intelligent recognition inference model in a file influence domain intelligent recognition engine; identifying a file influence domain based on the change element descriptions and in combination with the factory configuration file library to obtain an initial recommendation result; and generating a decision recommendation result based on the initial recommendation result, in combination with configuration change rules and manual review and decision. The beneficial effect is that the method helps to quickly determine the file influence range corresponding to the configuration change, automatically generates operation and maintenance information decisions adapted to the current scenario, and significantly improves the update speed and landing efficiency of the operation and maintenance specifications after the configuration change.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of factory operation and maintenance management and artificial intelligence, and in particular to a method, system and storage medium for generating factory operation and maintenance information decisions. Background Technology

[0002] The safe and stable operation of a factory relies on standardized facility configuration management. With accumulated operating time and technological updates, factory equipment will require maintenance, repair, and upgrades, leading to changes in facility configuration. According to relevant safety standards, any facility configuration change during operation necessitates updating the operation and maintenance (O&M) specifications; this is a crucial step in ensuring safe and compliant operation. Currently, factories primarily rely on manual experience-based judgment, matching pre-set rules, or a complete overhaul of O&M specifications when facility configuration changes occur.

[0003] However, factory systems are highly complex and subject to various change scenarios. Facility configuration changes involve both core safety-level configurations and non-safety-level auxiliary facility configurations, while also requiring strict safety regulatory traceability requirements. Existing technologies are difficult to adapt to these special needs, resulting in a relatively slow update speed of operation and maintenance specifications after facility configuration changes. In particular, when adjusting management methods based on the problems that arise, it is often difficult to immediately form new operation and maintenance specifications suitable for the current situation. Summary of the Invention

[0004] Based on this, and to address the aforementioned issues, this invention provides a method, system, and storage medium for generating factory operation and maintenance information decisions. By constructing a factory configuration file library, acquiring and preprocessing change information to generate change element descriptions, training and deploying an intelligent recognition and reasoning model, identifying the file impact domain based on the change element descriptions to obtain initial recommendation results, and combining configuration change rules with manual review to generate decision recommendation results, a complete operation and maintenance specification update scheme is formed, encompassing file library construction, change information processing, intelligent recognition, and decision output. This approach helps to quickly determine the file impact scope corresponding to configuration changes, automatically generate operation and maintenance information decisions adapted to the current scenario, significantly improve the update speed and implementation efficiency of operation and maintenance specifications after configuration changes, effectively reduce reliance on manual labor and experience bias, and quickly form new operation and maintenance specifications that fit the actual situation, thereby ensuring the compliance, timeliness, and stability of factory operation and maintenance management.

[0005] This invention is implemented as follows: a factory operation and maintenance information decision generation method, comprising: Build a plant configuration file library that includes a database of plant facility configuration files; Obtain change information from the preceding change management processes and preprocess it to generate change element descriptions; The intelligent recognition reasoning model is trained to obtain the required intelligent recognition reasoning model, and the required intelligent recognition reasoning model is set in the document influence domain intelligent recognition engine; Based on the description of the change elements, the file influence domain is identified using the factory configuration file library to obtain initial recommendation results; Based on the initial recommendation results, combined with configuration change rules and manual review and decision-making, a decision recommendation result is generated.

[0006] The factory configuration file library includes file storage, file retrieval, and retrieval enhancement generation functions.

[0007] The enhanced search generation function includes: The files in the factory configuration file library are transformed into vectors to build a file vector knowledge base; Based on the changed elements, a large model is used to semantically expand the content of the retrieved files and generate an impact description related to the changed elements. An initial prompt word vector is constructed based on the description of the changed elements and the description of the impact. Similarity matching is then performed using the file vector knowledge base to amplify the semantics of the prompt words. By drawing upon professional technical data and combining it with the amplified prompt word vectors, implicit cross-professional associations can be mined to supplement the association information not covered by explicit retrieval.

[0008] The process of training the intelligent recognition reasoning model to obtain the desired intelligent recognition reasoning model includes: The model is trained based on the factory facility configuration file database to obtain an initial intelligent recognition and reasoning model; The initial intelligent recognition reasoning model is tested and verified to obtain the desired intelligent recognition reasoning model.

[0009] The step of identifying the file impact domain based on the description of the change elements and in conjunction with the factory configuration file library to obtain the initial recommendation result includes: Based on the changed elements and their impact descriptions, a change feature vector is generated; The similarity between the changed feature vector and the file feature vector in the file vector knowledge base is calculated to obtain a set of explicitly associated files; Starting with the aforementioned set of explicit related files, path reasoning is performed based on the factory facility configuration knowledge graph to mine cross-professional implicit related files; The explicit and implicit related files are input into the LLM to perform generative reasoning, which parses the professional relationship logic between the files and changes and marks the degree of impact. Integrate explicit and implicit related documents and their degree of influence to obtain initial recommendation results.

[0010] The method further includes: The decision recommendation results are used as labeled data and matched with the descriptions of the change elements and the descriptions of the impacts to form training samples. Based on the training samples, a small-sample fine-tuning technique is used to update the semantic association weights between the changed parameters and the associated files; The factory facility configuration knowledge graph is updated synchronously, and exclusive association nodes for the change subject and related files are added; The optimized semantic association weights and the updated knowledge graph are input into the required intelligent recognition and reasoning model.

[0011] The description of the change elements includes the subject of the change, the type of change, the safety level, the change parameters, and related information. The type of change includes parameter modification, structural adjustment, and functional upgrade. The subject of the change is a factory system, equipment, or component.

[0012] The configuration change rules include time rules, scope rules, classification rules, and responsibility attribution rules; the time rules are the revision time requirements for different types of documents; the scope rules are the document scope logic affected by the change.

[0013] The factory operation and maintenance information decision generation system includes: The plant configuration file library contains a database of plant facility configuration files, as well as file storage, file retrieval, and enhanced generation capabilities. The change information interface and preprocessing module are used to obtain change information from the preceding change management process and preprocess it to generate a description of the change elements. The model training and deployment module is used to train the intelligent recognition reasoning model to obtain the required intelligent recognition reasoning model, and to set the required intelligent recognition reasoning model in the file influence domain intelligent recognition engine; The intelligent recognition engine is used to identify the file influence domain based on the description of the change elements and the factory configuration file library, so as to obtain an initial recommendation result; The decision-making module is used to generate decision-making recommendation results based on the initial recommendation results, combined with configuration change rules and manual review and decision-making.

[0014] A computer-readable storage medium storing computer program instructions that, when executed by a processor, perform steps in a method.

[0015] The advantages of this invention are as follows: The factory operation and maintenance information decision generation method, system, and storage medium of this invention, by constructing a factory configuration file library, acquiring and preprocessing change information to generate change element descriptions, training and deploying an intelligent recognition and reasoning model, identifying the file influence domain based on the change element descriptions to obtain initial recommendation results, and combining configuration change rules with manual review to generate decision recommendation results, form a full-process operation and maintenance specification update scheme from file library construction, change information processing, intelligent recognition to decision output. This approach helps to quickly determine the file influence scope corresponding to configuration changes, automatically generate operation and maintenance information decisions adapted to the current scenario, significantly improve the update speed and implementation efficiency of operation and maintenance specifications after configuration changes, effectively reduce reliance on manual labor and experience bias, and quickly form new operation and maintenance specifications that fit the actual situation, thereby ensuring the compliance, timeliness, and stability of factory operation and maintenance management. Attached Figure Description

[0016] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. For example, the flowcharts and block diagrams in the drawings illustrate the architecture, functions, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention.

[0017] Figure 1 This is a flowchart of the factory operation and maintenance information decision generation method provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the structure of the factory operation and maintenance information decision generation method system provided in the embodiments of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] A method for generating factory operation and maintenance information decisions includes: constructing a factory configuration file library containing a database of factory facility configuration files; obtaining and preprocessing change information from the change management pre-process to generate change element descriptions; training an intelligent recognition and reasoning model to obtain the required intelligent recognition and reasoning model, and setting the required intelligent recognition and reasoning model in a file influence domain intelligent recognition engine; identifying file influence domains based on change element descriptions and the factory configuration file library to obtain initial recommendation results; and generating decision recommendation results based on the initial recommendation results, combined with configuration change rules and manual review and decision-making.

[0020] The factory configuration file library includes file storage, file retrieval, and retrieval enhancement generation functions, enabling unified management, rapid retrieval, and deep semantic enhancement of factory configuration files. This effectively improves the completeness, accuracy, and efficiency of file influence domain identification, reduces the risk of missed detections and misjudgments, and better ensures the security and compliance of factory configuration changes.

[0021] The enhanced retrieval generation function includes: vectorizing files in the factory configuration file library to build a file vector knowledge base; semantically expanding the content of retrieved files based on change elements using a large model to generate impact descriptions related to the change element descriptions; constructing initial prompt word vectors based on the change element descriptions and impact descriptions, and performing similarity matching with the file vector knowledge base to amplify the semantics of the prompt words; and calling upon professional technical materials and combining them with the amplified prompt word vectors to mine cross-professional implicit associations to supplement the association information not covered by explicit retrieval. Through this enhanced retrieval generation function, file vectorization, semantic expansion, prompt word amplification, and cross-professional association mining can be achieved, significantly improving the analytical depth of the relationship between changes and files, effectively compensating for the shortcomings of traditional explicit retrieval, greatly reducing the false negative rate of cross-professional implicit files, and improving the comprehensiveness, accuracy, and intelligence level of file impact domain identification.

[0022] The intelligent recognition and reasoning model is trained to obtain the required intelligent recognition and reasoning model, including: training the model based on the factory facility configuration file database to obtain an initial intelligent recognition and reasoning model; and testing and verifying the initial intelligent recognition and reasoning model to obtain the required intelligent recognition and reasoning model. By conducting model training, testing, and reasoning verification based on the factory facility configuration file database, an intelligent recognition and reasoning model that fits the factory's professional scenario is constructed, improving the model's accuracy and stability in recognizing changes and document relationships.

[0023] Based on the description of change elements, and combined with the factory configuration file library, the impact domain of documents is identified to obtain initial recommendation results. This includes: generating change feature vectors based on change elements and impact descriptions; calculating the similarity between the change feature vectors and document feature vectors in the document vector knowledge base to obtain a set of explicitly related documents; starting from the set of explicitly related documents, performing path reasoning based on the factory facility configuration knowledge graph to mine cross-professional implicit related documents; inputting the set of explicitly related documents and implicit related documents into LLM to perform generative reasoning, parsing the professional association logic between documents and changes, and labeling the degree of impact; integrating the explicitly related documents, implicit related documents, and degree of impact to obtain the initial recommendation results. Through layer-by-layer identification using vector similarity calculation, knowledge graph path reasoning, and LLM generative reasoning, explicitly related documents can be completely obtained and cross-professional implicit related documents can be deeply mined. The professional association logic can be accurately parsed and the degree of impact can be clearly defined, significantly improving the completeness, accuracy, and professionalism of the initial recommendation results, effectively reducing missed detections and misjudgments, and providing a reliable basis for subsequent decision-making.

[0024] The decision recommendation results are used as labeled data and matched with the descriptions of change elements and their impacts to form training samples. Based on these training samples, a small-sample fine-tuning technique is employed to update the semantic association weights between change parameters and associated files. Simultaneously, the factory facility configuration knowledge graph is updated, adding dedicated association nodes between the change subject and associated files. The optimized semantic association weights and the updated knowledge graph are then input into the required intelligent recognition and reasoning model. By using the decision recommendation results as labeled data to form training samples, employing small-sample fine-tuning to update semantic association weights, and simultaneously improving the knowledge graph, the intelligent recognition and reasoning model can be continuously optimized. This continuously improves the matching accuracy between changes and associated files and the ability to uncover implicit associations, achieving closed-loop self-optimization of the recognition system. This makes the identification of the impact domain of subsequent similar changes more accurate and efficient.

[0025] The change element description includes the change subject, change type, safety level, change parameters, and related information. The change type includes parameter modification, structural adjustment, and functional upgrade. The change subject is the factory system, equipment, or component. It can provide standardized and structured input for the subsequent identification process based on the change element description and its impact description, accurately reflecting the core change information of the factory system, equipment, or component in scenarios such as parameter modification, structural adjustment, and functional upgrade. It effectively improves the accuracy and consistency of subsequent semantic expansion, vector generation, association matching, and inference identification, and reduces identification bias caused by missing elements or non-standard descriptions.

[0026] Configuration change rules include time rules, scope rules, classification rules, and responsibility attribution rules. Time rules specify the revision time requirements for different types of documents; scope rules define the document scope logic affected by the change. By setting configuration change rules that include time rules, scope rules, classification rules, and responsibility attribution rules, a unified and compliant basis for judgment can be provided for document impact domain decisions, standardizing the document revision sequence, clarifying impact boundaries and responsible parties, ensuring that decision results are rigorous, consistent, traceable, and verifiable, and fully meeting the safety and compliance requirements of factory change management.

[0027] The method also includes visualizing the decision recommendation results and connecting them with subsequent business processes and related systems of factory change management, enabling professional users to quickly and accurately grasp the impact of documents, hierarchical relationships and core revision points, significantly reducing the time cost and decision-making difficulty of manual review, and enhancing the overall technical solution's implementation efficiency and user adaptability.

[0028] Preprocessing includes denoising, format standardization, and entity standardization of change information. Entity standardization involves using the corresponding professional terminology and thesaurus for the factory. Denoising the change information removes redundant and invalid data, preventing interference with subsequent identification and reasoning processes and ensuring data purity. Format standardization addresses the issues of diverse change information sources and inconsistent expression formats, ensuring the consistency and standardization of change data and providing a unified input standard for subsequent processes. Entity standardization uses the corresponding professional terminology and thesaurus for the factory, which normalizes diverse expressions of professional terms in the corresponding field (such as safety level, equipment type, etc.), accurately extracts core elements such as the change subject and safety level, and avoids identification errors caused by terminological ambiguity.

[0029] The enhanced retrieval generation includes using a large model to semantically expand the content of the retrieved documents to generate impact descriptions related to the descriptions of change elements; deeply mining implicit associations in factory change scenarios to overcome the limitation of traditional keyword retrieval that can only match explicit associations; and accurately parsing professional terms and document semantic logic to upgrade the association between change elements and document content from superficial matching to in-depth interpretation. The generated impact descriptions provide richer decision-making basis for intelligent recognition and reasoning, effectively reducing the risk of missing cross-professional documents.

[0030] Manual review and decision-making involve human review of the completeness, accuracy, and granularity of the initial recommendation results to adjust their scope and priority. This process, conducted by professionals, aims to accurately identify potentially overlooked cross-disciplinary files or incorrectly included unrelated files by the intelligent identification system, overcoming the limitations of purely technical identification. Furthermore, by adjusting the scope and priority of the results, prioritizing core security-level documents, the process ensures that decisions align with the factory's actual operational needs and key security management priorities, further reducing the rates of missed detections and false positives.

[0031] The visualization includes using heatmaps to present the impact of documents and a tree structure to show the hierarchical relationship between documents. The heatmaps visually represent the impact of documents, enabling professional users to quickly identify highly relevant core documents, clearly define identification priorities, and avoid random filtering among massive amounts of documents. The tree structure presents the hierarchical relationship between documents from "security-level documents to non-security-level documents," precisely meeting the specific needs of factory document hierarchical management and clearly outlining the document association logic and classification. The collaboration of these two methods significantly reduces the information comprehension cost during manual review, improves decision-making efficiency, helps users quickly focus on core security-related documents, and enhances the relevance and readability of the identification results.

[0032] A factory operation and maintenance information decision generation system includes: a factory configuration file library, containing a database of factory facility configuration files and file storage, retrieval, and enhanced retrieval generation functions; a change information interface and preprocessing module, used to obtain change information from the change management pre-process and preprocess it to generate change element descriptions; a model training and deployment module, used to train an intelligent recognition and reasoning model to obtain the required intelligent recognition and reasoning model, and set the required intelligent recognition and reasoning model in the file influence domain intelligent recognition engine; an intelligent recognition engine, used to identify the file influence domain based on the change element descriptions and the factory configuration file library to obtain initial recommendation results; and a decision module, used to generate decision recommendation results based on the initial recommendation results, combined with configuration change rules and manual review and decision-making.

[0033] The change information interface and preprocessing module includes a change information interface unit and a preprocessing unit. The change information interface unit connects to the preceding change management process to obtain change information, avoiding the tediousness and errors of manual entry and significantly improving the efficiency of change information retrieval. The preprocessing unit cleans and standardizes the change information, extracts change element descriptions, uses a professional thesaurus for entity standardization, and combines cleaning and format unification to solve the problems of inconsistent change information formats and diverse professional terminology, ensuring the standardization and accuracy of change element extraction. The change element description includes the change subject, change type, security level, change parameters, and related data. The change type includes parameter modification, structural adjustment, and functional upgrade. The change subject is the factory system, equipment, or component. The module comprehensively extracts core elements such as the change subject (factory system, equipment, or component), change type (parameter modification, structural adjustment, functional upgrade), security level, and change parameters, providing high-quality structured data support for the subsequent intelligent recognition and reasoning engine to accurately retrieve files and match rules. This reduces invalid retrieval and reasoning ambiguity from the source, achieving efficient and accurate recognition of the overall technical solution.

[0034] The factory configuration file library includes a file storage unit, a retrieval unit, and a retrieval enhancement unit. The file storage unit stores the factory facility configuration file database, enabling standardized management of massive files and providing a structured data source for retrieval and subsequent model training. The retrieval unit performs precise searches based on factory configuration information files and metadata to obtain search results, quickly identifying explicitly related files and avoiding invalid traversal of all files, significantly improving retrieval efficiency. The retrieval enhancement unit combines an intelligent recognition engine to semantically enhance the search results, effectively identifying indirect relationships such as parameter changes and detection procedures and regulations, overcoming the limitation of traditional retrieval which can only match explicit keywords. The retrieval enhancement unit includes knowledge graph-to-vector library technology, using similarity calculations based on the vector library to improve the accuracy of retrieval enhancement generation.

[0035] The factory configuration file library stores files including safety-level files and non-safety-level files. The library categorizes safety-level files and non-safety-level files by safety level, precisely aligning with the differentiated control requirements of nuclear safety regulations for files of different levels. During the retrieval and identification process, priority can be given to core safety-level files, avoiding unordered traversal of all files and improving the targeting and efficiency of the retrieval. At the same time, it effectively avoids misjudgments and missed detections caused by confusion between safety-level and non-safety-level files, ensuring the accurate identification of core safety-related files.

[0036] The model training and deployment module includes a model training unit, a model testing unit, a model inference unit, and a model porting and deployment unit. The model training unit trains the model based on the factory configuration file library to obtain an initial intelligent recognition inference model, enabling the model to deeply embed professional logic. The model testing and model inference units test and infer from the initial intelligent recognition inference model to obtain the required intelligent recognition inference model. Through multiple rounds of verification and optimization, the model is ensured to meet stringent requirements for false negative and false positive rates, guaranteeing reliable recognition results. The model porting and deployment unit portes, deploys, and converts the required intelligent recognition inference model and sets it up in the document influence domain intelligent recognition engine, specifically adapting it to the factory's hardware and software environment, enabling rapid model deployment and operation. The model training and deployment module provides high-performance model support tailored to corresponding scenarios for the document influence domain intelligent recognition engine. Furthermore, through a standardized training-testing-deployment process, it ensures efficient and controllable model iteration and application, further enhancing the accuracy and practicality of the overall technical solution.

[0037] The model training unit also includes optimizing the required intelligent recognition and reasoning model based on the decision recommendation results, effectively addressing the pain point of recurring similar errors in existing technologies; optimizing data sources from real-world factory change scenarios, continuously strengthening the model's understanding of its specific relational logic, and further improving the model's domain adaptability; quickly adapting to change scenarios without the need for extensive expert intervention, reducing long-term maintenance costs; and simultaneously driving a continuous reduction and stabilization of the model's false negative and false positive rates within control standards, providing core support for the accuracy and sustainability of the overall technical solution.

[0038] The model reasoning unit includes similarity calculation between the modified feature vector and the file feature vector, knowledge graph path reasoning, and LLM generative reasoning mode. Similarity calculation enables rapid identification of explicitly related files, improving recognition efficiency. Knowledge graph path reasoning deeply mines the unique multi-layered implicit associations corresponding to the current scenario, avoiding missed detection of key cross-professional files. LLM generative reasoning accurately parses professional concepts and semantic logic, adapting to complex file association scenarios. The combination of these three effectively compensates for the limitations of a single reasoning method, ensuring recognition speed while significantly improving recognition accuracy and domain adaptability.

[0039] The decision-making module comprises a rule base unit, an intelligent agent unit, and a human interaction unit. The rule base unit stores configuration change rules, including time rules, scope rules, classification rules, and responsibility attribution rules, providing a unified and compliant standard for decision-making, avoiding judgment biases caused by differences in human experience, and conforming to the normative requirements of nuclear safety regulations for change review. The intelligent agent unit makes intelligent decisions based on initial recommendations and configuration change rules, significantly reducing the cost of human intervention and improving decision-making efficiency. The human interaction unit receives human review and information-based decision-making, ensuring the accuracy and rigor of key decisions, while providing high-quality feedback data for model iteration. Specifically, time rules define the revision time requirements for different types of documents, and scope rules define the document scope logic affected by the change. Time rules clearly define the revision time requirements for different types of documents, ensuring that the revision rhythm of security-level and non-security-level documents accurately matches the equipment change and installation progress, avoiding delays in document revision that could affect change implementation or cause compliance risks. Scope rules define the document scope logic affected by the change, providing clear correlation standards for the identification process and effectively avoiding the omission of cross-disciplinary documents or the misinclusion of unrelated documents.

[0040] The system also includes a human-computer interaction and system integration module, which comprises a visualization unit and a system interface unit. The visualization unit is used to visualize the decision recommendation results, allowing users to quickly and accurately grasp the scope of impact of the document and the core revision points, reducing the time cost and decision-making difficulty of manual review. The system interface unit is used to connect with subsequent change management processes, ensuring smooth process integration and providing support for document revision tracking and operation trajectory retention, in line with the nuclear safety regulations' requirements for full-process traceability.

[0041] A computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the steps of the method provided in the second aspect of the present invention.

[0042] Example: Please see Figure 1 , Figure 1 The present invention provides a method for generating factory operation and maintenance information decisions, which includes the following steps: S11: Build a plant configuration file library that contains a database of plant facility configuration files.

[0043] S12: Obtain change information from the preceding change management process and preprocess it to generate a description of the change elements.

[0044] S13: Train the intelligent recognition reasoning model to obtain the required intelligent recognition reasoning model, and set the required intelligent recognition reasoning model in the document influence domain intelligent recognition engine.

[0045] S14: Based on the description of the change elements, identify the file influence domain in conjunction with the factory configuration file library to obtain the initial recommendation result.

[0046] S15: Based on the initial recommendation results, combined with configuration change rules and manual review and decision-making, generate decision recommendation results.

[0047] This invention is applicable to the generation of operation and maintenance information decisions for various types of factories. The following explanation uses a nuclear power plant as an example.

[0048] In step S11, please refer to Figure 1 Step S11 involves building a plant configuration file library that includes a plant facility configuration file database.

[0049] In step S11 above, safety-level documents, non-safety-level documents, and related metadata of nuclear power plants can be uniformly and standardizedly stored. Configuration file retrieval and retrieval enhancement generation functions are also enabled. The files in the library are vectorized to build a file vector knowledge base, providing standardized and structured data support for subsequent change element matching, semantic retrieval, association mining, and intelligent recognition reasoning. In specific implementation, the configuration files such as operation and maintenance procedures, technical specifications, safety analysis reports, and testing procedures corresponding to various professional systems, equipment, and components are uniformly stored in the library. They are classified and indexed according to safety level, professional category, and equipment code, and a file vector knowledge base that can be used for vector similarity calculation is formed simultaneously. This provides a complete, searchable, and augmentable data foundation for quickly and accurately identifying the impact domain of files corresponding to facility configuration changes, thereby improving the efficiency of operation and maintenance specification updates and quickly forming new operation and maintenance specifications adapted to the current change scenario.

[0050] In step S12, please refer to Figure 1 In step S12, change information from the preceding change management process is obtained and preprocessed to generate a change element description.

[0051] In step S12 above, the change information interface unit first connects to the business systems of the nuclear power plant's change management pre-process, including change applications, technical feasibility studies, and safety reviews. It automatically obtains all change-related information within a pre-specified scope, such as change application forms, feasibility reports, and review opinions, eliminating the need for manual input. This effectively avoids the tedious operation and human error associated with manual input and significantly improves the efficiency of change information acquisition. Taking the nuclear power plant application scenario as an example, it can accurately connect to the implementation method MOC change management front-end process system, automatically synchronizing target change documents and associated original data according to preset interface rules. It completely retains core content such as equipment ownership, model parameters, safety classification, and change motivation, preventing the loss of key information caused by manual relaying.

[0052] Furthermore, standardized preprocessing is carried out on original change information from multiple sources and in multiple formats. This involves eliminating redundant and invalid information, standardizing data formats, and normalizing professional terminology. Irrelevant notes, duplicate attachments, and other interfering content are removed, transforming scattered and heterogeneous data into standardized structured data. Industry-specific thesaurus is used to unify synonyms and eliminate terminological ambiguity. Based on this, core information is precisely extracted to generate a structured description of change elements, including the change subject, change type, security level, parameter differences, and related supporting documents. This clearly defines the core content of changes such as equipment component adjustments, parameter optimizations, and functional upgrades, while retaining effective related details and desensitizing and standardizing irrelevant sensitive information to form high-quality standardized input data. This method can quickly transform the change details of the physical entity of the facility into standardized and readable element information, laying a precise data foundation for subsequent intelligent matching of document impact domains, mining of implicit relationships, and generation of operation and maintenance decisions. It effectively solves the pain points of traditional change information sorting, which is cumbersome, time-consuming, and difficult to quickly adapt to on-site problem adjustments and timely output of corresponding operation and maintenance management basis. This significantly shortens the cycle of subsequent operation and maintenance standard correlation analysis and update implementation.

[0053] In step S13, please refer to Figure 1 In step S13, the intelligent recognition reasoning model is trained to obtain the required intelligent recognition reasoning model, and the required intelligent recognition reasoning model is set in the document influence domain intelligent recognition engine.

[0054] In step S13 above, the structured configuration file data and metadata in the factory configuration file library are first integrated, along with the document influence domain decision recommendation annotation data after manual review and calibration. Simultaneously, the standardized change element descriptions generated by preprocessing (including core content such as change subject, change type, and safety level) are imported to construct a dedicated training dataset that fits the factory change scenario. Iterative training is carried out by adjusting key parameters such as model feature extraction weights and similarity calculation thresholds to deeply precipitate the professional logic of the factory domain implementation methods: "change subject implementation method - implementation method change type implementation method - implementation method document association implementation method - implementation method compliance constraint". An initial intelligent recognition and reasoning model is generated. Taking the nuclear power plant application scenario as an example, based on the large model framework adapted to the industrial scenario and retrieval enhancement generation technology, the association logic of nuclear power safety level and non-safety level documents and nuclear safety regulatory constraints can be specifically integrated to improve the model's adaptability to the nuclear power professional scenario.

[0055] Furthermore, a multi-dimensional model testing and inference verification scheme was designed to simulate typical and complex change scenarios such as factory equipment parameter modifications, structural adjustments, and functional upgrades. File data covering different security levels and professional categories was selected as the test set. Performance testing was conducted with the false negative rate and false positive rate as the core evaluation indicators. Simultaneously, the model's inference performance was verified. For issues discovered during testing, such as false negatives of cross-professional files and false positives of unrelated files, the feature weights of the training dataset and model parameters were adjusted in reverse. Through multiple rounds of testing-implementation-implementation-inference-implementation-optimization cycles, the core performance of the model was ensured to meet the stringent requirements of factory change management, resulting in a model that conforms to practical real-world conditions. The system employs an intelligent recognition and reasoning model to meet application requirements. Finally, the optimized model undergoes format conversion to adapt to the factory's industrial-grade hardware and software operating environment. This completes the deployment and debugging of the model within the file impact domain intelligent recognition engine, ensuring seamless collaboration between the model and stages such as change information acquisition, preprocessing, and configuration file libraries. This enables efficient and accurate recognition of the file impact domain corresponding to changes, addressing the pain points of traditional manual recognition—low efficiency, poor consistency, difficulty in quickly adapting to complex change scenarios, and inability to promptly output updated operational specifications—from a core technical perspective. This provides core technical support for the rapid updating and accurate implementation of subsequent operational specifications, ensuring the compliance and timeliness of factory operational management.

[0056] In step S14, please refer to Figure 1 In step S14, the influence domain of the file is identified based on the description of the change element and in conjunction with the factory configuration file library to obtain the initial recommendation result.

[0057] In step S14 above, the intelligent recognition engine first receives the standardized change element description output by the change information interface and the preprocessing module, and fully obtains core information such as the change subject, change type, security level, change parameters and related information, which serves as the core basis for document impact domain identification. Then, it calls the intelligent recognition reasoning model that has been iteratively optimized through model training and deployment modules and meets the stringent requirements of nuclear power scenarios, driving the nuclear power plant configuration file library to carry out orderly retrieval and enhancement operations.

[0058] The retrieval unit in the nuclear power plant configuration file library conducts multi-dimensional and precise searches based on the core information of change elements, quickly filtering out configuration files that have an explicit relationship with the change, effectively avoiding invalid traversal of all files, and improving retrieval accuracy and efficiency. It also triggers the retrieval enhancement unit to start the retrieval enhancement process, splitting and vectorizing all configuration files in the file library into text blocks, constructing a file vector knowledge base covering various professional fields of nuclear power, and then generating prompt word vectors based on change elements. These prompt words are then accurately matched with the vector knowledge base to achieve prompt word amplification. At the same time, it uses a large model to analyze nuclear power professional technical data, explores cross-professional implicit relationships between changes and files, supplements key information not covered by explicit searches, and makes up for the limitations of traditional keyword searches.

[0059] The intelligent recognition and reasoning model combines the explicit related files obtained from retrieval with the implicit related information obtained from enhanced mining. Based on professional rules such as the safety classification of nuclear power documents and the priority of related logic, it intelligently determines the degree of impact of changes on each document and the priority of revision, and integrates them to form an initial recommendation result. The result covers the list of related documents, the safety level of each document, the related dimensions, and the preliminary revision direction.

[0060] In step S15, please refer to Figure 1 Step S15 generates decision recommendation results based on the initial recommendation results, combined with configuration change rules and manual review and decision-making.

[0061] In step S15 above, the decision module first receives the initial recommendation results output by the intelligent recognition engine, fully obtains the core information such as the list of associated files, the degree of impact of each file, and preliminary revision suggestions, and then calls the preset configuration change rule library to automatically verify the compliance and relevance of the initial recommendation results. It focuses on checking the file association logic, security level matching degree, and compliance requirements, and eliminates content that does not comply with nuclear safety specifications or whose association logic is invalid. It initially screens out the scope of file impact and revision suggestions that meet the requirements.

[0062] Furthermore, a manual review and decision-making process was initiated. Professional reviewers used a visual interactive interface to comprehensively verify the completeness and accuracy of the initial recommendation results against the nuclear power plant operation and maintenance management specifications and safety regulations. The implementation method focused on confirming whether any cross-disciplinary related documents were missing, whether the document revision priority met the safety level requirements, and whether the related logic aligned with the actual operation and maintenance scenario. At the same time, based on the actual on-site operation, the initial recommendation results were adjusted accordingly, supplementing any missed cross-disciplinary related documents, correcting parameter matching deviations, and clarifying the revision sequence, responsibility attribution, and specific revision requirements for each document.

[0063] Based on this, the system integrates automated rule verification results with manual review opinions to generate decision recommendation results that include a clear revision list, priority ranking, responsibility attribution, and completion deadlines. It clearly defines the revision standards and implementation requirements of each related document, realizing closed-loop management of the implementation method: "rule verification implementation method - manual review implementation method - decision output implementation method." This effectively solves the problems of low efficiency, poor consistency, and insufficient compliance of traditional manual judgment, ensuring that the decision recommendation results are consistent with the actual situation on site and comply with nuclear safety regulations. It provides accurate and implementable basis for subsequent document revision and operation and maintenance scheduling, significantly improving the compliance and efficiency of nuclear power plant configuration change management. At the same time, it accumulates high-quality labeled data for subsequent model iteration and optimization, further strengthening the system's adaptability to various change scenarios.

[0064] It is worth noting that the decision recommendation results confirmed by manual review can be used as high-precision labeled data, which can be automatically associated and matched with the original change element description and impact description to form a structured training sample of "change element - impact description - confirmed impact domain". Then, this sample is input into the model training and deployment module, and the intelligent recognition and reasoning model is optimized in a targeted manner using small sample fine-tuning technology. The semantic association weights between change parameters and related files are updated, the factory facility configuration knowledge graph is updated simultaneously, and exclusive association nodes between the subject of this change and related files are added. Finally, the optimized semantic association weights and the updated knowledge graph are re-injected into the file impact domain intelligent recognition engine to complete the online iterative update of the model.

[0065] Please see Figure 2 , Figure 2 This invention provides a schematic diagram of a factory operation and maintenance information decision generation system. The system may include: a change information interface and preprocessing module 10, a factory configuration file library 20, a model training and deployment module 30, an intelligent recognition engine 40, and a decision module 50. The change information interface and preprocessing module 10 obtains change information from the preceding change management process and performs preprocessing to generate change element descriptions. The factory configuration file library 20 includes a factory facility configuration file database, file storage, file retrieval, and enhanced retrieval generation functions. The model training and deployment module 30 trains the intelligent recognition inference model to obtain the required intelligent recognition inference model and sets the required intelligent recognition inference model in the file influence domain intelligent recognition engine 40. The intelligent recognition engine 40 identifies the file influence domain based on the change element descriptions and the factory configuration file library to obtain initial recommendation results. The decision module 50 generates decision recommendation results based on the initial recommendation results, combined with configuration change rules and manual review and decision-making.

[0066] In one optional implementation, the change information interface and preprocessing module 10 includes a first change information interface unit and a second change information interface unit. The first change information interface unit interfaces with the change management preprocess to obtain change information, avoiding the tediousness and errors of manual entry and significantly improving the efficiency of change information acquisition. The second change information interface unit cleans and standardizes the change information, extracts change element descriptions, uses a professional thesaurus for entity standardization, and combines cleaning and format unification processing to solve the problems of inconsistent change information formats and diverse professional terminology, ensuring the standardization and accuracy of change element extraction. The change element descriptions include the change subject, change type, safety level, change parameters, and related data. The change subject is clearly defined as a specific system, equipment, or component of the plant, such as the reactor coolant system, steam generator, control rod drive mechanism, and other core objects. The change types are divided into three categories: parameter modification, structural adjustment, and functional upgrade. Parameter modification covers... The adjustments to technical indicators such as operating pressure and temperature thresholds, structural adjustments including the addition, removal, or layout optimization of equipment components, and functional upgrades involving control logic iteration and the addition of monitoring and early warning functions; the safety level corresponds to the classification standards of the field, clearly defining the safety level 1, safety level 2, or non-safety level attributes of the objects involved in the changes and related documents; the change parameters are key technical data directly related to the changes, including numerical comparisons before and after parameter modification, dimensional specifications of structural adjustments, and performance indicators of functional upgrades; the related materials include supporting materials such as change application forms, technical feasibility demonstration reports, compliance review documents, and history of similar changes; by comprehensively extracting the above core elements and forming structured data, high-quality data input is provided for the subsequent intelligent recognition and reasoning engine's accurate document retrieval, knowledge graph association matching, and rule verification of the decision module 50, reducing invalid retrieval and reasoning ambiguity from the source, and helping the overall technical solution achieve efficient and accurate identification and deep adaptation to the change scenario.

[0067] In one optional implementation, the factory configuration file library 20 includes a file storage unit, a retrieval unit, and a retrieval enhancement unit. The file storage unit stores factory configuration information files and metadata, enabling standardized management of massive files and providing a structured data source for retrieval and subsequent model training. The retrieval unit performs precise retrieval based on the factory configuration information files and metadata to obtain retrieval results, enabling rapid identification of explicitly related files, avoiding invalid traversal of all files, and significantly improving retrieval efficiency. The retrieval enhancement unit, in conjunction with the intelligent recognition engine 40, performs semantic enhancement on the retrieval results, effectively identifying indirect associations such as parameter changes with testing procedures and regulatory clauses. It integrates knowledge graph and vector library technologies, constructing a multi-layered association network of "change-facilities-documents-regulations" based on the corresponding domain knowledge graph, and combining the feature vector transformation and similarity calculation functions of the vector library to effectively identify cross-professional indirect associations such as parameter changes with testing procedures and regulatory clauses, generating a file impact description that fits the change requirements, making up for the deficiency of traditional keyword retrieval that can only match explicit associations, and further improving the comprehensiveness and practicality of retrieval results.

[0068] The factory configuration file library 20 stores files including security-level files and non-security-level files. The factory configuration file library 20 stores security-level files and non-security-level files according to their security level, which accurately meets the differentiated control requirements of security regulations for files of different levels. During the retrieval and identification process, priority can be given to core security-level files, avoiding disordered traversal of all files and improving the targeting and efficiency of the retrieval. At the same time, it effectively avoids misjudgment and missed detection caused by the confusion between security-level and non-security-level files, ensuring the accurate identification of core security-related files.

[0069] In one optional implementation, the model training and deployment module 30 includes a model training unit, a model testing unit, a model inference unit, and a model porting and deployment unit. The model training unit trains the model based on the factory configuration file library 20 to obtain an initial intelligent recognition inference model. It conducts model training based on the massive structured file data, metadata, and manually reviewed decision recommendation annotation data stored in the factory configuration file library 20 to create an initial intelligent recognition inference model that meets the actual needs of change management. The model testing unit and the model inference unit test and infer the initial intelligent recognition inference model to obtain the required intelligent recognition inference model. The model testing unit designs a multi-dimensional verification scheme for the initial model and conducts performance tests simulating complex change scenarios. The model inference unit simultaneously verifies the inference effect. Through multiple rounds of test-inference-optimization cycles, it strictly controls the model's core indicators such as the false negative rate and false positive rate. Through multiple rounds of verification and optimization, it ensures that the model meets the stringent requirements for false negative rate and false positive rate, ensuring the reliability of the recognition results. The model porting and deployment unit ports, deploys, and converts the required intelligent recognition inference model and sets it on the file influence domain intelligent recognition engine 40, specifically adapting it to the factory's hardware and software environment to achieve rapid model deployment and operation.

[0070] The model reasoning unit includes similarity calculation between the modified feature vector and the file feature vector, knowledge graph path reasoning, and LLM generative reasoning mode. Similarity calculation enables rapid identification of explicitly related files, improving recognition efficiency. Knowledge graph path reasoning deeply mines unique multi-layered implicit associations, avoiding missed detection of key cross-disciplinary files. LLM generative reasoning accurately analyzes professional concepts and semantic logic, adapting to complex file association scenarios. The combination of these three effectively compensates for the limitations of a single reasoning method, ensuring recognition speed while significantly improving recognition accuracy and domain adaptability.

[0071] The model training unit also includes optimizing the required intelligent recognition and reasoning model based on the decision recommendation results, effectively addressing the pain point of recurring similar errors in existing technologies; the optimized data source is from real-world change scenarios, which can continuously strengthen the model's understanding of specific relational logic and further improve the model's domain adaptability; it can quickly adapt to change scenarios without the need for extensive expert intervention, reducing long-term maintenance costs; at the same time, it drives the model's false negative rate and false positive rate to continuously decrease and stabilize within the control standards, providing core support for the accuracy and sustainability of the overall technical solution; the model training unit also supports receiving manually reviewed and labeled data output by the decision module, and realizes online iterative updates of the model based on small sample fine-tuning, forming a system self-learning optimization mechanism.

[0072] In one optional implementation, the decision-making module 50 includes a rule base unit 51, an intelligent agent unit, and a human interaction unit. The rule base unit 51 stores configuration change rules that meet the needs of change management, specifically including time rules, scope rules, classification rules, and responsibility attribution rules. This provides a unified and compliant standard basis for the decision-making process, effectively avoiding judgment biases caused by differences in human experience, and conforming to the normative requirements of security regulations for change review. The time rules specify the revision time requirements for different types of documents, ensuring that the revision rhythm of security-level documents and non-security-level documents is accurately matched with the equipment change and installation progress, avoiding the impact of delayed document revisions on change implementation or the occurrence of compliance risks. The scope rules define... The logic for defining the scope of documents affected by changes provides clear association standards for the identification process, effectively avoiding the omission of cross-disciplinary documents or the misinclusion of unrelated documents. The intelligent agent unit conducts automated intelligent decision-making based on initial recommendation results and configuration change rules, significantly reducing the cost of manual intervention and significantly improving decision-making efficiency. The human interaction unit receives review opinions and decision adjustment information from professionals, ensuring the accuracy and rigor of decisions on key change matters, and providing high-quality real business scenario feedback data for model iteration and optimization. The three work together to build a multi-level decision-making mechanism of "rule constraints - intelligent decision-making - manual verification", comprehensively ensuring the compliance, efficiency and reliability of factory document impact domain decisions.

[0073] The time rules define the revision time requirements for different types of documents, while the scope rules define the document scope logic affected by the change. The time rules clarify the revision time requirements for different types of documents, ensuring that the revision rhythm of security-level and non-security-level documents is accurately matched with the equipment change and installation progress, avoiding the impact of delayed document revisions on the implementation of changes or the occurrence of compliance risks. The scope rules define the document scope logic affected by the change, providing a clear correlation standard for the identification process and effectively avoiding the problems of missing cross-disciplinary documents or mistakenly including unrelated documents.

[0074] In an optional implementation, the system further includes a human-computer interaction and system integration module 60, which interacts with users and subsequent change management processes based on decision recommendation results. The decision module, as the core hub connecting intelligent identification decisions with actual business processes and user operations, has a built-in visualization unit that visually presents the impact and hierarchical relationship of documents through heatmaps and tree structures, assisting professionals in quickly locating security-level core documents and key revisions, reducing the information comprehension cost of manual review. It also features a human interaction unit, supporting user review and adjustment of the completeness, accuracy, and granularity of decision recommendation results. The generated feedback data can be directly input into the model training and deployment module 30, providing calibration basis in real business scenarios for the iterative optimization of the intelligent identification inference model. Addressing the characteristics of multi-professional collaboration and high-security control, the decision module supports hierarchical permission management and possesses good software and hardware compatibility, adapting to existing factory digital platforms. It can be quickly deployed without large-scale modifications, significantly reducing system integration costs and timelines, and comprehensively improving the user adaptability and process collaboration of the overall technical solution.

[0075] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the steps of the method described in the first aspect above.

[0076] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for generating factory operation and maintenance information decisions, characterized in that, include: Build a plant configuration file library that includes a database of plant facility configuration files; Obtain change information from the preceding change management processes and preprocess it to generate change element descriptions; The intelligent recognition reasoning model is trained to obtain the required intelligent recognition reasoning model, and the required intelligent recognition reasoning model is set in the document influence domain intelligent recognition engine; Based on the description of the change elements, the file influence domain is identified using the factory configuration file library to obtain initial recommendation results; Based on the initial recommendation results, combined with configuration change rules and manual review and decision-making, a decision recommendation result is generated.

2. The method as described in claim 1, characterized in that: The factory configuration file library includes file storage, file retrieval, and retrieval enhancement generation functions.

3. The method as described in claim 2, characterized in that: The enhanced search generation function includes: The files in the factory configuration file library are transformed into vectors to build a file vector knowledge base; Based on the changed elements, a large model is used to semantically expand the content of the retrieved files and generate an impact description related to the changed elements. An initial prompt word vector is constructed based on the description of the changed elements and the description of the impact. Similarity matching is then performed using the file vector knowledge base to amplify the semantics of the prompt words. By drawing upon professional technical data and combining it with the amplified prompt word vectors, implicit cross-professional associations can be mined to supplement the association information not covered by explicit retrieval.

4. The method as described in claim 1, characterized in that: The process of training the intelligent recognition reasoning model to obtain the desired intelligent recognition reasoning model includes: The model is trained based on the factory facility configuration file database to obtain an initial intelligent recognition and reasoning model; The initial intelligent recognition reasoning model is tested and verified to obtain the desired intelligent recognition reasoning model.

5. The method as described in claim 1 or 3, characterized in that: The step of identifying the file impact domain based on the description of the change elements and in conjunction with the factory configuration file library to obtain the initial recommendation result includes: Based on the changed elements and their impact descriptions, a change feature vector is generated; The similarity between the changed feature vector and the file feature vector in the file vector knowledge base is calculated to obtain a set of explicitly associated files; Starting with the aforementioned set of explicit related files, path reasoning is performed based on the factory facility configuration knowledge graph to mine cross-professional implicit related files; The explicit and implicit related files are input into the LLM to perform generative reasoning, which parses the professional relationship logic between the files and changes and marks the degree of impact. Integrate explicit and implicit related documents and their degree of influence to obtain initial recommendation results.

6. The method as described in claim 3, characterized in that, The method further includes: The decision recommendation results are used as labeled data and matched with the descriptions of the change elements and the descriptions of the impacts to form training samples. Based on the training samples, a small-sample fine-tuning technique is used to update the semantic association weights between the changed parameters and the associated files; The factory facility configuration knowledge graph is updated synchronously, and exclusive association nodes for the change subject and related files are added; The optimized semantic association weights and the updated knowledge graph are input into the required intelligent recognition and reasoning model.

7. The method as described in claim 1, characterized in that: The description of the change elements includes the subject of the change, the type of change, the safety level, the change parameters, and related information. The type of change includes parameter modification, structural adjustment, and functional upgrade. The subject of the change is a factory system, equipment, or component.

8. The method as described in claim 1, characterized in that: The configuration change rules include time rules, scope rules, classification rules, and responsibility attribution rules; the time rules are the revision time requirements for different types of documents; the scope rules are the document scope logic affected by the change.

9. A factory operation and maintenance information decision generation system, characterized in that, include: The plant configuration file library contains a database of plant facility configuration files, as well as file storage, file retrieval, and enhanced generation capabilities. The change information interface and preprocessing module are used to obtain change information from the preceding change management process and preprocess it to generate a description of the change elements. The model training and deployment module is used to train the intelligent recognition reasoning model to obtain the required intelligent recognition reasoning model, and to set the required intelligent recognition reasoning model in the file influence domain intelligent recognition engine; The intelligent recognition engine is used to identify the file influence domain based on the description of the change elements and the factory configuration file library, so as to obtain an initial recommendation result; The decision-making module is used to generate decision-making recommendation results based on the initial recommendation results, combined with configuration change rules and manual review and decision-making.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions that, when executed by a processor, perform the steps of the method according to any one of claims 1-8.

Images