Nuclear power plant DCS intelligent training method and platform based on knowledge questions and answers

Abstract

The invention belongs to the technical field of nuclear power plant DCS intelligent training, aims to solve the problems of knowledge redundancy, noise interference, interaction defects and lack of real-time performance in nuclear power plant DCS intelligent training, and discloses a nuclear power plant DCS intelligent training method and platform based on knowledge questions and answers. Two-stage optimization is carried out based on a DeepSeek model, an MLA / MoE layer is firstly frozen to adjust a classification head to realize field adaptation, then a top-layer network is finely adjusted through a LoRA technology, a question and answer knowledge base is enhanced by utilizing multi-round dialogue extension and prompt annotation, intelligent retrieval, real-time question and answer and personalized recommendation are integrated to construct a simulation training platform, a closed-loop feedback mechanism is established, and the real-time question and answer knowledge base is established. Updating of the knowledge base and iteration of the model are triggered through expert labeling errors, and dynamic optimization of the training system is achieved; the platform comprises an application layer, a business function layer, a service layer, a basic support layer and an edge layer. The system can improve the training efficiency and quality, and guarantees the professional skill and safety awareness of operation and maintenance personnel.

Description

Technical Field

[0001] The present application belongs to the technical field of nuclear power plant DCS intelligent training, and in particular relates to a nuclear power plant DCS intelligent training method and platform based on knowledge question and answer. Background Art

[0002] With the development of nuclear power technology, the requirements for operators' professional skills are becoming increasingly stringent. Existing nuclear power plant DCS training platforms provide a unified learning plan, rigidly teaching based on existing teaching resources without exploring and optimizing the deeper information within these resources. This means that trainees cannot get timely and effective answers to their questions, which affects learning efficiency and effectiveness.

[0003] Existing nuclear power plant DCS-related training materials, operating manuals, fault cases, and other information contain large amounts of information and are relatively complex, filled with a large amount of irrelevant information. Directly using large models for training will result in the model not being accurate or direct enough when answering specific questions.

[0004] While standard structured data related to nuclear power plant DCS systems can better meet the needs of precise queries, the data volume is relatively small. Although large models can theoretically capture complex patterns, this capability is often underutilized in the absence of data, resulting in poor performance.

[0005] Specifically, existing nuclear power plant DCS intelligent training suffers from knowledge redundancy and noise interference. Existing nuclear power plant DCS training materials contain a large amount of unstructured, redundant, or low-relevance information (such as general theories, non-core equipment parameters, and outdated specifications). Directly using this information for large model training can lead to ambiguity or errors in the model's answers to specialized questions due to noise interference. Key operational logic can easily be overwhelmed by irrelevant information, reducing the accuracy and directness of answers.

[0006] Existing nuclear power plant DCS intelligent training suffers from shortcomings in interaction and real-time performance. Existing nuclear power plant DCS training is mostly delivered on a one-off basis, lacking an interactive feedback mechanism for trainees. Furthermore, with evolving technology and safety standards, traditional training materials are difficult to update in real time, resulting in lagging model training data and impacting the timeliness and accuracy of responses. Summary of the Invention

[0007] The main purpose of this application is to provide a knowledge question-answering-based nuclear power plant DCS intelligent training method and platform to solve the problems of knowledge redundancy and noise interference in nuclear power plant DCS intelligent training.

[0008] Another object of the present application is to provide a knowledge question-answering-based nuclear power plant DCS intelligent training method and platform to solve the problems of interaction defects and lack of real-time performance in nuclear power plant DCS intelligent training.

[0009] In order to achieve the above objectives, this application provides the following technical solutions:

[0010] On the one hand, the present application provides a knowledge question-answering-based intelligent training method for a nuclear power plant DCS, comprising:

[0011] Step 1: Collect nuclear power plant DCS training materials, operation manuals, and fault case data to build an original database covering theory, operation, and safety specifications;

[0012] Step 2: Clean and standardize the data to generate a structured training set and domain corpus;

[0013] Step 3: Perform two-stage fine-tuning based on the DeepSeek model, freeze the parameters of the MLA layer and MoE layer and adjust the classification head to complete general domain adaptation. Then, insert the low-rank matrix through the LoRA adapter to optimize the top attention layer and MLP layer;

[0014] Step 4: Build a question-answer knowledge base based on structured question-answer pairs, and use multi-round dialogue expansion and domain hint annotation for data enhancement;

[0015] Step 5: Integrate intelligent retrieval, question-answering engine, and personalized recommendation modules to build a training platform and generate a simulated operation interface with a real-time question-answering window;

[0016] Step 6: Establish a feedback loop mechanism to dynamically maintain the knowledge question and answer database, and trigger knowledge base correction and model iterative training by having experts mark incorrect answers.

[0017] In step 3, the strategy of freezing the bottom layer and fine-tuning the top layer is adopted. The general model is fine-tuned using common data in the nuclear power DCS field. The parameters of the MLA layer and MoE layer of the DeepSeek model are frozen, the bottom-level general language features are retained, and only the last layer of classification head is adjusted and replaced.

[0018] In step 4, the LoRA adapter is used to fine-tune the model in step 3. A low-rank matrix is inserted into the top-level attention layer and MLP layer to adjust the model output, and a Dropout layer is added to prevent the model from overfitting. The low-parameter characteristics of LoRA are used to adapt to small amounts of structured data, while the bottom layer is frozen to ensure general knowledge, forming a lightweight fine-tuning paradigm that collaborates with general and domain knowledge.

[0019] In step 4, the general model is initially fine-tuned using basic domain knowledge to obtain a domain generalization model, and then the domain generalization model is fine-tuned again based on the structured data of professional knowledge questions and answers to obtain a domain enhancement model.

[0020] On the other hand, the present application provides a knowledge question-answering-based nuclear power plant DCS intelligent training platform, comprising:

[0021] The application layer provides a user interaction portal and a personalized operation interface, supporting access to training courses, simulation operations, online examinations, and document review.

[0022] The business function layer, including the student side and the administrator side, is used to implement training plan formulation, course arrangement, examination management, question bank maintenance, question and answer interaction, and training record tracing;

[0023] The service layer is used to provide identity authentication, permission control, log auditing, message push, security protection and load balancing services;

[0024] The basic support layer is used to implement data governance, interface communication and network security protection through server clusters, storage devices and operating systems;

[0025] The edge layer is used to complete real-time data collection, preprocessing and low-latency transmission through data collectors and smart sensors, supporting localized decision-making and closed-loop control.

[0026] In some embodiments, the business function layer includes:

[0027] Dynamic course generation module, used to automatically adjust the training difficulty based on the trainees' operation records;

[0028] A multimodal question-answering engine that supports multi-dimensional knowledge interaction using text, voice, and diagrams.

[0029] In some embodiments, the service layer includes:

[0030] Operation playback service module, used to record and reconstruct the trainees' simulated operation process;

[0031] The abnormal behavior detection module is used to identify illegal operation behaviors based on the preset rule base.

[0032] In some embodiments, the edge layer comprises:

[0033] Industrial protocol adapter module for compatibility with Modbus, PROFIBUS and OPC UA communication protocols;

[0034] The local cache database is used to store the training operation data of the last 24 hours.

[0035] In some embodiments, the student terminal includes:

[0036] The planning center unit is used to help students learn in stages according to the training plan assigned by the administrator, test their learning results through the interweaving of courses and question-and-answer modules, and dynamically adjust their learning progress;

[0037] The course center unit is used to integrate all course resources on the platform and supports classification and screening;

[0038] The examination center unit is used for the management of the entire online examination process;

[0039] The Q&A center unit is used to centrally manage the data of the phased Q&A modules participated by students and display the Q&A results and score details;

[0040] The data center unit is used to store and display various file resources of the training platform and supports hierarchical authority management;

[0041] The training record unit is used to record the execution of the training plan throughout the entire process and associate the plan, course, and examination data through batch numbers.

[0042] In some embodiments, the administrator terminal includes:

[0043] The plan management unit is used to develop training plans that include basic information and course / question and answer / exam requirements, and supports administrators to publish plans to designated personnel;

[0044] The course management unit is used to create and edit course content, arrange teaching schedules, and monitor course progress and student completion rates in real time;

[0045] The exam management unit is used to configure exam rules, intelligently generate exam papers, publish online exams, and manage scores;

[0046] The marking center unit is used to handle the manual marking of subjective questions, support administrators to perform multi-dimensional scoring on short-answer questions, and automatically generate comprehensive test score reports;

[0047] Question and answer test unit, used for periodic learning achievement assessment, by setting course-related questions and implementing scoring feedback;

[0048] Resource management unit, used to centrally manage training materials and establish a structured knowledge base system;

[0049] Intelligent question bank unit, used to maintain question resources and support classification management in professional fields;

[0050] Training record management unit, used for full record data, multi-dimensional retrieval and visual analysis.

[0051] Compared with the existing technology, the knowledge question-answering-based nuclear power plant DCS intelligent training method and platform provided by this application has the following beneficial effects:

[0052] This application utilizes a domain knowledge base construction and structured processing. Through question-and-answering, knowledge fragments directly related to the core functions of a nuclear power plant DCS are extracted from training materials, operating manuals, and troubleshooting cases, filtering out irrelevant information. Core knowledge is converted into structured data to establish an expert knowledge base. Domain tags, such as "nuclear safety regulations," "pressure control," and "emergency response," are added to the knowledge base content to facilitate the model's rapid identification of key information and improve the focus of responses.

[0053] This application utilizes dynamic adaptation and fine-tuning of large models. Through interactive question-and-answer sessions, a review and feedback loop is established between domain experts and trainees. Through multiple rounds of question-and-answer sessions, incorrect or outdated information is flagged and stored in batches in the knowledge base. The knowledge base is connected to the question-and-answer data from the nuclear power plant DCS training system in real time, triggering incremental updates to the knowledge base at regular intervals. This allows for continuous fine-tuning of the model and enhances its dynamic responsiveness.

[0054] This application is a method for implementing a special model for the nuclear power industry based on a large model, and an advanced teaching platform that uses artificial intelligence technology to enhance the educational experience and optimize teaching resources.

[0055] This application aims to improve employees' understanding and application of distributed control systems (DCS) to ensure the safe operation of nuclear power plants. By using large-scale modeling technology to refine the content of question-and-answer modules in the platform into a knowledge base, the model is trained to generate question-and-answer results for evaluating training effectiveness, thereby improving training efficiency and quality and ensuring the professional skills and safety awareness of nuclear power plant DCS operation and maintenance personnel.

[0056] This application introduces a large model technology training expert knowledge question and answer database to establish a personalized intelligent training platform, which can not only utilize the generalization ability of the large model, but also respond to user query requests quickly and accurately.

[0057] This application combines the advantages of both to design a hybrid strategy based on a large amount of knowledge pre-training model and then fine-tuning with a small batch of data. Through large-scale pre-training, a wide range of basic knowledge related to nuclear power plant DCS training is obtained, and then fine-tuning is performed through structured data of professional questions and answers to improve the performance of the model in the field of intelligent question and answer training. The advantages of the general large model are efficient data processing and computing capabilities, wide applicability and flexibility, and are suitable for supporting basic capabilities in multiple fields. The advantages of the domain knowledge base are professionalism, accuracy and efficiency, which can make up for the shortcomings of the general model in specific fields and ensure practicality through dynamic updates. The combination of the two (such as "large model + knowledge base") can further improve the level of intelligence. In nuclear power plant DCS training, the knowledge base can filter out irrelevant information, and the large model provides accurate answers, jointly achieving efficient and safe training goals.

[0058] This application can improve the quality of training for nuclear power plant trainees. First, by leveraging the powerful data processing capabilities of large models, educational resources can be managed and allocated more effectively, enabling preliminary exploration of key information from training materials, operating manuals, fault cases, and other sources, acquiring a broad knowledge base and improving information utilization efficiency. Second, fine-tuning is performed based on the pre-trained generalized model. By training on the labeled question-and-answer pairs in the "Daily Question" teaching module, the number of question-and-answer pairs is relatively small but more concentrated, meaning that training can be completed in a shorter time and it is easier to optimize the model's performance for specific tasks, providing trainees with immediate and accurate answers to their questions and enhancing the interactive experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solution of this application, the following is a brief introduction to the drawings required for the technical description.

[0060] Figure 1 Flowchart of the knowledge question-answering-based intelligent training method for nuclear power plant DCS provided for this application;

[0061] Figure 2 The overall architecture diagram of the knowledge question-answering-based nuclear power plant DCS intelligent training platform provided for this application;

[0062] Figure 3 The student-side business process diagram provided for this application;

[0063] Figure 4 The administrator-side business process diagram provided for this application. DETAILED DESCRIPTION

[0064] The following is further explained in detail through specific implementation methods.

[0065] like Figure 1 As shown, this application provides a knowledge question-answering-based DCS intelligent training method for nuclear power plants, including:

[0066] Step 1: Collect nuclear power plant DCS training materials, operation manuals, and fault case data to build an original database covering theory, operation, and safety specifications;

[0067] Step 2: Clean and standardize the data to generate a structured training set and domain corpus;

[0068] Step 3: Perform two-stage fine-tuning based on the DeepSeek model. First, freeze the parameters of the MLA layer and MoE layer and adjust the classification head to complete general domain adaptation. Then, insert the low-rank matrix through the LoRA adapter to optimize the top attention layer and MLP layer.

[0069] Step 4: Build a question-answer knowledge base based on structured question-answer pairs, and use multi-round dialogue expansion and domain hint annotation for data enhancement;

[0070] Step 5: Integrate intelligent retrieval, question-answering engine, and personalized recommendation modules to build a training platform and generate a simulated operation interface with a real-time question-answering window;

[0071] Step 6: Establish a feedback loop mechanism to dynamically maintain the knowledge Q&A database. Experts' annotation of incorrect answers triggers knowledge base corrections and iterative model training. Within this feedback loop mechanism, users can interact with the model through multiple rounds of Q&A, marking incorrect or outdated information, and giving likes or dislikes to Q&A outputs. This feedback data is categorized and annotated and stored in batches in the knowledge base, triggering incremental updates to the knowledge base and continuously fine-tuning the model to enhance its answer accuracy and user satisfaction.

[0072] In step 1, comprehensively collect existing nuclear power plant DCS-related training materials, operating manuals, fault case studies, and other data to establish an original database. Ensure that the data covers as many aspects as possible, including theoretical foundations, actual operating procedures, safety regulations, and other content.

[0073] In step 2, preprocess the data by cleaning and transforming it to obtain a training set and build a large-scale domain corpus. Leveraging existing teaching resources, the collected data is cleaned to remove irrelevant or erroneous information, and the format is standardized to prepare the training dataset.

[0074] In step 3, based on the training of the open source model DeepSeek, the existing model is adjusted to adapt to the terminology and expressions in the field. DeepSeek, an open source large language model suitable for the Chinese environment and of a certain scale, is selected as the infrastructure. The model is fine-tuned and trained based on the terminology and technical details unique to the nuclear power industry to make it better adapt to application scenarios in specific fields. A freeze-bottom-level fine-tuning top-level strategy is adopted, and general model fine-tuning is performed using general data in the nuclear power DCS field. The parameters of the MLA layer and MoE layer of the DeepSeek model are frozen, the underlying general language features are retained, and only the last layer of classification head is adjusted and replaced. Through preliminary fine-tuning, the video memory and computing requirements required by the model are significantly reduced, and it can accurately identify terminology in the nuclear power field. The description of complex processes is more in line with industry standards, reducing the generalization error of the general model.

[0075] In step 4, a question-and-answer knowledge base is established based on a structured dataset of standard "Daily Question" question-and-answer pairs, and the model is fine-tuned. Leveraging the existing "Daily Question" database of question-and-answer pairs, large-scale modeling techniques are used to extract key knowledge points from the Q&A database, forming a structured knowledge base. An intelligent Q&A engine is then developed to automate the process from user input to retrieval of the most relevant answers. Data augmentation techniques are used to annotate the standard Q&A knowledge within the training platform, adding domain-specific prompts such as nuclear power safety, emergency response, and fault diagnosis. The Q&A process is also expanded to a multi-turn dialogue format to improve data quality. Based on the high-quality Q&A data, the model from step 3 is further fine-tuned using the LoRa adapter. Low-rank matrices are inserted into the top-level attention and MLP layers to adjust the model output, and a dropout layer is added to prevent overfitting. Leveraging LoRa's low parameter count, the model adapts to small amounts of structured data while preserving general knowledge at the bottom layer. This creates a lightweight fine-tuning paradigm that leverages "general-domain" collaboration, further enhancing the model's predictive capabilities in specialized domains.

[0076] In step 5, a personalized intelligent training platform based on knowledge Q&A is built based on the domain-enhanced model. This intelligent engine incorporates multiple functions, including intelligent search, intelligent Q&A, knowledge learning, and exam assessment. Students enhance their training effectiveness through interactive learning. Based on their historical training and Q&A records, the platform provides personalized learning resource recommendations, such as similar courses and materials, to help them quickly grasp core concepts and related knowledge.

[0077] In step 6, a feedback loop mechanism is established to dynamically maintain the knowledge Q&A database and continuously optimize the knowledge Q&A engine's answer performance. Power plant experts manually review the model's responses, correcting or supplementing knowledge items associated with incorrect answers. This triggers incremental updates to the Q&A knowledge database and fine-tunes the model. The intelligent training platform continuously learns and improves based on user feedback. As new knowledge is generated, the knowledge base content is regularly updated to ensure both answer accuracy and knowledge freshness.

[0078] In one embodiment, a large-scale corpus is constructed to establish a comprehensive DCS-related database, including training materials, operation manuals, fault cases and other data, to ensure that all necessary knowledge points are covered.

[0079] In one embodiment, domain adaptability adjustments are performed. Based on the characteristics of the nuclear power industry, the general model is initially fine-tuned using basic domain knowledge to obtain a domain generalization model. This is then combined with the structured data of professional knowledge questions and answers to perform secondary fine-tuning on the domain generalization model to obtain a domain enhancement model. This improves its applicability and accuracy in this field, provides students with immediate and accurate answers to questions, and enhances the interactive experience. For example, a freeze-bottom-level fine-tuning strategy is used for initial fine-tuning, followed by data enhancement technology for data adaptation, and further fine-tuning using LoRA technology.

[0080] In one embodiment, a loop feedback mechanism is used to iteratively update the question and answer data generated by users, dynamically maintain the knowledge question and answer database, continuously optimize the knowledge extraction of the pre-trained model, and create a lighter but equally efficient knowledge question and answer engine.

[0081] In addition, if Figures 2 to 4 As shown, this application provides a knowledge question-answering-based nuclear power plant DCS intelligent training platform, including:

[0082] The application layer provides a user interaction portal and a personalized operation interface, supporting access to training courses, simulation operations, online examinations, and document review.

[0083] The business function layer is used to implement training plan formulation, course arrangement, examination management, question bank maintenance, question-and-answer interaction, and training record tracing;

[0084] The service layer is used to provide identity authentication, permission control, log auditing, message push, security protection and load balancing services;

[0085] The basic support layer is used to implement data governance, interface communication and network security protection through server clusters, storage devices and operating systems;

[0086] The edge layer is used to complete real-time data collection, preprocessing and low-latency transmission through data collectors and smart sensors, supporting localized decision-making and closed-loop control.

[0087] Specifically, the application layer serves as the entry point for user interaction with the system, providing an intuitive operational experience. This layer includes training module interfaces for trainees and a management console for administrators. It enables users to access training courses, simulate operating environments, take online exams, and access learning materials. The application layer enables user interaction and a personalized experience for the DCS training platform.

[0088] The business function layer implements all specific training-related functions, including student-side and administrator-side functions, such as training plan development, course scheduling, exam management, question bank management, Q&A interaction, training material management, and training record maintenance, to fully support the implementation of the training process. The business function layer implements the business logic for specific scenarios of the DCS training platform.

[0089] The service layer provides a range of common technical services to support the development, operation, and expansion of upper-layer applications, including identity authentication, permission management, logging, message notifications, inter-system communication, security assurance, load balancing, and more. Through the service layer, the DCS training platform achieves security, reliability, and scalability, and provides a convenient integration path for different business functions.

[0090] The foundational support layer provides the necessary infrastructure for the entire platform, including software and hardware resources such as operating systems, servers, and storage devices. It is responsible for open interface communications, data management and storage, computing resource allocation, and network security. This foundational support layer provides the DCS training platform with elastic resource scalability and a communication technology foundation.

[0091] The edge layer is used for collaborative call processing of on-site training resource libraries, collecting, processing, transmitting, and storing data through data collectors, smart sensors, and other edge computing devices. This layer enables terminal adaptation, data integration, and a closed-loop control system for the DCS training platform, reducing data transmission latency and improving local decision-making efficiency.

[0092] In one embodiment, the application layer includes:

[0093] The student operation interface module integrates the virtual DCS operation panel and the 3D equipment model;

[0094] Management configuration interface module, providing visual arrangement tools for training courses.

[0095] In one embodiment, the business function layer includes:

[0096] Dynamic course generation module automatically adjusts training difficulty based on trainees' operation records;

[0097] A multimodal question-answering engine that supports multi-dimensional knowledge interaction using text, voice, and diagrams.

[0098] In one embodiment, the service layer includes:

[0099] Operation playback service module, used to record and reconstruct the trainees' simulated operation process;

[0100] The abnormal behavior detection module identifies illegal operation behaviors based on the preset rule base.

[0101] In one embodiment, the base support layer comprises:

[0102] Real-time data mirroring module to synchronize the operating status data of the nuclear power plant DCS system;

[0103] Distributed computing framework that supports parallel processing of more than 200 training sessions.

[0104] In one embodiment, the edge layer comprises:

[0105] Industrial protocol adapter module, compatible with Modbus, PROFIBUS and OPC UA communication protocols;

[0106] The local cache database stores the training operation data of the last 24 hours.

[0107] The student side includes a planning center unit, a course center unit, an examination center unit, a question-and-answer center unit, a data center unit, and a training record unit.

[0108] The planning center unit is used to help students learn in stages (such as beginner, intermediate, and advanced) according to the training plan assigned by the administrator. The learning outcomes can be verified through the interspersed courses and question-and-answer modules, and the learning progress can be dynamically adjusted. Students can view the plan cycle, learning progress, and stage completion status in real time to ensure that progress is made as needed. Students study according to the training plan assigned by the administrator. Each plan can include courses in multiple stages, such as beginner, intermediate, and advanced. Question-and-answer modules are interspersed between courses at different stages to verify learning outcomes. Only after passing the test can they proceed to the next stage of learning. Students can view the plan cycle and learning progress by themselves, and the progress completion status can be adjusted in conjunction with the completion status of the course learning.

[0109] The functions of the planning center unit include:

[0110] (1) Plan List Function: Trainees can view the list of training plans they have participated in, and view the plan number, plan name, batch name, batch number, plan start time, plan end time, plan progress, plan introduction, plan details, plan status and plan category of the current training in real time. Plan categories are divided into department-level training plans, section-level training plans, and team-level training plans. Plan status includes not started, in progress, and completed.

[0111] (2) Plan details function: View the detailed information of the plan. Each training plan may contain course information for multiple stages and question and answer information.

[0112] (3) Start learning function: For plans that have not yet started or are in progress, start learning, view the learning content of the current stage, and start learning related courses, including elementary, intermediate, and advanced course information.

[0113] (4) Course learning function: a detailed interface for course learning according to the corresponding stage course, including all the courseware content of the course.

[0114] (5) Question and answer test function: After completing the course, you will enter the stage question and answer session, which will prompt you that you have completed the course for this stage. You will move to the question and answer center for question and answer test. After passing the question and answer test, you will enter the next stage of learning.

[0115] (6) Examination and testing function: After completing all stages of the course and passing the question-and-answer test, you will enter the examination and testing phase, which will prompt you that you have completed all the planned content and you will enter the examination center to take the examination.

[0116] (7) Query function: can screen training plans based on fuzzy matching conditions.

[0117] The course center unit is used to integrate all course resources on the platform, supports classification and filtering by name, type, stage, etc., and provides columns such as popular courses, boutique courses and personal courses. Students can complete the courses in the plan or independently expand the content outside the learning plan, and realize flexible learning through the course details page. The course center includes all course names, types, stages and other information. You can select the classification category to filter the courses. Click on the course to enter the learning details interface. The course center provides course query and filtering functions, specifically including popular courses, boutique courses, my courses and other columns. In addition to studying according to the plan assigned by the administrator, students can also study courses outside the plan on their own.

[0118] The functions of the course center unit include:

[0119] (1) Course List: Students can view all course information, including course name, course type, course category, stage, number of views, course status, course duration, release date, course credits, course rating, course stage, course progress, etc. Course types include videos and PPTs, course categories include job skills and experience feedback, and course stages are divided into beginner courses, intermediate courses, and advanced courses.

[0120] (2) My Courses function: You can check the courses you have participated in and the courses you have collected, including the courses you have planned and the courses you have studied independently.

[0121] (3) Popular courses function: Displays information of several courses with the highest number of views for users’ reference and reading.

[0122] (4) Latest course function: Combined with the release time of the course, several courses updated in the recent period are displayed for users' reference and reading.

[0123] (5) Course collection function: Save the collected courses in My Courses for viewing.

[0124] (6) Course details function: Enter the course details interface, which contains specific course content information.

[0125] (7) Query function: You can perform fuzzy queries on courses based on key fields such as course name and category to locate the specific courses you need to find.

[0126] The exam center unit manages the entire online exam process, allowing students to enter the exam interface with one click. Features like real-time timing and automatic answer saving ensure an efficient and standardized exam process. The system automatically records answer data for subsequent score analysis and archiving. Exams are conducted online. By clicking "Start Exam," candidates enter the online exam interface and begin answering. The system provides real-time timing and automatic answer saving to ensure a smooth exam process.

[0127] The functions of the test center unit include:

[0128] (1) Exam list function: displays information about the exams that students need to take, including exam number, exam name, exam type, exam duration, exam start date and time, exam end date and time, subject name, subject number, exam location, exam status, planned batch, batch number, total exam score, passing score, exam creator, subjective question score, objective question score, exam results, certification results, operation bar, etc. Exam status includes not started, in progress, and completed; certification results include pending review, passed, and failed; and the operation bar includes start exam, continue exam, and view exam paper.

[0129] (2) Start exam function: For an exam that has not started, click the "Join exam" button in the operation bar to participate in the exam and enter the corresponding answering interface, which contains multiple question types such as selection, judgment, and short answer.

[0130] (3) Continue test function: If you exit the test due to unexpected circumstances, you can click Continue test when you enter again and enter the corresponding answering interface to continue answering questions.

[0131] (4) Edit and save function: The modified answers can be automatically saved during the process of editing the test paper.

[0132] (5) Test paper submission function: After completing the test paper, click the Submit button to complete the test content. If you do not click Submit before the test countdown ends, the system will automatically submit and end the test.

[0133] (6) View test paper function: For completed test items, click the View Test Paper button in the operation bar to view the specific content and scores of the test paper.

[0134] (7) Query function: You can perform fuzzy query on exam items based on key fields such as exam name and time to locate the specific exam arrangement you need to take.

[0135] The Q&A Center centrally manages data from student-engaged Q&A modules, displaying detailed Q&A results and scores. This helps students identify gaps and provides a basis for transitioning to the next stage of learning. The Q&A Center includes all Q&A sessions participated in by students, centrally managing each phase of the training plan and allowing students to easily view their own Q&A results and scores.

[0136] The functions of the Q&A center unit include:

[0137] (1) Question and answer list function: displays the question and answer information of students, including the questioner, team, question type, question time, answerer, question content, question answer, answer time, answer status, master, apprentice, question and answer credits, plan name, plan number, batch name, batch number, course name, course stage, operation column, etc. The operation column includes the functions of re-editing and initiating questions.

[0138] (2) Start answering questions: Click the Enter Exam button in the operation bar after the exam item to participate in the exam and enter the corresponding answering interface, which contains multiple question types such as selection, judgment, and short answer.

[0139] (3) Edit and save function: The answers can be edited and modified, and the modified content can be automatically saved during the editing process.

[0140] (4) Submit Q&A function: After completing the question and answer, click the Submit button to complete the Q&A content.

[0141] (5) Questioning function: You can ask questions about the current Q&A content, edit the questions and submit them to the answerer for answering.

[0142] (6) Re-editing function: Before the administrator reviews the reply, you can edit and modify it freely. After the administrator reviews it, you cannot modify it.

[0143] (7) Query function: You can perform fuzzy queries on question and answer items based on key fields such as question and answer name, question time, etc.

[0144] The Data Center unit is used to store and display various file resources (such as Word, Excel, PPT, and compressed files) on the training platform. It supports hierarchical permission management, allowing students to browse or download them online based on their permissions, meeting diverse learning needs. The Data Center displays all the training platform's materials, including Word, Excel, PPT, and compressed files. Users can browse or download relevant materials based on their permissions.

[0145] The functions of the data center unit include:

[0146] (1) Data list function: You can view all data information, including courseware name, courseware type, courseware size, upload time, file category, creator, last update time, number of views, number of downloads, favorites, etc.

[0147] (2) Data details function: Click on the data to enter the data details interface, which contains specific data content information.

[0148] (3) Query function: You can perform fuzzy query on data based on key fields such as data name and data type.

[0149] (4) Most downloaded function: displays information about several materials ranked at the top in terms of download volume.

[0150] (5) Most viewed function: displays information of several materials with the highest number of views.

[0151] (6) Most collected data function: displays the information of the most collected data.

[0152] (7) My Information function: displays students’ collected information and information download history.

[0153] (8) Download function: Provides the function of downloading data. Click to download to the local device.

[0154] (9) Collection function: Provides the collection function of information. Click to add it to the favorites.

[0155] The training record unit is used to record the execution of training plans throughout the entire process, linking plan, course, and exam data through batch numbers. Trainees can view the progress of current plans or historical plan details and track corresponding exam scores, forming a complete closed-loop training management system. With the training plan as the core, the entire training cycle is recorded. By linking plan and exam information through batch numbers, trainees can easily view the details of each ongoing training plan, as well as the details of completed training plans and corresponding exam scores.

[0156] The functions of the training record unit include:

[0157] (1) Record list function: You can view the complete training plan and examination records you have participated in, including fields such as plan name, plan number, batch name, batch number, batch type, plan start time, plan end time, training credits, exam name, exam number, exam results, certification results, score effective time, score expiration time, and whether you missed the exam. Training credit calculation requires the combination of course credits and question-and-answer credits.

[0158] (2) Query function: You can perform fuzzy query on record items based on key fields such as plan name, plan number, test name, batch number, etc. to locate the training records you need to view.

[0159] like Figure 3 As shown, the student side mainly realizes the functions of training and reference, and its business process is as follows: First, the student enters the plan center, where he can view the details of the plan he participates in, and click on the plan that has not started or is in progress to start training. Each batch of plans is divided into three stages: elementary, intermediate and advanced, which correspond to the course learning in the elementary, intermediate and advanced stages respectively. After completing each stage of learning, a question-and-answer test will be conducted, and students can obtain credits after answering the questions; secondly, after completing the study, they can enter the examination link and take an online test in the examination center. If they fail the exam, they need to retake the exam; finally, after passing the exam, they can query the detailed information of the plan and the exam results associated with the plan in the training record.

[0160] In addition, students can view relevant course information in the course center, view all materials in the information center, view and modify the question and answer records generated during their training in the question and answer center, and can also raise questions.

[0161] The administrator side includes a plan management unit, a course management unit, an examination management unit, a marking center unit, a question and answer center unit, a data management unit, a question bank management unit, and a training record management unit.

[0162] The plan management unit is used to develop training plans that include basic information such as plan name / number / batch, as well as course / question / answer / exam requirements. Administrators can distribute plans to designated personnel to ensure the orderly progress of training activities and the adequacy of participant preparation. Administrators can manage all training plans and develop detailed training plans, including information such as plan name, plan number, batch number, and required course information, question and answer tests, and exam information. These plans can then be distributed to relevant personnel, helping to ensure the orderly progress of training activities and the adequacy of participant preparation.

[0163] The functions of the planning management unit include:

[0164] (1) Plan List Function: Administrators can manage all training plan lists and view the plan number, plan name, batch name, batch number, plan start time, plan end time, plan progress, plan introduction, plan details, plan status, and plan category of the current training in real time. Plan categories are divided into department-level training plans, section-level training plans, and team-level training plans. Plan status includes not started, in progress, and completed.

[0165] (2) Plan details function: Click on the plan details to display the progress of each student participating in the plan, including the plan name, plan type, batch number, student name, student type, learning progress, start time, end time and other information.

[0166] (3) New functions: Administrators can add training plans. Click to enter the new interface, fill in the plan details, check the phased courses of the plan, and the relevant trainees who will participate in the training plan. The training plan content is reasonably allocated according to the trainees' situations. At the same time, the course stage questions are set and the examination papers are compiled to form a complete training plan.

[0167] (4) Modification function: supports editing and modifying the published training plan, and can only make changes and adjustments to the trainees.

[0168] (5) Query function: Training plans can be screened based on fuzzy matching conditions.

[0169] (6) Deletion function: supports deletion of training plans.

[0170] (7) Batch deletion function: supports selecting multiple plans for batch deletion.

[0171] (8) Save function: supports saving the training plan currently being compiled.

[0172] (9) Publishing function: supports submitting and publishing the current training plan and pushing it to trainees who need to participate in the training.

[0173] The course management unit is used to create and edit course content, schedule teaching schedules, and monitor course progress and student completion rates in real time to ensure the precise achievement of teaching objectives. Administrators can manage training courses, including creating, editing, deleting, and scheduling them. They also need to monitor course progress and completion to ensure that training objectives are achieved.

[0174] The functions of the course management unit include:

[0175] (1) Course List: Administrators can manage all course information, including course name, course type, course category, stage, views, course status, course duration, release date, course credits, course rating, course stage, course progress, etc. Course types include videos, PPT, etc., course categories include job skills, experience feedback, etc., and course stages are divided into beginner courses, intermediate courses, and advanced courses.

[0176] (2) New functions: Administrators can add new courses. The specific functions should include course name, course type, course classification, course stage, and included courseware.

[0177] (3) Modification function: Modify and update existing course content.

[0178] (4) Query function: Courses can be filtered based on fuzzy matching conditions.

[0179] (5) Deletion function: supports deletion of courses.

[0180] (6) Batch deletion function: supports batch deletion of multiple courses.

[0181] (7) Save function: supports saving the course currently being edited.

[0182] (8) Publishing function: Administrators submit and publish courses to facilitate students’ learning.

[0183] (9) Statistics function: Statistics of popular courses and latest courses based on the number of views and release date.

[0184] The exam management unit is used to configure exam rules, intelligently generate exam papers, publish online exams, and manage scores. This standardized assessment process ensures fair evaluation and provides data support for quantifying training effectiveness. Administrators can set exam rules, generate exam papers, publish exams, and manage scores, helping to ensure exam fairness and accuracy and providing a basis for training effectiveness evaluation.

[0185] The functions of the course management unit include:

[0186] (1) Examination list function: Administrators can manage all examination information, including examination number, examination name, examination type, examination duration, examination start date and time, examination end date and time, subject name, subject number, examination location, examination status, planned batch, batch number and other information.

[0187] (2) Exam details function: The function displays the exam results of each student participating in the program, including the total score, passing score, exam creator, subjective question score, objective question score, exam results, certification results, and other information.

[0188] (3) New functions: Administrators can add new exams, edit relevant exam information, extract questions from the question bank, and add questions to form exam papers.

[0189] (4) Modification function: supports modification of exam information and test paper information that have not yet started.

[0190] (5) Query function: supports querying specific examination information through fuzzy matching.

[0191] (6) Deletion function: supports deletion of related test information and test paper information.

[0192] (7) Batch deletion function: supports batch deletion of multiple-choice test information and test paper information.

[0193] (8) Save function: supports editing of basic examination information and test paper information, and supports saving the basic examination information and test paper information currently being edited.

[0194] (9) Publishing function: supports submitting edited basic examination information and test paper information.

[0195] The Grading Center handles manual grading of subjective questions, supports administrators in multi-dimensional scoring of short-answer questions, and automatically generates comprehensive exam score reports. This allows administrators to review papers pending and reviewed, assign scores to short-answer questions, and generate scores for subjective questions.

[0196] The functions of the marking center unit include:

[0197] (1) Marking list function: Administrators can manage all marking information, including exam name, exam number, creator, exam time, plan number, plan name, batch number, batch name, number of papers to be marked, number of papers marked, and paper status. Paper status is divided into paper to be marked and paper marked.

[0198] (2) Scoring details function: query the detailed examination papers of all candidates in the current examination, including the specific scores.

[0199] (3) Query function: supports querying relevant marking action information, and can be filtered according to the test number and test name.

[0200] (4) Start review function: supports manual review and scoring of the short-answer question module of the test paper.

[0201] (5) Modification function: supports modification of scores before submission.

[0202] (6) Save function: supports automatic saving of scoring information.

[0203] (7) Submission function: supports submission of graded test papers. After submission, the test paper status changes to reviewed.

[0204] (8) Delete function: supports deleting marking items.

[0205] (9) Batch deletion function: supports batch deletion of marking items.

[0206] The Q&A verification unit is used to assess learning outcomes at specific stages. By setting relevant questions and providing scoring and feedback, students can identify and address gaps in their learning. This unit provides a Q&A function, allowing administrators to verify learning outcomes at specific stages of the planned course, provide scoring for responses, and provide feedback to students.

[0207] The functions of the Q&A center unit include:

[0208] (1) Question and answer list function: Administrators can manage all question and answer information, including the questioner, team, question type, question time, answerer, question content, question answer, answer time, answer status, master, apprentice, question and answer credits, plan name, plan number, batch name, batch number, course name, course stage and other information.

[0209] (2) Question and answer details function: After clicking on the details, the three-stage question and answer situation of the current student will be displayed, including specific questions and answers.

[0210] (3) Query function: supports querying relevant Q&A action item information, and can be filtered based on plan number, question content, questioner and other information.

[0211] (4) Start the Q&A function: support answering questions and editing relevant answer information.

[0212] (5) Question and answer scoring function: supports scoring and evaluating students' answers to form question and answer credits.

[0213] (6) Modification function: supports modification of answer content and score before submission.

[0214] (7) Save function: supports automatic saving of answer content and scoring information.

[0215] (8) Submission function: supports submission of questions that have been answered. After submission, the answer status changes to answered.

[0216] (9) Delete function: supports deleting question and answer items.

[0217] (10) Batch deletion function: supports batch deletion of question and answer items.

[0218] The resource management unit is used to centrally manage training materials such as videos, documents, and PowerPoint presentations, establish a structured knowledge base, and ensure the integrity and usability of teaching resources. Administrators are responsible for managing training resources such as training documents, videos, and PowerPoint presentations, ensuring their comprehensiveness and smooth progress.

[0219] The functions of the resource management unit include:

[0220] (1) Data List: Administrators can manage all data information, including courseware name, courseware type, courseware size, upload time, file category, creator, last update time, number of views, downloads, and number of favorites. It supports categorization based on courseware type. File categories include video, PDF, Word, Excel, PPT, and other formats.

[0221] (2) Material details function: View the details of the material, including courseware content information in video, PPT and other formats.

[0222] (3) New function: support for adding new data information.

[0223] (4) Modification function: supports editing and modifying courseware information.

[0224] (5) Query function: Courseware can be filtered based on fuzzy matching conditions.

[0225] (6) Deletion function: supports courseware deletion.

[0226] (7) Batch deletion function: supports batch deletion of multiple selected courseware.

[0227] (8) Save function: supports automatic saving of the courseware information currently being modified.

[0228] (9) Publishing function: supports courseware deletion.

[0229] (10) Upload function: supports batch deletion of multiple selected courseware.

[0230] (11) Download function: supports local download of courseware.

[0231] (12) Statistics function: Statistics the most viewed, downloaded and collected courseware based on the number of views, downloads and collections.

[0232] The intelligent question bank maintains a variety of test resources, including multiple-choice, true-or-false, and short-answer questions. It supports categorized management by specialized fields, such as DCS operation and maintenance and inspection, to meet the needs of multi-scenario assessments. Administrators can add, edit, delete, and categorize test questions and question banks based on training content and requirements. The test supports a variety of question types, including multiple-choice, true-or-false, and short-answer questions, to meet the needs of different assessment scenarios. The question bank supports a variety of types, including DCS maintenance, operation and maintenance, and inspection, to meet the needs of different assessment scenarios.

[0233] The functions of the intelligent question bank unit include:

[0234] (1) Question list function: manage all question-related information, including question name, question code, question type, question difficulty, creation time, creator, etc.

[0235] (2) Question details function: supports viewing the specific content of the test questions.

[0236] (3) New function: Click Add to enter the secondary interface and write the specific content of the test question, including the test content, test name, test type (single choice, multiple choice, judgment, short answer), and fill in the question bank to which the test question belongs.

[0237] (4) Modification function: supports editing and modifying test questions.

[0238] (5) Query function: supports fuzzy query of information related to test questions, and filters according to test question name, test question type, etc.

[0239] (6) Delete function: After modification, the question information in the question bank will also be deleted synchronously.

[0240] (7) Batch deletion function: supports batch deletion of multiple test questions.

[0241] (8) Save function: supports saving the test questions being edited.

[0242] (9) Import function: supports importing existing test questions.

[0243] (10) Question bank list function: manage all question bank related information, including question bank name, question bank code, number of question types, question bank description, question bank type, etc.

[0244] (11) Question bank details function: supports viewing the distribution of question types in the question bank and information such as question titles, answers, etc.

[0245] (12) New function: Click Add to enter the secondary interface to complete the question bank information. After creating a blank question bank, click Question Management in the operation bar to add questions to the question bank.

[0246] (13) Modification function: supports editing and modifying the question bank.

[0247] (14) Query function: Fuzzy query matching based on question bank name, type and other conditions.

[0248] (15) Deletion function: supports deletion of question bank information.

[0249] (16) Batch deletion function: supports batch deletion of question bank information.

[0250] (17) Save function: supports saving the question bank being edited.

[0251] (18) Import function: supports importing existing local question banks.

[0252] The training record management unit comprehensively records training time, content, participation, and assessment results, providing multi-dimensional search and visual analysis capabilities for traceable management of the training process. The training record management unit aggregates all training plan records, including training time, content, participants, and training outcomes, improving information management efficiency and enabling managers to more easily access and query relevant information.

[0253] The functions of the training record management unit include:

[0254] (1) Record list function: Administrators can view all training plans and examination records, including plan name, plan number, batch name, batch number, batch type, plan start time, plan end time, training credits, examination name, examination number, examination results, certification results, result effective time, result expiration time, whether the examination was absent, and examination candidate field information.

[0255] (2) Record details function: The details include all the information of the trainees who participated in the training program, including the trainees’ names, grades, training scores, etc.

[0256] (3) Query function: Fuzzy query matching based on conditions such as plan name, plan number, batch name, batch number, exam name, and exam number.

[0257] (4) Deletion function: supports deletion of training records.

[0258] (5) Batch deletion function: supports batch deletion of training records.

[0259] like Figure 4 As shown, the administrator side primarily manages plans, courses, exams, materials, and question banks, as well as grading and question-answering functions. The main business processes for plan management are as follows: First, administrators can create training plans in the plan management module, setting information such as the plan name, included phased courses, phased questions, and trainees; second, if this plan requires an assessment, select the exam papers prepared in the exam management module and associate them with this training; finally, publish the training plan. If no assessment is required, the plan can be published directly.

[0260] In addition, administrators can also add new courses in the course center, and publish a complete set of courses for students to view and study by editing course types, course stages, uploading courseware, etc.; through material management, all courseware materials can be operated; in the question bank management, test questions and question bank information can be edited and imported; in the marking center, administrators can view the items to be marked and the items that have been marked, and score the subjective questions of different students' test papers; in the question and answer center, administrators can view the items to be answered and the items that have been answered, and score according to the students' specific questions and answers; in the training record management, the training records of all students can be viewed to exercise macro control over the entire training.

[0261] The service layer provides a range of common services to support the development, operation, and expansion of upper-layer applications, such as identity authentication, permission management, logging, message notification, inter-system communication, security assurance, data processing, and more. These services ensure the security, reliability, and scalability of the system and facilitate the integration of different business functions.

[0262] The foundational support layer provides the necessary infrastructure for the entire platform, offering high-performance hardware resources such as computers, servers, and storage devices. It is responsible for open interface communications, data management and storage, computing resource allocation, and network security. This layer not only ensures the normal operation of the system but also provides strong support for upper-layer applications and services, enabling the training platform to flexibly respond to a variety of complex business scenarios and technical challenges.

[0263] The edge layer mainly involves the collaborative call processing capabilities of the on-site training resource library, and designs data collection, processing and transmission, including data collectors, smart sensors and other edge computing devices, aiming to reduce data transmission delays and improve local decision-making efficiency.

[0264] This platform has:

[0265] (1) Security: Ensure that only authorized users can access and operate data.

[0266] (2) Scalability: Support the scalability of training programs to cope with the increase in training content and functions and ensure the performance of the system.

[0267] (3) Ease of use: The training management module needs to be easy to use and understand, with reduced technical requirements to meet the needs of business personnel.

[0268] (4) Fault tolerance: The platform needs to have an automatic recovery function. When the training report editing is abnormal, it can provide an alarm message and automatically save the current input information to maintain the stable operation of the system.

[0269] (5) Reliability: The training management module is designed and developed using relatively mature technologies to avoid internal failures of the development platform.

[0270] This application combines the advantages of large-scale model generalization capabilities with domain expertise to build a highly intelligent and responsive nuclear power plant DCS training platform. The platform has the ability to self-learn and evolve, continuously improving teaching quality over time. Trainees can deepen their understanding of complex concepts through interactive learning, effectively shortening the training cycle and comprehensively improving the professional skills and safety awareness of nuclear power plant DCS operation and maintenance personnel. By establishing a feedback loop mechanism and continuously updating and improving the content of the knowledge base, the model becomes more professional and the prediction accuracy is continuously improved. The main technologies used are dynamic updating and maintenance of the domain knowledge base, dynamic model adaptation and fine-tuning, interactive question-and-answer data enhancement, and a loop feedback spiral mechanism.

[0271] This application enables comprehensive and efficient management of training. This module can implement multiple functions such as training plans, training courses, exams, Q&A, training materials, training records, and training question banks. Through this module, students can conduct training efficiently, while administrators can grasp the training dynamics of each student, comprehensively improving the refinement and standardization of training management.

[0272] In addition, the present application also provides an electronic device including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the above-mentioned knowledge question-and-answer-based nuclear power plant DCS intelligent training method is implemented.

[0273] In addition, the present application also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed, the above-mentioned knowledge question-and-answer-based nuclear power plant DCS intelligent training method is implemented.

[0274] Computer-readable media include permanent and non-permanent, removable and non-removable media that can store information using any method or technology. The information can be computer-readable instructions, data structures, program modules or other data.

[0275] The above description is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed in this application should be covered by the scope of protection of the present application.

Claims

1. A knowledge question-answering-based intelligent training method for DCS in nuclear power plants, characterized in that: include: Step 1: Collect nuclear power plant DCS training materials, operation manuals, and fault case data to build an original database covering theory, operation, and safety specifications; Step 2: Clean and standardize the data to generate a structured training set and domain corpus; Step 3: Perform two-stage fine-tuning based on the DeepSeek model, freeze the parameters of the MLA layer and MoE layer and adjust the classification head to complete general domain adaptation. Then, insert the low-rank matrix through the LoRA adapter to optimize the top attention layer and MLP layer; Step 4: Build a question-answer knowledge base based on structured question-answer pairs, and use multi-round dialogue expansion and domain hint annotation for data enhancement; Step 5: Integrate intelligent retrieval, question-answering engine, and personalized recommendation modules to build a training platform and generate a simulated operation interface with a real-time question-answering window; Step 6: Establish a feedback loop mechanism to dynamically maintain the knowledge question and answer database, and trigger knowledge base correction and model iterative training by having experts mark incorrect answers.

2. The method for intelligent training of nuclear power plant DCS based on knowledge question and answer according to claim 1, characterized in that: In step 3, the strategy of freezing the bottom layer and fine-tuning the top layer is adopted. The general model is fine-tuned using common data in the nuclear power DCS field. The parameters of the MLA layer and MoE layer of the DeepSeek model are frozen, the bottom-level general language features are retained, and only the last layer of classification head is adjusted and replaced.

3. The method for intelligent training of nuclear power plant DCS based on knowledge question and answer according to claim 1, characterized in that: In step 4, the LoRA adapter is used to fine-tune the model in step 3. A low-rank matrix is inserted into the top-level attention layer and MLP layer to adjust the model output, and a Dropout layer is added to prevent the model from overfitting. The low-parameter characteristics of LoRA are used to adapt to small amounts of structured data, while the bottom layer is frozen to ensure general knowledge, forming a lightweight fine-tuning paradigm that collaborates with general and domain knowledge.

4. The method for intelligent training of nuclear power plant DCS based on knowledge question and answer according to claim 1, characterized in that: In step 4, the general model is initially fine-tuned using basic domain knowledge to obtain a domain generalization model, and then the domain generalization model is fine-tuned again based on the structured data of professional knowledge questions and answers to obtain a domain enhancement model.

5. A knowledge question-answering-based DCS intelligent training platform for nuclear power plants, characterized by: include: The application layer provides a user interaction portal and a personalized operation interface, supporting access to training courses, simulation operations, online examinations, and document review. The business function layer, including the student side and the administrator side, is used to implement training plan formulation, course arrangement, examination management, question bank maintenance, question and answer interaction, and training record tracing; The service layer is used to provide identity authentication, permission control, log auditing, message push, security protection and load balancing services; The basic support layer is used to implement data governance, interface communication and network security protection through server clusters, storage devices and operating systems; The edge layer is used to complete real-time data collection, preprocessing and low-latency transmission through data collectors and smart sensors, supporting localized decision-making and closed-loop control.

6. The nuclear power plant DCS intelligent training platform based on knowledge question and answer according to claim 5 is characterized in that: The business function layer includes: Dynamic course generation module, used to automatically adjust the training difficulty based on the trainees' operation records; A multimodal question-answering engine that supports multi-dimensional knowledge interaction using text, voice, and diagrams.

7. The nuclear power plant DCS intelligent training platform based on knowledge question and answer according to claim 5 is characterized in that: The service layer includes: Operation playback service module, used to record and reconstruct the trainees' simulated operation process; The abnormal behavior detection module is used to identify illegal operation behaviors based on the preset rule base.

8. The nuclear power plant DCS intelligent training platform based on knowledge question and answer according to claim 5 is characterized in that: The edge layer includes: Industrial protocol adapter module for compatibility with Modbus, PROFIBUS and OPC UA communication protocols; The local cache database is used to store the training operation data of the last 24 hours.

9. The nuclear power plant DCS intelligent training platform based on knowledge question and answer according to claim 5 is characterized in that: The student side includes: The planning center unit is used to help students learn in stages according to the training plan assigned by the administrator, test their learning results through the interweaving of courses and question-and-answer modules, and dynamically adjust their learning progress; The course center unit is used to integrate all course resources on the platform and supports classification and screening; The examination center unit is used for the management of the entire online examination process; The Q&A center unit is used to centrally manage the data of the phased Q&A modules participated by students and display the Q&A results and score details; The data center unit is used to store and display various file resources of the training platform and supports hierarchical authority management; The training record unit is used to record the execution of the training plan throughout the entire process and associate the plan, course, and examination data through batch numbers.

10. The nuclear power plant DCS intelligent training platform based on knowledge question and answer according to claim 5 is characterized in that: The administrator side includes: The plan management unit is used to develop training plans that include basic information and course / question and answer / exam requirements, and supports administrators to publish plans to designated personnel; The course management unit is used to create and edit course content, arrange teaching schedules, and monitor course progress and student completion rates in real time; The exam management unit is used to configure exam rules, intelligently generate exam papers, publish online exams, and manage scores; The marking center unit is used to handle the manual marking of subjective questions, support administrators to perform multi-dimensional scoring on short-answer questions, and automatically summarize and generate comprehensive test score reports; Question and answer test unit, used for periodic learning achievement assessment, by setting course-related questions and implementing scoring feedback; Resource management unit, used to centrally manage training materials and establish a structured knowledge base system; Intelligent question bank unit, used to maintain question resources and support classification management in professional fields; Training record management unit, used for full record data, multi-dimensional retrieval and visual analysis.

Images