Knowledge-driven multi-agent collaborative reactor scheme demonstration system, design method, medium and equipment thereof

Abstract

The application relates to a knowledge-driven multi-agent cooperative reactor scheme demonstration system and a design method, medium and equipment thereof, the system comprising: a knowledge internalization unit for converting unstructured and semi-structured documents into a dynamic knowledge source which can be queried and utilized by a model in real time; a knowledge structuring unit for extracting core entities, relationships and attributes from the knowledge source and constructing a professional knowledge graph of a reactor design field; and a knowledge application unit for constructing intelligent agents and a collaborative working mechanism of multi-professional intelligent agents based on the knowledge source and the knowledge graph, and constructing an intelligent question and answer interface. Through automatic knowledge management and intelligent agent collaborative work, the application significantly reduces the time and effort of manual intervention and improves the efficiency of reactor scheme demonstration.

Description

Technical Field

[0001] This invention relates to reactor design technology, specifically to a knowledge-driven multi-agent collaborative reactor scheme demonstration system and its design method, medium, and equipment. Background Technology

[0002] Large-scale modeling technology has cross-domain generalization capabilities in natural language understanding and generation, but it still struggles to reach a professional level in specific fields. The nuclear industry is characterized by its technological complexity, long time span, and information sensitivity. Nuclear engineering scheme demonstration is a highly complex, knowledge-intensive, and deeply interdisciplinary engineering problem, and its decision-making process integrates knowledge from multiple fields such as neutron physics, thermal hydraulics, fuel materials, safety analysis, and economics.

[0003] Traditional feasibility studies rely heavily on the experience and intuition of experts from different fields, requiring multiple iterations by designers from various disciplines before finalization. Designers must expend significant time and effort manually retrieving, extracting, integrating, and verifying the necessary, fragmented knowledge from numerous technical reports, design guidelines, historical schemes, and research papers. This human-dominated knowledge management and application model has become a bottleneck restricting the improvement of reactor feasibility studies and design efficiency. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the prior art by providing a knowledge-driven multi-agent collaborative reactor scheme demonstration system and its design method, medium and equipment, so as to realize multi-dimensional evaluation and iterative optimization of the complex problem of nuclear engineering scheme demonstration, which involves close coupling of multiple disciplines.

[0005] To achieve the above objectives, in one aspect, embodiments of the present invention provide a knowledge-driven multi-agent collaborative reactor scheme demonstration system, comprising:

[0006] Knowledge internalization units are used to transform unstructured and semi-structured documents into dynamic knowledge sources that can be queried and utilized by models in real time.

[0007] Knowledge structuring units are used to extract core entities, relationships, and attributes from the knowledge source to construct a professional knowledge graph in the reactor design domain.

[0008] The knowledge application unit, based on the knowledge source and knowledge graph, constructs intelligent agents and collaborative working mechanisms for multiple professional intelligent agents for different professions, and builds an intelligent question-and-answer interface.

[0009] Furthermore, in a specific embodiment, in the knowledge-driven multi-agent collaborative reactor scheme demonstration system described above, the knowledge internalization unit includes:

[0010] The data collection module is used to acquire professional literature, design specifications, and calculation models in the process of reactor design, analysis, and safety assessment.

[0011] The data preprocessing module is used to clean the collected data, remove irrelevant content, and convert all documents into a lightweight markup language format using a large visual language model.

[0012] The vector database building module is used to split and vectorize data, store vectorized data in a database environment, and build indexes to support efficient retrieval operations.

[0013] Furthermore, in a specific embodiment, the knowledge-driven multi-agent collaborative reactor scheme demonstration system described above, wherein the knowledge structuring unit includes:

[0014] The entity recognition module is used to perform entity recognition on text data by leveraging the semantic understanding and logical reasoning capabilities of large models, and extract core entities as nodes of the knowledge graph; and combined with the ontology dictionary and context-aware mechanism in the field of nuclear engineering, it performs fine-grained classification and disambiguation of professional terms in the field of nuclear engineering to ensure the consistency and accuracy of node types.

[0015] The relation extraction module is used to extract semantic relationships between entities from text data using natural language processing techniques, which are then used as edges in the knowledge graph. The large model identifies relationships with engineering semantics through prompt engineering and few-shot learning strategies, and maps them to predefined relation patterns to ensure the logical rigor of the knowledge graph.

[0016] The attribute extraction module is used to extract the attribute information of the entity and the relationships between different attributes;

[0017] The graph construction module is used to organize the extracted entities, relations and attributes into a knowledge graph with a specific structure. The system optimizes the graph structure of the initially generated triples, including entity alignment, redundancy resolution, contradiction detection and confidence weighting, and finally persists the high-quality knowledge graph in the graph database.

[0018] Furthermore, in a specific embodiment, the knowledge-driven multi-agent collaborative reactor scheme demonstration system described above, wherein the knowledge application unit includes:

[0019] The prompt word design module is used to initialize multi-disciplinary intelligent agents, enabling them to possess knowledge and skills in specific fields; the multi-disciplinary intelligent agents include systems engineering intelligent agents, neutron physics intelligent agents, thermal hydraulics intelligent agents, fuel materials intelligent agents, and safety analysis intelligent agents;

[0020] The agent building module is used to design the basic architecture of the agent, enabling the agent to extract key information, decompose tasks, and call various tools and resources to complete specific tasks.

[0021] The collaborative work mechanism construction module is used to establish a collaborative work mechanism among multiple professional intelligent agents through a multi-agent communication protocol.

[0022] The intelligent question-and-answer interface building module is used to build the final intelligent question-and-answer interface, allowing users to input questions through the interface, and the big model to generate answers based on the retrieved content and return them to the user through the interface.

[0023] Furthermore, the basic architecture of the intelligent agent includes:

[0024] The perception module is used to analyze external input and extract key information;

[0025] The decision-making module is used to formulate task execution strategies, decompose tasks, and assign the decomposed subtasks to intelligent agents or tools of different specialties.

[0026] The execution module is used to coordinate the work of various professional intelligent agents and call on various tools and resources to complete specific tasks.

[0027] Furthermore, among the multi-disciplinary intelligent agents, the systems engineering agent is responsible for requirement identification and task decomposition, transforming the requirements described in natural language into explicit engineering requirements; the neutron physics agent is responsible for reactor physics analysis, including reactor core power, critical dimensions, and neutron flux distribution range; the thermal-hydraulic agent is responsible for reactor thermal-hydraulic analysis, including reactor inlet and outlet temperature changes, coolant medium selection, and flow rate settings; the fuel and material agent is responsible for selecting reactor fuel and material materials, fuel enrichment, and irradiation damage thresholds that are compatible with core power and temperature; and the safety analysis agent is responsible for verifying the proposed solutions, analyzing from a safety perspective whether the solutions can meet the safety requirements of each specialty and the overall safety threshold requirements. The system adopts a single-agent execution and multi-agent iterative interaction approach to achieve a closed loop of solution proposal and verification.

[0028] On the other hand, embodiments of the present invention provide a design method for the above-mentioned knowledge-driven multi-agent collaborative reactor scheme demonstration system, comprising the following steps:

[0029] S1) Transform unstructured and semi-structured documents into dynamic knowledge sources that can be queried and utilized by the model in real time; including data collection, data preprocessing, and vector database construction;

[0030] S2) Extract core entities, relationships, and attributes from the knowledge source to construct a professional knowledge graph for reactor design.

[0031] S3) Based on the knowledge source and knowledge graph, construct intelligent agents and collaborative working mechanisms of multi-professional intelligent agents for different professions, and construct an intelligent question-and-answer interface.

[0032] Furthermore, in a specific embodiment, the data collection in step S1) of the design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system described above includes acquiring professional literature, design specifications and calculation models from multiple fields such as neutron physics, thermal hydraulics, fuel materials, and safety analysis.

[0033] Data preprocessing includes cleaning the collected data, removing irrelevant content, and converting all documents into a lightweight markup language format using a large visual language model.

[0034] The construction of a vector database includes splitting and vectorizing the data, storing the vectorized data in a database environment, and building indexes to support efficient retrieval operations.

[0035] Furthermore, the vector database construction described in step S1) includes the following steps:

[0036] S11) Data splitting: breaking long text data into smaller paragraphs or sentences;

[0037] S12) Vectorization processing: The pre-trained large model is used to vectorize the split text segments to generate high-dimensional vector representations.

[0038] S13) Data entry: The vectorized data is stored in the database, and an index is created to support efficient retrieval operations.

[0039] Furthermore, in a specific embodiment, the design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system described above, step S2) specifically includes:

[0040] S21) Utilize the semantic understanding and logical reasoning capabilities of large models to perform entity recognition on text data, extract core entities as nodes of the knowledge graph; and combine the ontology dictionary and context-aware mechanism in the field of nuclear engineering to perform fine-grained classification and disambiguation of professional terms in the field of nuclear engineering, ensuring the consistency and accuracy of node types.

[0041] S22) Using natural language processing technology, semantic relationships between entities are extracted from text as edges of the knowledge graph; the large model identifies relationships with engineering semantics through cue engineering and few-shot learning strategies, and maps them to predefined relationship patterns to ensure the logical rigor of the knowledge graph.

[0042] S23) Extract the attribute information of the entity and extract the relationships between different attributes;

[0043] S24) The extracted entities, relations and attributes are organized into a knowledge graph in the form of a Resource Description Framework (RDF), with each entity as a node, relations as edges and attributes as attributes of the nodes. The system optimizes the graph structure of the initially generated triples, including entity alignment, redundancy resolution, contradiction detection and confidence weighting, and finally persists the high-quality knowledge graph in the graph database.

[0044] Furthermore, the core entities mentioned in step S21) include: component name, material type, and physical parameters.

[0045] Furthermore, in a specific embodiment, the design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system described above, step S3) specifically includes:

[0046] S31) Based on the characteristics of each major, a special prompt word template was designed to initialize the multi-major intelligent agent and dynamically adjust the prompt words according to the user's input and feedback;

[0047] S32) Design the basic architecture of the intelligent agent so that the intelligent agent can extract key information, decompose tasks, and call various tools and resources to complete specific tasks;

[0048] S33) Establish a collaborative working mechanism among multiple professional intelligent agents through a multi-agent communication protocol;

[0049] S34) Construct the final intelligent question-answering interface, allowing users to input questions through the interface, and the large model to generate answers based on the retrieved content, which are then returned to the user through the interface.

[0050] Furthermore, the multi-disciplinary intelligent agents mentioned in step S31) include a systems engineering intelligent agent, a neutron physics intelligent agent, a thermal-hydraulic intelligent agent, a fuel and materials intelligent agent, and a safety analysis intelligent agent. The systems engineering intelligent agent is responsible for requirement identification and task decomposition, transforming natural language descriptions of requirements into explicit engineering requirements. The neutron physics intelligent agent is responsible for reactor physics analysis, including reactor core power, critical dimensions, and neutron flux distribution range. The thermal-hydraulic intelligent agent is responsible for reactor thermal-hydraulic analysis, including reactor inlet and outlet temperature changes, coolant medium selection, and flow rate settings. The fuel and materials intelligent agent is responsible for selecting reactor fuel and materials suitable for core power and temperature, fuel enrichment, and irradiation damage thresholds. The safety analysis intelligent agent is responsible for verifying the proposed solutions, analyzing from a safety perspective whether the solutions meet the safety requirements of each specialty and the overall safety threshold requirements. The system adopts a single-agent execution and multi-agent iterative interaction approach to achieve a closed loop of solution proposal and verification.

[0051] Furthermore, step S33) establishes a collaborative working mechanism among multiple professional intelligent agents, specifically including:

[0052] Message passing: Each agent processes specific sub-tasks based on its own expertise and skills, and passes the results to other agents through the message passing mechanism. Multiple agents work together to complete the complex reactor demonstration task.

[0053] Task scheduling is carried out by a systems engineering professional intelligent agent to design task scheduling, rationally allocate tasks, and coordinate the work of various professional intelligent agents.

[0054] Collaborative optimization enables efficient processing and optimization of complex tasks through the collaborative work of multiple specialized intelligent agents.

[0055] On the other hand, the present invention further provides a storage medium storing a computer program, which is executed by a processor to implement the above-mentioned knowledge-driven multi-agent collaborative reactor scheme demonstration system.

[0056] Meanwhile, the present invention also provides a device including a storage medium and a processor, wherein the storage medium stores a computer program, and the processor executes the computer program to realize the above-mentioned knowledge-driven multi-agent collaborative reactor scheme demonstration system.

[0057] The beneficial effects of this invention are as follows: The knowledge-driven multi-agent collaborative reactor scheme demonstration system provided by this invention significantly reduces the time and effort required for manual intervention and improves the efficiency of reactor scheme demonstration through automated knowledge management and agent collaboration. This invention combines large models, RAG (Research Aggregate Algorithm), and knowledge graphs to ensure that every key point is verifiable, improving the accuracy and reliability of scheme demonstration. By supporting cross-domain collaborative work through a multi-agent system, it promotes knowledge sharing and exchange among different disciplines. This invention significantly outperforms the deep thinking and network search capabilities of large models on test sets of complex, multi-disciplinary coupled reactor problems. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a schematic diagram of the reactor scheme demonstration system for knowledge-driven multi-agent collaboration in a specific embodiment of the present invention;

[0060] Figure 2 and Figure 3 This is a schematic diagram of the distribution of data document topics collected by the knowledge internalization unit in a specific embodiment of the present invention;

[0061] Figure 4 This is a schematic diagram of a knowledge graph constructed using the Neo4j platform in a specific embodiment of the present invention;

[0062] Figure 5 This is a flowchart illustrating the multi-specialty intelligent agent workflow in a reactor according to a specific embodiment of the present invention.

[0063] Figure 6 This is a schematic diagram of the intelligent question-and-answer interface in a specific embodiment of the present invention;

[0064] Figure 7 This is an example of the final question-and-answer result in a specific embodiment of the present invention;

[0065] Figure 8 This diagram illustrates the evaluation results of a multi-specialty collaborative intelligent agent compared to a general large model (DeepSeek) and a networked search combining a general large model, as shown in a specific embodiment of the present invention. Detailed Implementation

[0066] 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.

[0067] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0068] The terms “comprising”, “including”, etc., as used herein indicate the presence of the steps, features, operations, or components, but do not preclude the addition of one or more other steps, features, operations, or components.

[0069] The design, analysis, and safety assessment of nuclear reactors involve a vast amount of professional literature, complex design specifications, and sophisticated computational models. This invention, tailored to the characteristics of the nuclear engineering field, integrates specialized knowledge and large-scale models to propose a novel knowledge-driven, multi-agent collaborative reactor scheme demonstration model. It constructs a reactor collaborative scheme design system that integrates extensive professional knowledge and possesses deep reasoning capabilities. By fusing three major technologies—Retrieval-Augmented Generation (RAG), Knowledge Graph (KG), and Multi-Agent System—a multi-disciplinary intelligent agent collaborative design mechanism is built. This provides intelligent design and analysis capabilities for multiple reactor domains, enabling multi-dimensional evaluation and iterative optimization of the complex problem of nuclear engineering scheme demonstration, which involves close coupling of multiple disciplines.

[0070] like Figure 1 As shown, the present invention provides a knowledge-driven multi-agent collaborative reactor scheme demonstration system, comprising:

[0071] Knowledge internalization units are used to transform unstructured and semi-structured documents into dynamic knowledge sources that can be queried and utilized by models in real time.

[0072] Knowledge structuring units are used to extract core entities, relationships, and attributes from the knowledge source to construct a professional knowledge graph in the reactor design domain.

[0073] The knowledge application unit, based on the knowledge source and knowledge graph, constructs intelligent agents and collaborative working mechanisms for multiple professional intelligent agents for different professions, and builds an intelligent question-and-answer interface.

[0074] 1. Knowledge Internalization Unit

[0075] The knowledge internalization unit includes: a data collection module, a data preprocessing module, and a vector database construction module.

[0076] 1.1 Data Collection Module

[0077] The design, analysis, and safety assessment of nuclear reactors involve a vast amount of specialized literature, complex design specifications, and sophisticated computational models. To construct a collaborative reactor design system that integrates extensive professional knowledge and possesses deep reasoning capabilities, data collection is the first step. The literature collected in this invention covers a wide range of topics, including neutron physics, thermal hydraulics, fuel materials, and safety analysis, ensuring the comprehensiveness and diversity of the data.

[0078] 1.2 Data Preprocessing Module

[0079] First, the collected documents and standards underwent initial cleaning to remove irrelevant content such as headers, footers, and footnotes. Next, the Vision Language Model (VLM) was used to convert all documents into a lightweight markup language (Markdown) format. This process not only simplified the file format but also effectively extracted images, formulas, charts, and other content from the files, facilitating subsequent reading and processing. Multimodal recognition technology ensured the integrity of text and image information, laying a solid foundation for subsequent vectorization processing and knowledge graph construction.

[0080] 1.3 Vector Database Construction Module

[0081] After data preprocessing, a database environment was deployed to ensure consistency and portability. Database connectivity and management ensured data security and stability. Next, the text data was split, vectorized, stored in the database, and retrieved. The specific steps are as follows:

[0082] (1) Data splitting: The long text data is split into smaller paragraphs or sentences to facilitate subsequent vectorization processing. This step ensures that the length of each paragraph or sentence is appropriate for model processing and storage.

[0083] (2) Vectorization: The pre-trained large model is used to vectorize the split text segments to generate high-dimensional vector representations. These vectors will be used for subsequent similarity search and knowledge graph construction.

[0084] (3) Data storage: The vectorized data is stored in the database and an index is created to support efficient retrieval operations. This step ensures that the data can be queried and accessed quickly.

[0085] Ultimately, a large amount of vectorized data was successfully stored in the database, and an efficient index was established, providing reliable data support for subsequent knowledge graph construction and intelligent question answering.

[0086] 2. Knowledge Structured Units

[0087] Building upon knowledge internalization, further knowledge structuring is achieved through an automated knowledge graph generation pipeline based on a large-scale model's logical capabilities. This pipeline automatically identifies and generates professional knowledge graphs for the reactor design domain. Leveraging the powerful semantic understanding and logical reasoning capabilities of the large-scale model, key entities (i.e., "nodes" in the knowledge graph) are automatically identified from massive amounts of unstructured text data (such as technical manuals, design specifications, accident reports, research papers, and operation logs). The semantic relationships (i.e., "edges") between these entities are then precisely extracted, thereby constructing a structured knowledge network for the reactor design domain. The knowledge structuring units include: an entity identification module, a relationship extraction module, an attribute extraction module, and a graph construction module.

[0088] 2.1 Entity Recognition Module

[0089] By leveraging the semantic understanding and logical reasoning capabilities of large-scale models, entity recognition is performed on text data, extracting core entities as nodes in the knowledge graph, such as component names, material types, and physical parameters. Furthermore, by combining an ontology dictionary and context-aware mechanisms in the field of nuclear engineering, fine-grained classification and disambiguation are performed on professional terms in the field of nuclear engineering, such as "pressurized water reactor," "control rods," "moderators," "neutron flux," "fuel cladding," and "LOCA accident." This process ensures the accuracy and comprehensiveness of entity (node ​​type) identification.

[0090] 2.2 Relationship Extraction Module

[0091] Natural language processing techniques are used to extract semantic relationships between entities from text, which are then used as edges in the knowledge graph, such as "Component A is made of material B" and "Component C is located at location D". The large model employs prompt engineering and few-shot learning strategies to identify relationships with engineering semantics, such as "contains", "causes", "influences", "depends on", "used for cooling", and "occurs at", and maps them to predefined relationship patterns to ensure the logical rigor of the knowledge graph. This process not only extracts explicit relationships but also infers implicit relationships through context, ensuring the richness and accuracy of the relationships.

[0092] 2.3 Attribute Extraction Module

[0093] Extracting entity attribute information, such as size, temperature, and pressure, and identifying relationships between different attributes. This process ensures a detailed description of each entity, laying the foundation for subsequent deep reasoning and complex queries.

[0094] 2.4 Map Construction Module

[0095] The extracted entities, relations, and attributes are organized into a knowledge graph in the form of a Resource Description Framework (RDF). Each entity is a node, relations are edges, and attributes are attributes of the nodes. This process ensures that the knowledge graph has a clear structure, is easy to query, and can be expanded. The system optimizes the graph structure of the initially generated triples (subject-relation-object), including entity alignment, redundancy resolution, contradiction detection, and confidence weighting. Finally, the high-quality knowledge graph is persistently stored in a graph database (such as Neo4j or Nebula Graph), supporting efficient querying and dynamic updates.

[0096] Through the extraction and construction of knowledge structure units, a large and structured professional knowledge graph in the reactor design domain was successfully built. This graph not only contains rich entities and relationships, but also supports complex queries and reasoning, providing strong knowledge support for subsequent collaborative work of intelligent agents.

[0097] This knowledge graph not only achieves a unified representation of multi-source heterogeneous knowledge in nuclear engineering, but also serves as a foundation for intelligent analysis, supporting complex reasoning tasks in cross-disciplinary collaborative scenarios. For example, in reactor safety assessments, it can automatically trace the multi-dimensional chain effects that a failure of a certain equipment may trigger, including those related to thermal-hydraulic, neutron physics, and material properties; in design optimization, it can assist engineers in quickly identifying coupling constraints between different subsystems; and in training and emergency response, it can provide intelligent question-and-answer and decision-making suggestions based on causal chains. The entire pipeline significantly improves the automation level and interdisciplinary analysis capabilities of nuclear engineering knowledge management, laying the knowledge infrastructure for the research and development of next-generation intelligent nuclear energy systems.

[0098] 3. Knowledge Application Unit

[0099] The knowledge application unit includes: a prompt word design module, an intelligent agent construction module, a collaborative work mechanism construction module, and an intelligent question-and-answer interface construction module.

[0100] 3.1 Prompt Design Module

[0101] After the knowledge graph is constructed, prompt word engineering is performed to initialize multi-specialty agents, equipping them with domain-specific knowledge and skills. Specifically:

[0102] (1) Initialization: Based on the characteristics of each major, a special prompt word template was designed to initialize the intelligent agent.

[0103] (2) Dynamic adjustment: Based on user input and feedback, the prompts are dynamically adjusted to optimize the agent's performance. This process ensures that the agent can adapt to different application scenarios and provide more accurate and personalized services.

[0104] 3.2 Intelligent Agent Construction Module

[0105] Based on the prompt word engineering, a basic architecture for the intelligent agent was designed, including a perception module, a decision-making module, and an execution module. The specific functions of each module are as follows:

[0106] (1) Perception Module: The agent receives external input, such as user questions and task descriptions, through the perception module. The perception module is responsible for parsing the input, extracting key information, and providing a basis for subsequent decision-making and execution.

[0107] (2) Decision Module: The agent decomposes the task into multiple sub-tasks through the decision module and assigns them to different sub-agents or tools. The decision module is responsible for formulating the optimal task execution strategy to ensure the efficient completion of the task.

[0108] (3) Execution Module: The agent calls various tools and resources through the execution module to complete specific tasks. The execution module is responsible for coordinating the work of various professional agents to ensure the smooth progress of tasks.

[0109] To address the various disciplines involved in reactor design feasibility studies, a multi-disciplinary intelligent agent was established. First, a systems engineering agent was responsible for requirement identification and task breakdown, translating natural language descriptions of requirements into explicit engineering specifications such as power, volume, weight, and inlet / outlet temperature ranges. Second, a neutron physics agent was responsible for reactor physics analysis, including core power, critical dimensions, and neutron flux distribution. Third, a thermal-hydraulic agent was responsible for reactor thermal-hydraulic analysis, including inlet / outlet temperature variations, coolant selection, and flow rate settings. Fourth, a fuel and materials agent was responsible for selecting reactor fuel and materials suitable for core power and temperature, considering factors such as fuel enrichment and irradiation damage thresholds. Finally, a safety analysis agent was responsible for verifying the finalized design, analyzing whether it met the safety requirements of each discipline and the overall safety threshold requirements.

[0110] The system adopts a single-agent execution and multi-agent iterative interaction approach to achieve a closed loop of solution proposal and demonstration. The verification module significantly improves the reliability of the agent solution demonstration.

[0111] 3.3 Collaborative Working Mechanism Construction Module

[0112] To enable collaborative work among multiple specialized intelligent agents, a collaborative verification mechanism was established based on the Multi-Agent Communication Protocol (MCP). The specific steps are as follows:

[0113] (1) Message passing: Multiple agents work together through a message passing mechanism to complete complex reactor demonstration tasks. Each agent can handle specific sub-tasks based on its own expertise and skills and pass the results to other agents.

[0114] (2) Task scheduling: The system engineering professional intelligent agent is used to schedule the design tasks, allocate tasks reasonably, and improve the overall efficiency of the system. The system engineering professional intelligent agent is responsible for coordinating the work of each sub-intelligent agent to ensure the orderly progress of tasks.

[0115] (3) Collaborative optimization: Through the collaborative work of multiple agents, it is possible to efficiently process and optimize complex tasks. Each agent can leverage its own strengths to jointly complete a high-quality reactor design scheme.

[0116] 3.4 Intelligent Question-Answering Interface Construction Module

[0117] The final intelligent question-answering interface was built using interactive interface design tools (such as Gradio). Users can input questions through this interface, and the system will retrieve the top n most relevant paragraphs from the vector database, serving as the context for the larger model's answer. The intelligent question-answering interface operates as follows:

[0118] (1) User input: Users input questions through the interface, such as "Please calculate the neutron flux of component A under given conditions".

[0119] (2) Retrieval: The system retrieves the most relevant paragraphs from the vector database as context for the large model's response.

[0120] (3) Answer: The large model generates the answer based on the retrieved paragraphs and returns it to the user through the interface.

[0121] Through the above system architecture and implementation methods, this invention successfully constructs a knowledge-driven, multi-agent collaborative reactor scheme demonstration system, providing strong support for intelligent design in the field of nuclear engineering.

[0122] The above system architecture can be implemented by a computer program. The present invention further provides a storage medium storing a computer program, which is executed by a processor to implement the above knowledge-driven, multi-agent collaborative reactor scheme demonstration system.

[0123] Meanwhile, the present invention also provides a device including a storage medium and a processor, wherein the storage medium stores a computer program, and the processor executes the computer program to realize the above-mentioned knowledge-driven, multi-agent collaborative reactor scheme demonstration system.

[0124] Example

[0125] Regarding the architecture of the aforementioned knowledge-driven multi-agent collaborative reactor scheme demonstration system, the design method of the system is described below with specific embodiments.

[0126] The first step is knowledge internalization. Massive amounts of unstructured and semi-structured documents (papers, books, standards) are transformed into dynamic knowledge sources that the model can query and utilize in real time using advanced Retrieval Enhancement Generation (RAG) technology, ensuring that every key point in the model's answer is verifiable.

[0127] The knowledge internalization part mainly includes data collection, data preprocessing, and vector database construction. In this embodiment, the data collection part uses web crawling and internal knowledge channels to obtain more than 3,000 articles related to nuclear engineering and nuclear technology. It also obtains 6 nuclear engineering monographs and 22 reactor design standards or specifications through libraries and other internal resources. The final distribution of the crawled literature topics is shown in Figures 2 and 3, including engineering, physics, thermal engineering, fuel materials, simulation, etc., and keywords were extracted.

[0128] In the data preprocessing section, text cleaning is performed first, initially cleaning the collected documents and standards to remove irrelevant content such as headers, footers, and footnotes. Then, uniform format conversion and multimodal recognition are conducted. A Vision Language Model (VLM) is used to convert all documents into a lightweight markup language (Markdown) format, effectively extracting images, formulas, charts, and other content for easier subsequent reading and processing. Multimodal recognition technology ensures the integrity of text and image information, laying a solid foundation for subsequent vectorization processing and knowledge graph construction.

[0129] After acquiring the knowledge base, three large models were downloaded using the open-source cross-platform large model tool ollama: qwen3:8B, bge-m3:latest, and qwen3:32B. These were used to quickly verify the technical approach, embedding, and final answer, respectively.

[0130] In the vector database construction, the Pgvector database environment was deployed using Docker containerization technology, ensuring environment consistency and portability. Database connection and management were performed using VSCode and a MySQL plugin, ensuring data security and stability. Data splitting, vectorization, database insertion, and retrieval were performed using the Large Language Model (LLM) programming framework langchain, with the specific steps as follows:

[0131] Data splitting involves breaking long text data down into smaller paragraphs or sentences to facilitate subsequent vectorization processing. This step ensures that each paragraph or sentence is of appropriate length, making it easy for the model to process and store.

[0132] Vectorization is performed using a pre-trained large model (bge-m3) to vectorize the split text segments, generating high-dimensional vector representations. These vectors will be used for subsequent similarity searches and knowledge graph construction.

[0133] Data import involves storing vectorized data in the Pgvector database and creating indexes to support efficient retrieval operations. This step ensures that the data can be quickly queried and accessed.

[0134] The second step is knowledge structuring. RAG technology solves the problem of "knowing what," while knowledge graphs solve the problem of "knowing why." This embodiment will leverage the capabilities of large models to automatically extract core entities, relationships, and attributes from a vast knowledge source, constructing a massive professional knowledge graph for the reactor design domain, laying the foundation for deep reasoning and complex queries.

[0135] Retrieval Enhanced Generation (RAG) enhances the accuracy and traceability of generated results by coupling a parametric language model with a non-parametric external knowledge base. Its advanced paradigm, Graph Retrieval Enhanced Generation (GraphRAG), further introduces a knowledge graph as a retrieval source, utilizing graph algorithms to mine deep, multi-step semantic relationships between entities, overcoming the limitations of traditional vector retrieval in complex reasoning. This invention integrates RAG and GraphRAG to construct a two-layer retrieval-generation mechanism. The RAG component, based on vector similarity retrieval, quickly locates relevant segments from unstructured documents, handling factual and detailed queries; GraphRAG, relying on the Neo4j graph database, performs semantic retrieval, addressing system-level, multi-constraint complex problems (such as solution trade-off analysis and security impact deduction), extracting logical subgraphs through graph queries, providing context rich in relational semantics for large language models. The combination of these two significantly improves the coherence and engineering rationality of the generated content, providing reliable knowledge support for subsequent multi-agent collaborative decision-making.

[0136] This embodiment utilizes the knowledge graph-based enhanced retrieval and generation framework GraphRAG and the graph database Neo4j to automatically generate a knowledge graph and perform knowledge retrieval based on it. Considering the numerous attributes of various components, devices, and materials during the core design process, the Resource Description Framework (RDF) is ultimately chosen to construct the knowledge graph. The specific steps are as follows:

[0137] (1) Entity Recognition: Entity recognition is performed on the text data using a large model (qwen3:32B) to extract core entities, such as component names, material types, and physical parameters. In conjunction with an ontology dictionary and context-aware mechanisms in the field of nuclear engineering, fine-grained classification and disambiguation are performed on professional terms in the field of nuclear engineering, such as "pressurized water reactor," "control rods," "moderator," "neutron flux," "fuel cladding," and "LOCA accident." This process ensures the accuracy and comprehensiveness of the entities.

[0138] (2) Relation Extraction: Semantic relationships between entities are extracted from the text using natural language processing techniques, such as "Component A is made of material B" and "Component C is located at location D". The large model uses prompt engineering and few-shot learning strategies to identify relationships with engineering semantics, such as "contains", "causes", "affects", "depends on", "used for cooling", and "occurs in", and maps them to predefined relational patterns to ensure the logical rigor of the knowledge graph. This process not only extracts explicit relationships but also infers implicit relationships through context, ensuring the richness and accuracy of the relationships.

[0139] (3) Attribute extraction: Extract the attribute information of the entity, such as size, temperature, pressure, etc., extract the relationships between different attributes, and then give the output in JSON format. This process ensures a detailed description of each entity, laying the foundation for subsequent deep reasoning and complex queries.

[0140] (4) Knowledge Graph Construction: Using the Neo4j graph database, the extracted entities, relations, and attributes are organized into an RDF-formatted knowledge graph. Each entity is a node, relations are edges, and attributes are attributes of the nodes. This process ensures that the knowledge graph has a clear structure, is easy to query, and can be expanded. The system optimizes the graph structure of the initially generated triples (subject-relation-object), including entity alignment, redundancy resolution, contradiction detection, and confidence weighting. Finally, the high-quality knowledge graph is persistently stored in a graph database (such as Neo4j or Nebula Graph), supporting efficient querying and dynamic updates. The Neo4j platform used to construct the knowledge graph in this embodiment is as follows: Figure 4 As shown.

[0141] The third step is knowledge application (AI Agents). Based on RAG and knowledge graphs, a series of AI agents simulating nuclear domain experts will be built. These agents possess "independent thinking" capabilities and dedicated "toolkits," enabling them to autonomously decompose tasks and collaborate within a collaborative framework, ultimately completing complex, cross-domain analysis tasks that only nuclear engineering experts can handle.

[0142] The knowledge application section mainly includes the implementation of prompt words, intelligent agent construction, and collaborative work mechanism construction for majors such as systems engineering, physics, thermodynamics, and mechanics.

[0143] The prompt word engineering primarily initializes various specialized agents, equipping them with domain-specific knowledge and skills. For example, an agent specializing in systems engineering might need to understand project requirements and performance evaluation metrics, while an agent specializing in physics would need knowledge in areas such as neutron physics and radiation protection. Based on user input and feedback, the prompt words are dynamically adjusted to optimize the agent's performance. This process ensures that the agent can adapt to different application scenarios, providing more accurate and personalized services.

[0144] The intelligent agent construction builds upon the prompt word engineering, designing the basic architecture of the intelligent agent, including a perception module, a decision-making module, and an execution module. The intelligent agent is constructed based on the Large Language Model (LLM) programming framework, Langchain. The intelligent agent can autonomously decompose complex tasks into multiple sub-tasks and assign them to different sub-agents or tools.

[0145] Perception Module: The agent receives external input, such as user questions and task descriptions, through the perception module. The perception module is responsible for parsing the input, extracting key information, and providing a basis for subsequent decision-making and execution.

[0146] Decision Module: The agent decomposes the task into multiple sub-tasks through the decision module and assigns them to different sub-agents or tools. The decision module is responsible for formulating the optimal task execution strategy to ensure the efficient completion of the task.

[0147] Execution Module: The agent uses the execution module to invoke various tools and resources to complete specific tasks. The execution module is responsible for coordinating the work of various specialized agents to ensure the smooth progress of tasks.

[0148] This embodiment establishes a multi-disciplinary intelligent agent for the various disciplines involved in reactor scheme demonstration. First, there is a systems engineering intelligent agent, primarily responsible for requirement identification and task decomposition, translating natural language descriptions of requirements into explicit engineering specifications such as power, volume, weight, and inlet / outlet temperature ranges. Second, there is a neutron physics intelligent agent, primarily responsible for reactor physics analysis, including reactor core power, critical dimensions, and neutron flux distribution range. Third, there is a thermal-hydraulic intelligent agent, primarily responsible for reactor thermal-hydraulic analysis, including inlet / outlet temperature changes, coolant medium selection, and flow rate settings. Fourth, there is a fuel and material intelligent agent, primarily responsible for selecting reactor fuel and material materials suitable for core power and temperature, considering fuel enrichment, radiation damage thresholds, and other conditions. Finally, there is a safety analysis intelligent agent, primarily responsible for verifying the formed scheme, analyzing from a safety perspective whether the scheme meets the safety requirements of each discipline and the overall safety threshold requirements.

[0149] The system adopts a single-agent execution and multi-agent iterative interaction approach to achieve a closed loop of solution proposal and demonstration. The verification module significantly improves the reliability of the agent solution demonstration.

[0150] The collaborative working mechanism is built upon the MCP model context protocol to establish a multi-agent collaborative demonstration mechanism. Multiple agents can collaborate through message passing to jointly complete complex reactor demonstration tasks. Design tasks are scheduled and allocated rationally by systems engineering agents, improving the overall efficiency of the system. Through the collaborative work of multiple agents, complex tasks can be efficiently processed and optimized. Each agent can leverage its strengths to collaboratively produce a high-quality reactor design scheme.

[0151] This system constructs an agent service layer based on the Model Context Protocol (MCP), integrating multi-scale language models and hybrid retrieval enhancement mechanisms to support multidisciplinary collaborative design and analysis in nuclear engineering. The MCP server, as the core hub for agents to access external tools and knowledge resources, coordinates the collaborative work of three main types of models: the embedded model bge-m3 is responsible for text vectorization and semantic matching, providing the foundation for retrieval tasks; Qwen3-8B, as a lightweight model, is used for rapid verification of inter-agent communication and process scheduling; and Qwen3-32B, as a heavyweight generative model, undertakes the final solution synthesis and comprehensive reasoning tasks, ensuring the professionalism and accuracy of the output.

[0152] In the collaborative workflow, the system employs multi-role intelligent agents to mimic a nuclear engineering expert team, achieving professional division of labor. The system engineer agent first uses the MBSE methodology and prompt word engineering techniques to decompose user requirements layer by layer, constructing a system functional and logical model, and then distributing tasks to a cluster of specialized analytical agents. Physics, thermal, and mechanics expert agents are responsible for the analysis and feasibility assessment of schemes related to neutron physics characteristics, thermal-hydraulic performance, and structural mechanics, respectively. Each specialized agent can access two types of knowledge services via MCP: RAG multi-way recall retrieves the most relevant Top-K text fragments from a vector library, providing direct knowledge support; GraphRAG extracts multi-hop relationship subgraphs between entities from the Neo4j knowledge graph, revealing deep coupling relationships and constraints between parameters. Numerical simulation tools can also be integrated when necessary to achieve data-driven, precise analysis.

[0153] All professional analysis results are ultimately submitted to a security analysis expert agent for system-level security and compliance verification. This agent verifies the integrated solution based on specifications and guidelines in the knowledge base. If the verification fails, feedback is generated and the system is reverted to the corresponding professional agent for re-analysis, forming a closed-loop iterative optimization mechanism until all design requirements are met. Finally, the system automatically outputs a complete design solution report (Word format). This process, through the deep integration of technologies such as large language models, hybrid retrieval enhancement, and graph reasoning, achieves knowledge integration, coupled reasoning, and iterative optimization in the multidisciplinary collaborative design process, significantly improving the automation and reliability of complex nuclear engineering system design. The specific process is as follows: Figure 5 As shown.

[0154] Finally, the intelligent question-answering interface was built using the Python library (gradio) for building interactive interfaces, as shown in Figure 6.

[0155] Users can input questions through this interface. When a question is asked, the system retrieves the top n most relevant paragraphs from the vector database, which serve as the context for the large model's answer. The final question-and-answer result is as follows: Figure 7 As shown.

[0156] For example, if a user's question is how to build a reactor, the intelligent solution demonstration system searches the vector database. The most relevant paragraphs retrieved include the composition and structure of a pressurized water reactor, the structure of the pressure vessel and core, the function and structure of control rods, and the structure of the Hualong One reactor. The intelligent question-and-answer interface then presents the retrieved contextual fragments as the content of the large model's answer to the user.

[0157] To objectively evaluate the system's performance, a specialized test suite (NuclEval-Bench) covering five core disciplines—physics, thermal engineering, fuel, safety, and systems—was constructed, comprising 20 questions. These questions are derived from exercises in classic nuclear engineering textbooks and real-world engineering scenarios, and are categorized into three difficulty levels: L1 (basic factual), L2 (professional analysis and derivation), and L3 (multidisciplinary collaborative design). A general large model (DeepSeek) and a networked search combining the general large model were selected as the baseline for comparison. Quantitative evaluation was conducted across three dimensions: accuracy, illusion rate, and response time. The evaluation results are as follows: Figure 8 As shown, this invention significantly outperforms the capabilities of deep thinking and networked search in solving complex, multi-disciplinary coupled reactor problems on a test set.

[0158] Those skilled in the art will understand that the specific order of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order of steps in the process can be rearranged without departing from the scope of the invention. The appended methods provide elements of various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0159] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. Thus, the invention also intends to include such variations and adaptations if they fall within the scope of the claims and their equivalents.

[0160] The above embodiments are merely illustrative examples of the present invention. The present invention may also be implemented in other specific ways or forms without departing from its spirit or essential characteristics. Therefore, the described embodiments should be considered illustrative rather than limiting in any respect. The scope of protection of the present invention should be defined by the claims, and any variations equivalent to the intent and scope of the claims should also be included within the scope of the present invention.

Claims

1. A knowledge-driven, multi-agent collaborative reactor scheme demonstration system, characterized in that, include: Knowledge internalization units are used to transform unstructured and semi-structured documents into dynamic knowledge sources that can be queried and utilized by models in real time. Knowledge structuring units are used to extract core entities, relationships, and attributes from the knowledge source to construct a professional knowledge graph in the reactor design domain. The knowledge application unit, based on the knowledge source and knowledge graph, constructs intelligent agents and collaborative working mechanisms for multiple professional intelligent agents for different professions, and builds an intelligent question-and-answer interface.

2. The knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 1, characterized in that, The knowledge internalization unit includes: The data collection module is used to acquire professional literature, design specifications, and calculation models in the process of reactor design, analysis, and safety assessment. The data preprocessing module is used to clean the collected data, remove irrelevant content, and convert all documents into a lightweight markup language format using a large visual language model. The vector database building module is used to split and vectorize data, store vectorized data in a database environment, and build indexes to support efficient retrieval operations.

3. The knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 1, characterized in that, The knowledge structuring unit includes: The entity recognition module is used to perform entity recognition on text data by leveraging the semantic understanding and logical reasoning capabilities of large models, and extract core entities as nodes of the knowledge graph; and combined with the ontology dictionary and context-aware mechanism in the field of nuclear engineering, it performs fine-grained classification and disambiguation of professional terms in the field of nuclear engineering to ensure the consistency and accuracy of node types. The relation extraction module is used to extract semantic relationships between entities from text data using natural language processing technology as edges of the knowledge graph. The large model identifies relationships with engineering semantics through prompting engineering and few-shot learning strategies, and maps them to predefined relation patterns to ensure the logical rigor of the knowledge graph. The attribute extraction module is used to extract the attribute information of the entity and the relationships between different attributes; The graph construction module is used to organize the extracted entities, relations and attributes into a knowledge graph with a specific structure. The system optimizes the graph structure of the initially generated triples, including entity alignment, redundancy resolution, contradiction detection and confidence weighting, and finally persists the high-quality knowledge graph in the graph database.

4. The knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 1, characterized in that, The knowledge application unit includes: The prompt word design module is used to initialize multi-disciplinary intelligent agents, enabling them to possess knowledge and skills in specific fields; the multi-disciplinary intelligent agents include systems engineering intelligent agents, neutron physics intelligent agents, thermal hydraulics intelligent agents, fuel materials intelligent agents, and safety analysis intelligent agents; The agent building module is used to design the basic architecture of the agent, enabling the agent to extract key information, decompose tasks, and call various tools and resources to complete specific tasks. The collaborative work mechanism construction module is used to establish a collaborative work mechanism among multiple professional intelligent agents through a multi-agent communication protocol. The intelligent question-and-answer interface building module is used to build the final intelligent question-and-answer interface, allowing users to input questions through the interface, and the big model to generate answers based on the retrieved content and return them to the user through the interface.

5. The knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 4, characterized in that, The basic architecture of the intelligent agent includes: The perception module is used to analyze external input and extract key information; The decision-making module is used to formulate task execution strategies, decompose tasks, and assign the decomposed subtasks to intelligent agents or tools of different specialties. The execution module is used to coordinate the work of various professional intelligent agents and call on various tools and resources to complete specific tasks.

6. The knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 4, characterized in that, Among the multi-disciplinary intelligent agents, the systems engineering agent is responsible for requirement identification and task decomposition, transforming the requirements described in natural language into explicit engineering requirements; the neutron physics agent is responsible for reactor physics analysis, including reactor core power, critical dimensions, and neutron flux distribution range; the thermal-hydraulic agent is responsible for reactor thermal-hydraulic analysis, including reactor inlet and outlet temperature changes, coolant medium selection, and flow rate settings; the fuel and materials agent is responsible for selecting reactor fuel and materials that are compatible with core power and temperature, fuel enrichment, and irradiation damage thresholds; and the safety analysis agent is responsible for verifying the proposed solutions, analyzing from a safety perspective whether the solutions can meet the safety requirements of each specialty and the overall safety threshold requirements. The system adopts a single-agent execution and multi-agent iterative interaction approach to achieve a closed loop of solution proposal and verification.

7. A design method for a knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in any one of claims 1-6, characterized in that, Includes the following steps: S1) Transform unstructured and semi-structured documents into dynamic knowledge sources that can be queried and utilized by the model in real time; including data collection, data preprocessing, and vector database construction; S2) Extract core entities, relationships, and attributes from the knowledge source to construct a professional knowledge graph for reactor design. S3) Based on the knowledge source and knowledge graph, construct intelligent agents and collaborative working mechanisms of multi-professional intelligent agents for different professions, and construct an intelligent question-and-answer interface.

8. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 7, characterized in that, The data collection in step S1) includes acquiring professional literature, design specifications and calculation models from multiple fields such as neutron physics, thermal hydraulics, fuel materials and safety analysis; Data preprocessing includes cleaning the collected data, removing irrelevant content, and converting all documents into a lightweight markup language format using a large visual language model. The construction of a vector database includes splitting and vectorizing the data, storing the vectorized data in a database environment, and building indexes to support efficient retrieval operations.

9. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 8, characterized in that, The vector database construction described in step S1) includes the following steps: S11) Data splitting: breaking long text data into smaller paragraphs or sentences; S12) Vectorization processing: The pre-trained large model is used to vectorize the split text segments to generate high-dimensional vector representations. S13) Data entry: The vectorized data is stored in the database, and an index is created to support efficient retrieval operations.

10. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 7, characterized in that, Step S2) specifically includes: S21) Utilize the semantic understanding and logical reasoning capabilities of large models to perform entity recognition on text data, extract core entities as nodes of the knowledge graph; and combine the ontology dictionary and context-aware mechanism in the field of nuclear engineering to perform fine-grained classification and disambiguation of professional terms in the field of nuclear engineering, ensuring the consistency and accuracy of node types. S22) Using natural language processing technology, semantic relationships between entities are extracted from text as edges of the knowledge graph; the large model identifies relationships with engineering semantics through cue engineering and few-shot learning strategies, and maps them to predefined relationship patterns to ensure the logical rigor of the knowledge graph. S23) Extract the attribute information of the entity and extract the relationships between different attributes; S24) The extracted entities, relations and attributes are organized into a knowledge graph in the form of a resource description framework, with each entity as a node, relations as edges and attributes as attributes of the nodes; the system optimizes the graph structure of the initially generated triples, including entity alignment, redundancy resolution, contradiction detection and confidence weighting, and finally persists the high-quality knowledge graph in the graph database.

11. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 10, characterized in that, The core entities mentioned in step S21) include: component name, material type, and physical parameters.

12. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 7, characterized in that, Step S3) specifically includes: S31) Based on the characteristics of each major, a special prompt word template was designed to initialize the multi-major intelligent agent and dynamically adjust the prompt words according to the user's input and feedback; S32) Design the basic architecture of the intelligent agent so that the intelligent agent can extract key information, decompose tasks, and call various tools and resources to complete specific tasks; S33) Establish a collaborative working mechanism among multiple professional intelligent agents through a multi-agent communication protocol; S34) Construct the final intelligent question-answering interface, allowing users to input questions through the interface, and the large model to generate answers based on the retrieved content, which are then returned to the user through the interface.

13. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 12, characterized in that, The multi-disciplinary intelligent agents mentioned in step S31) include a systems engineering intelligent agent, a neutron physics intelligent agent, a thermal-hydraulic intelligent agent, a fuel and materials intelligent agent, and a safety analysis intelligent agent. The systems engineering intelligent agent is responsible for requirement identification and task decomposition, transforming natural language descriptions of requirements into explicit engineering requirements. The neutron physics intelligent agent is responsible for reactor physics analysis, including reactor core power, critical dimensions, and neutron flux distribution range. The thermal-hydraulic intelligent agent is responsible for reactor thermal-hydraulic analysis, including reactor inlet and outlet temperature changes, coolant medium selection, and flow rate settings. The fuel and materials intelligent agent is responsible for selecting reactor fuel and materials suitable for core power and temperature, fuel enrichment, and irradiation damage threshold conditions. The safety analysis intelligent agent is responsible for verifying the proposed solutions, analyzing from a safety perspective whether the solutions meet the safety requirements of each specialty and the overall safety threshold requirements. The system adopts a single-agent execution and multi-agent iterative interaction approach to achieve a closed loop of solution proposal and verification.

14. The design method of the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in claim 12, characterized in that, Step S33) establishes a collaborative working mechanism among multiple professional intelligent agents, which specifically includes: Message passing: Each agent processes specific sub-tasks based on its own expertise and skills, and passes the results to other agents through the message passing mechanism. Multiple agents work together to complete the complex reactor demonstration task. Task scheduling is carried out by a systems engineering professional intelligent agent to design task scheduling, rationally allocate tasks, and coordinate the work of various professional intelligent agents. Collaborative optimization enables efficient processing and optimization of complex tasks through the collaborative work of multiple specialized intelligent agents.

15. A storage medium storing a computer program, characterized in that, The computer program is executed by a processor to implement the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in any one of claims 1-6.

16. An apparatus comprising a storage medium and a processor, the storage medium storing a computer program, characterized in that, When the processor executes the computer program, it implements the knowledge-driven multi-agent collaborative reactor scheme demonstration system as described in any one of claims 1-6.

Images