Multi-intelligent-agent physical examination enhancement generation system and question answering method, equipment and product thereof

Abstract

The invention is suitable for the technical field of search enhancement generation, and provides a multi-intelligent-agent search enhancement generation system and a question answering method, equipment and product thereof. According to the question and answer method of the multi-agent search enhancement generation system provided by the embodiment of the invention, multiple agents such as the routing decision agent and the tool use agent are introduced, so that information can be obtained from an external database, a reasoning process can be autonomously planned based on domain knowledge, a search strategy is dynamically optimized, and the search efficiency is improved. And a specific analysis tool of a specific industry is flexibly called. Different from a static'retrieval-reading-generation 'mode of a retrieval enhancement generation system in the prior art, the multi-intelligent-body retrieval enhancement generation system adopts a multi-round interaction and intelligent optimization mechanism, and dynamically evaluates a retrieval result, adjusts a query mode and calls a professional tool in an answer generation process so as to ensure high-quality intelligent question and answer service.

Description

Technical Field

[0001] The present application belongs to the field of search enhancement generation technology, and in particular relates to a multi-agent search enhancement generation system and its question-answering method, equipment and product. Background Art

[0002] As the power industry becomes increasingly intelligent, traditional information retrieval and expert systems are no longer able to meet the growing demand for intelligent question-answering. For example, tasks such as power equipment operation and maintenance, fault diagnosis, and policy and regulatory inquiries often involve extensive expertise and require comprehensive analysis from multiple data sources. Consequently, the Retrieval-Augmented Generation (RAG) method has garnered widespread attention in the power industry.

[0003] The RAG approach combines information retrieval with a large language model (LLM). It can dynamically access information from external knowledge bases without retraining the model, thereby improving the accuracy and interpretability of question-answering systems. In the power industry, RAG can be used in scenarios such as grid operation knowledge query, power equipment document parsing, and accident case analysis, helping to improve operation and maintenance efficiency and decision-making quality. However, despite the excellent performance of traditional RAG approaches in knowledge retrieval and generation, they still have many limitations in practical applications, making them difficult to meet the needs of complex business scenarios in the power industry. These limitations include the following four points: 1. Limited information integration capabilities. When processing multi-source information, traditional RAG struggles to effectively integrate shared attributes across multiple data fragments, thus impacting the quality of comprehensive answers. For example, certain fault analyses require reasoning based on information from multiple sources, and traditional RAG has significant shortcomings in connecting key points and forming logical chains. 2. Weak understanding of large-scale documents. When processing massive amounts of power equipment manuals, industry standard documents, or operation reports, traditional RAG may be unable to accurately extract key semantic concepts, impacting overall understanding quality. 3. Lack of dynamic adaptability. Traditional RAGs use a fixed "retrieval-reading-generation" process that cannot dynamically adjust to task requirements, resulting in poor performance when handling complex fault diagnosis or multi-step reasoning tasks. 4. Lack of autonomous optimization capabilities. Traditional RAGs rely on static query patterns and lack the ability to self-correct failed queries or inaccurate answers, making it difficult to optimize output results during the interactive process. Summary of the Invention

[0004] The embodiments of the present application provide a multi-agent retrieval enhancement generation system and its question-answering method, equipment and product, which can solve problems such as limited information integration capabilities of the retrieval enhancement generation method in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a question-answering method for a multi-agent search and enhancement generation system, wherein the multi-agent search and enhancement generation system includes a routing decision agent and a tool use agent, and the method includes:

[0006] When receiving query information submitted by a user, determining one or more query tasks corresponding to the query information based on whether the query information has in-depth research requirements;

[0007] The routing decision agent inputs the one or more query tasks into the tool use agent according to the query requirements respectively corresponding to the one or more query tasks, and obtains one or more search results output by the tool use agent;

[0008] Performing quality assessment and system optimization based on the one or more search results to obtain high-quality processing results;

[0009] Determining whether the processing result satisfies the query information submitted by the user;

[0010] If the processing results satisfy the query information submitted by the user, summarizing the processing results to generate a comprehensive answer;

[0011] The comprehensive answer is provided to the user.

[0012] In a possible implementation of the first aspect, the multi-agent retrieval enhancement generation system further includes a query planning agent, and determining one or more query tasks corresponding to the query information based on whether the query information has an in-depth research requirement includes:

[0013] If the query information has an in-depth research requirement, the query planning agent determines one or more query tasks corresponding to the query information;

[0014] If the query information does not have an in-depth research requirement, the query information is determined as a query task.

[0015] In a possible implementation of the first aspect, after determining whether the processing result satisfies the query information submitted by the user, the method further includes:

[0016] If the processing result does not satisfy the query information submitted by the user, the process returns to the step of the query planning agent determining one or more query tasks corresponding to the query information.

[0017] In a possible implementation of the first aspect, inputting the one or more query tasks into a tool-using agent and obtaining one or more search results output by the tool-using agent includes:

[0018] Inputting the one or more query tasks into a tool-using agent, wherein the tool-using agent includes one or more retrieval agents and their corresponding retrieval tools;

[0019] The one or more retrieval agents respectively use corresponding retrieval tools to determine one or more retrieval results corresponding to the one or more query tasks.

[0020] In a possible implementation of the first aspect, the method further includes at least any one of the following:

[0021] When erroneous information is detected, the erroneous information is corrected to obtain correct information;

[0022] Adaptive optimization is performed based on the category of failure mode.

[0023] In a possible implementation of the first aspect, when erroneous information is detected, correcting the erroneous information to obtain correct information includes at least any one of the following:

[0024] When format error information is detected, correcting the format error information to obtain correct format information;

[0025] When it is detected that the retrieval results do not match the user's query intention, the retrieval process is updated to obtain updated retrieval results;

[0026] When it is detected that the data source is insufficient and the comprehensive answer is incomplete, the data source is updated to obtain an updated comprehensive answer;

[0027] When a task failure is detected, the abnormal task is recorded and manual review is requested.

[0028] In a possible implementation of the first aspect, the performing of adaptive optimization according to the category of the failure mode includes at least any one of the following:

[0029] If the failure mode is not obtaining high-quality results, adjust the search parameters and optimize the keywords;

[0030] If the failure mode is high-frequency failure, store the adjustment record and apply the verified optimization strategy when receiving similar query information;

[0031] If the failure mode is reasoning failure, use error analysis tools to track the question-answering execution path, identify the cause of the failure, and optimize the reasoning logic.

[0032] In a second aspect, an embodiment of the present application provides a multi-agent retrieval enhancement generation system, the multi-agent retrieval enhancement generation system comprising:

[0033] A first determination module is configured to, upon receiving query information submitted by a user, determine one or more query tasks corresponding to the query information based on whether the query information has an in-depth research requirement;

[0034] a routing decision agent, configured to input the one or more query tasks into a tool usage agent according to query requirements respectively corresponding to the one or more query tasks, and obtain one or more search results output by the tool usage agent;

[0035] A first evaluation module is used to perform quality evaluation and system optimization based on the one or more search results to obtain a processing result of good quality;

[0036] A second evaluation module is used to determine whether the processing result satisfies the query information submitted by the user;

[0037] A first summarizing module is configured to summarize the processing results and generate a comprehensive answer if the processing results meet the query information submitted by the user;

[0038] The first providing module is used to provide the comprehensive answer to the user.

[0039] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the electronic device implements the method described in the first aspect above.

[0040] In a fourth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed, enables the method described in the first aspect above to be executed.

[0041] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method of the first aspect described above is implemented.

[0042] The first aspect of the embodiment of the present application has the following beneficial effects compared with the prior art:

[0043] The question-answering method of the multi-agent retrieval enhancement generation system provided in the embodiment of the present application includes a routing decision agent and a tool use agent. The method includes: when receiving query information submitted by a user, determining one or more query tasks corresponding to the query information based on whether the query information has an in-depth research requirement; the routing decision agent inputs the one or more query tasks into the tool use agent based on the query requirements corresponding to the one or more query tasks, and obtains one or more retrieval results output by the tool use agent; performing quality assessment and system optimization based on the one or more retrieval results to obtain a high-quality processing result; determining whether the processing result meets the query information submitted by the user; if the processing result meets the query information submitted by the user, comprehensively summarizing the processing result to generate a comprehensive answer; and providing the comprehensive answer to the user. The question-answering method of the multi-agent retrieval enhancement generation system provided in the embodiment of the present application, by introducing multiple agents such as the routing decision agent and the tool use agent, can not only obtain information from an external database, but also autonomously plan the reasoning process based on domain knowledge, dynamically optimize the query strategy, and flexibly call specific analysis tools for specific industries. Unlike the static "retrieval-reading-generation" model of the retrieval enhancement generation system in the existing technology, the multi-agent retrieval enhancement generation system adopts multi-round interaction and intelligent optimization mechanism to dynamically evaluate the retrieval results, adjust the query method, and call professional tools during the answer generation process to ensure high-quality intelligent question and answer services.

[0044] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0046] Figure 1 This is a schematic diagram of the question-answering process based on the retrieval enhancement generation system in the prior art;

[0047] Figure 2 This is a flowchart of a question-answering method of a multi-agent retrieval enhancement generation system according to an embodiment of the present application;

[0048] Figure 3 This is a flowchart of a question-answering method of a multi-agent retrieval enhancement generation system according to another embodiment of the present application;

[0049] Figure 4 Schematic diagram of the structure of the multi-agent retrieval enhancement generation system provided in an embodiment of the present application;

[0050] Figure 5 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0052] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0053] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0054] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0055] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0056] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0057] Figure 1 It is a schematic diagram of the question-answering process based on the retrieval enhancement generation system in the prior art.

[0058] like Figure 1 As shown in Figure 2, the existing retrieval enhancement generation system mainly includes the following three core steps.

[0059] S11: The private knowledge base is fed into the preprocessing pipeline, and the knowledge base is constructed based on the processed objects output by the preprocessing pipeline. The processed objects are then sliced ​​and the documents are vectorized using an embedding model and stored in a vector database for subsequent retrieval.

[0060] S12, uses the embedding model to vectorize the user query, searches the vector database by searching the retrieval vector, uses the cosine similarity method to retrieve the top K most relevant (Top-K) results, and re-ranks them.

[0061] S13: The first K documents retrieved are used as context and input into the large language model together with the user query and prompt template for reasoning, and finally the final response (answer) is generated.

[0062] In actual applications, the above existing retrieval enhancement generation systems, although they combine large language models and information retrieval mechanisms and perform well in knowledge acquisition and generation, still have certain limitations. In particular, they show many deficiencies in handling complex tasks, dynamic reasoning and multi-step decision-making. For example, the retrieval enhancement generation system relies on a static retrieval mechanism, which makes it difficult to adjust reasoning according to task requirements. At the same time, there are challenges in cross-document reasoning and complex information integration. In comparison, the technical solution of the multi-agent based retrieval enhancement generation system proposed in this application introduces multiple agents, enabling the system to dynamically adjust the reasoning path and achieve stronger adaptability and intelligent reasoning capabilities. Compared with the technical solution of the retrieval enhancement generation system in the prior art, the multi-agent retrieval enhancement generation system proposed in this application optimizes the retrieval and reasoning process through the collaboration of multiple agents, can make dynamic decisions based on task requirements, and provide more accurate and logical answers in complex scenarios.

[0063] Figure 2 This is a flowchart of the question-answering method of the multi-agent retrieval enhancement generation system in one embodiment of the present application.

[0064] The method of the embodiment of the present application is applied to a multi-agent search enhancement generation system. The multi-agent search enhancement generation system includes a routing decision agent and a tool use agent.

[0065] S21 , when query information submitted by a user is received, one or more query tasks corresponding to the query information are determined based on whether the query information has an in-depth research requirement.

[0066] User-submitted query information includes, but is not limited to, query statements and query terms. Simple queries don't require in-depth research. Complex queries do require in-depth research. For example, if the query is "development trends in the new energy industry," it requires in-depth research and can be broken down into multiple query tasks, such as "policy impact," "market demand," and "technological breakthroughs."

[0067] S22, the routing decision agent inputs the one or more query tasks into the tool usage agent according to the query requirements respectively corresponding to the one or more query tasks, and obtains one or more retrieval results output by the tool usage agent.

[0068] The routing decision agent uses a large language model inference mechanism to evaluate query intent, information requirement type, data availability, and other query requirements. It then notifies the tool agent of the query task and its requirements, enabling the tool agent to retrieve the query task using appropriate retrieval agents and tools, obtaining results that meet the query requirements. The routing decision agent optimizes its decision-making process based on historical interaction data and model feedback, ensuring that query tasks are assigned to the most appropriate data source.

[0069] S23, performing quality assessment and system optimization based on the one or more search results to obtain a processing result of good quality.

[0070] After obtaining one or more search results, the multi-agent search enhancement generation system enters the steps of quality assessment, quality verification and system optimization. The multi-agent search enhancement generation system can filter and optimize data through the quality assessment module. The multi-agent search enhancement generation system uses natural language processing and machine learning algorithms to sort and filter one or more search results, and combines the reasoning ability of large language models to evaluate the completeness, relevance and reliability of the information. In addition, the multi-agent search enhancement generation system structures the processing results and updates the internal memory to achieve system optimization and enhance the accuracy and consistency of subsequent queries. For example, when dealing with power equipment fault diagnosis problems, the system will give priority to integrating historical fault cases and combine them with real-time monitoring data to ensure the accuracy of the answers.

[0071] S24: Determine whether the processing result satisfies the query information submitted by the user.

[0072] After quality assessment and system optimization, the multi-agent retrieval enhancement generation system enters the reflection stage to determine whether the data of the current processing results is sufficient to serve as an answer to support the query information submitted by the user.

[0073] S25: If the processing result satisfies the query information submitted by the user, the processing result is comprehensively summarized to generate a comprehensive answer.

[0074] If the processed data meets the requirements for answering the user's query, the multi-agent search and generation system proceeds to the comprehensive summary phase to generate a complete, integrated answer. The multi-agent search and generation system integrates all high-quality data from the processed results into a large language model. The large language model integrates contextual information to generate coherent, accurate, and readable answers. The comprehensive answer is then generated in a structured format, ensuring completeness and actionability.

[0075] S26, providing the comprehensive answer to the user.

[0076] The multi-agent retrieval enhancement generation system can provide comprehensive answers to users through various communication methods such as wired or wireless.

[0077] The question-answering method for the multi-agent retrieval enhancement generation system provided in the embodiments of this application, by introducing multiple agents such as routing decision agents and tool usage agents, can not only obtain information from external databases, but also autonomously plan reasoning processes based on domain knowledge, dynamically optimize query strategies, and flexibly call specific analysis tools for specific industries. Unlike the static "retrieval-reading-generation" model of retrieval enhancement generation systems in the prior art, the multi-agent retrieval enhancement generation system adopts multiple rounds of interaction and intelligent optimization mechanisms to dynamically evaluate retrieval results, adjust query methods, and call professional tools during the answer generation process to ensure high-quality intelligent question-answering services.

[0078] In one embodiment, the multi-agent retrieval enhancement generation system also includes a query planning agent. The above step S21 determines one or more query tasks corresponding to the query information based on whether the query information has in-depth research requirements, including the following steps S211 or S212.

[0079] S211: If the query information has an in-depth research requirement, the query planning agent determines one or more query tasks corresponding to the query information.

[0080] S212: If the query information does not have an in-depth research requirement, determine the query information as a query task.

[0081] The multi-agent enhanced search and generation system processes user-submitted queries in different steps depending on the query requirements. If the query doesn't require in-depth research, the multi-agent enhanced search and generation system is directly invoked for a quick answer, effectively defining the query as a task. If the query requires in-depth research, the query planning agent is used to decompose and plan the query.

[0082] The query planning agent can parse the query information submitted by the user and break it down into multiple query tasks that can be executed in parallel. Based on the complexity of the query content, the availability of data sources, and the importance of the query target, the multi-agent retrieval enhancement generation system can dynamically adjust the number and allocation strategy of query tasks. For example, when processing the "development trend of the new energy industry", the multi-agent retrieval enhancement generation system can split the query information into multiple query tasks such as "policy impact", "market demand", and "technological breakthroughs", and assign them to different retrieval pipelines for retrieval.

[0083] In one embodiment, after determining whether the processing result satisfies the query information submitted by the user in step S24, the question-answering method of the multi-agent retrieval enhancement generation system further includes step S251.

[0084] S251, if the processing result does not satisfy the query information submitted by the user, return to the step of the query planning agent determining one or more query tasks corresponding to the query information, that is, step S211.

[0085] If the multi-agent retrieval enhancement system determines that the processed information does not meet the response requirements of the user's query information, it returns to the step where the query planning agent determines one or more query tasks corresponding to the query information, rewrites the query tasks, adjusts the retrieval strategy, and re-enters the retrieval process to obtain more comprehensive data support. For example, when solving a smart grid dispatch optimization problem, if the multi-agent retrieval enhancement system finds a lack of real-time load data, it will prioritize supplementing the latest dispatch information to improve the reliability of the answer.

[0086] In one embodiment, the step S22 of the above embodiment inputs the one or more query tasks into the tool-using agent to obtain one or more retrieval results output by the tool-using agent, including the following steps S221 and S222.

[0087] S221: Input the one or more query tasks into a tool-using agent, where the tool-using agent includes one or more retrieval agents and their corresponding retrieval tools.

[0088] S222, the one or more retrieval agents respectively use corresponding retrieval tools to determine one or more retrieval results corresponding to the one or more query tasks.

[0089] The tool's usage agents include multiple types of retrieval agents, each optimized for different data sources. For example, retrieval agent A focuses on semantic vector searches and is suitable for extracting unstructured data from PDFs, books, and internal documents. Retrieval agent B handles structured queries and can interact with SQL databases to retrieve structured data. Retrieval agent C supports real-time web searches and is suitable for obtaining the latest public information. Retrieval agent D is used to access proprietary systems or external services to meet specific industry needs.

[0090] After determining the query requirements corresponding to one or more query tasks, the routing agent selects a retrieval agent that matches the query task as a retrieval pipeline. For example, for queries involving real-time information, the routing agent prioritizes the network retrieval agent. Alternatively, for queries related to technical documents, the routing agent selects the internal knowledge base retrieval agent. Each retrieval agent can execute retrieval tasks in parallel and select the most relevant data for subsequent processing based on the query intent.

[0091] The retrieval agent in the tool-using agent has the ability to interact with external tools. During the retrieval process, the retrieval agent can call retrieval tools such as API, database query or perform computing tasks to improve data acquisition capabilities.

[0092] Figure 3 This is a flowchart of the question-answering method of the multi-agent retrieval enhancement generation system in another embodiment of the present application.

[0093] S31: Upon receiving a query submitted by a user, if the query has an in-depth research requirement, the query planning agent determines one or more query tasks corresponding to the query. If the query does not have an in-depth research requirement, the query is determined as one query task.

[0094] The user submits query information, and the multi-agent retrieval enhancement generation system enters different processing flows based on the query requirements. When the query information does not require in-depth research, a quick answer is given directly, and the query information is determined as a query task. If the query requires in-depth research, the query planning agent is entered to decompose and plan the task. The query planning agent parses the user's query information and breaks it down into multiple query tasks that can be executed in parallel. Based on the complexity of the query content, the availability of the data source, and the importance of the query target, the multi-agent retrieval enhancement generation system dynamically adjusts the number of sub-queries and the allocation strategy. For example, when processing the "development trend of the new energy industry", the system can be split into sub-queries such as "policy impact", "market demand", and "technological breakthroughs", and assigned to different retrieval pipelines for execution. After the task decomposition is completed, the query routing decision stage begins.

[0095] S32: The routing decision agent inputs the one or more query tasks into a tool-using agent based on the query requirements corresponding to the one or more query tasks. The tool-using agent includes one or more retrieval agents and their corresponding retrieval tools. The one or more retrieval agents use the corresponding retrieval tools to determine one or more retrieval results corresponding to the one or more query tasks.

[0096] The routing agent uses a large language model inference mechanism to evaluate the query intent, information requirement type, and data availability, and selects the most appropriate retrieval agent pipeline. The routing agent optimizes its decision-making process based on historical interaction data and model feedback, ensuring that queries are assigned to the most appropriate data sources. For example, for queries involving real-time information, the web retrieval agent is prioritized; for queries related to technical documents, the internal knowledge base agent is selected.

[0097] Tool usage agents have the ability to interact with external tools. During the retrieval process, they can call APIs, query databases, or perform computing tasks to improve data acquisition capabilities. Specialized retrieval agents are optimized for different data sources, including: Retrieval agent A, which focuses on semantic vector search and is suitable for extracting unstructured data from PDFs, books, and internal documents. Retrieval agent B is responsible for structured queries and can interact with SQL databases to obtain structured data. Retrieval agent C is used to support real-time web searches and is suitable for obtaining the latest public information. Retrieval agent D is used to access proprietary systems or external services to meet specific industry needs. Each agent executes retrieval tasks in parallel and filters the most relevant data for subsequent processing based on query intent.

[0098] S33, performing quality assessment and system optimization based on the one or more search results to obtain a processing result of good quality.

[0099] After the search results are returned, the multi-agent search enhancement generation system first enters the quality verification phase, filtering and optimizing the data through the quality assessment module. The multi-agent search enhancement generation system uses natural language processing and machine learning algorithms to sort and filter the search results, and combines large-scale model reasoning to assess the completeness, relevance, and reliability of the information. The multi-agent search enhancement generation system then structures the processed results and updates its internal memory to enhance the accuracy and consistency of subsequent queries. For example, when diagnosing power equipment faults, the system prioritizes integrating historical failure cases and combining them with real-time monitoring data to ensure the accuracy of the answers.

[0100] S34: Determine whether the processing result satisfies the query information submitted by the user.

[0101] S351: If the processing result satisfies the query information submitted by the user, or if the maximum number of feedback iterations has been reached, the processing result is comprehensively summarized to generate a comprehensive answer.

[0102] S352: If the processing result does not satisfy the query information submitted by the user, return to the step of the query planning agent determining one or more query tasks corresponding to the query information.

[0103] After quality verification and memory updating, the multi-agent retrieval enhancement generation system enters the reflection phase, determining whether the data from the current processing results is sufficient to support the answer. If it determines that the information from the processing results is still insufficient, the query is automatically rewritten, the retrieval strategy is adjusted, and the retrieval process is re-entered to obtain more comprehensive data support. If the data from the processing results meets the answer requirements, the system proceeds to the final comprehensive summary phase to generate a complete answer. For example, when solving a smart grid dispatch optimization problem, if the system finds that real-time load data is lacking, it will prioritize supplementing the latest dispatch information to improve the reliability of the answer.

[0104] S36, providing the comprehensive answer to the user.

[0105] After completing quality verification and reflective optimization, the multi-agent retrieval enhancement generation system feeds all high-quality data into a large model for final information integration. The large model combines contextual information to generate a coherent, accurate, and readable comprehensive answer, which is returned to the user in a structured format to ensure completeness and actionability.

[0106] In the above embodiments, the query planning agent in S31 and the comprehensive summary module in S35 can be implemented using the same type of query-type large language model. More specifically, the query planning agent in S31 and the comprehensive summary module in S35 can be implemented using the same first large language model to reduce the number of large language models and improve efficiency. The routing decision agent in S32, the verification quality module in S33, and the reflection module in S34 that determines whether the processing result meets the query information submitted by the user can be implemented using the same type of decision-type large language model. More specifically, the reflection module in S34 that determines whether the processing result meets the query information submitted by the user can be implemented using the same second large language model to reduce the number of large language models and improve efficiency. The tool usage agent in S32 can be implemented by the third large language model.

[0107] In one embodiment, the question-answering method based on the multi-agent retrieval enhancement generation system also includes steps S31 and / or S32.

[0108] S31, when error information is detected, correct the error information to obtain correct information.

[0109] S32, performs adaptive optimization according to the category of failure mode.

[0110] The multi-agent retrieval enhancement generation system has an advanced self-correction mechanism that can detect potential erroneous information in real time during the question-answering process, correct the erroneous information, and obtain correct information. In addition, the multi-agent retrieval enhancement generation system can also perform adaptive optimization and adjustment for different categories of failure modes, thereby significantly improving the robustness of the system and the quality of question-answering, and enabling the system to automatically apply improved methods in future similar queries, thereby improving the overall intelligence of the system. Through the three core links of error information detection, intelligent correction processing, and adaptive optimization, this application can ensure that the multi-agent retrieval enhancement generation system can dynamically recover and continuously optimize under abnormal circumstances, thereby maintaining efficient and accurate question-answering capabilities.

[0111] The above steps S31 and S32 can be performed as follows: Figure 2 Each step in the embodiment shown. When error information is detected in each step, the above correction processing and adaptive optimization processing can be performed to improve the effectiveness and accuracy of the multi-agent retrieval enhancement generation system. If each step is successfully processed, then according to Figure 2 The steps shown are performed in sequence until the subsequent steps are completed.

[0112] In one embodiment, step S31 includes but is not limited to at least any one of the following: S311, S312, S313, S314.

[0113] S311: When format error information is detected, the format error information is corrected to obtain correct format information.

[0114] When the multi-agent search enhancement generation system detects an input query with formatting anomalies (such as unclear structure or ambiguous keywords), it automatically rewrites the query information or query task to conform to the standard format, ensuring that the query information or query task can be correctly parsed. For example, if the user enters an unstructured query such as "Interpretation of new energy policies in 2025?", the multi-agent search enhancement generation system will rewrite it to "What are the latest policies for the new energy industry in 2025?" to improve the accuracy of the search.

[0115] S312: When it is detected that the retrieval result does not match the user's query intention, the retrieval process is updated to obtain an updated retrieval result.

[0116] If the multi-agent search enhancement system identifies that the search results do not match the user's query intent, it will readjust the search strategy and perform a new search to obtain updated search results. For example, if a user queries for "smart grid architecture" but the system returns content related to "smart home grid," the multi-agent search enhancement system will adjust the query semantics to more accurately match the technical architecture of the smart grid.

[0117] S313, when it is detected that the comprehensive answer is incomplete due to insufficient data source, the data source is updated to obtain an updated comprehensive answer.

[0118] When the multi-agent search and enhancement system detects that the current data source is insufficient, resulting in an incomplete comprehensive answer, it proactively evaluates alternatives and supplements information from other knowledge bases, external data sources, or context to obtain an updated comprehensive answer and ensure its completeness. For example, when querying a technical standard, if the current database does not cover the latest revision, the multi-agent search and enhancement system will automatically attempt to obtain data from other authoritative channels.

[0119] S314, when a task failure is detected, the abnormal task is recorded and a manual review is requested.

[0120] In cases where core tasks fail or permissions are restricted (e.g., requiring professional review or unable to access a specific database), the multi-agent search and augmentation generation system will flag abnormal tasks and request human review to reduce the risk of errors. For example, if a rigorous interpretation of a law or regulation is involved, the system will prioritize requesting expert confirmation to ensure accuracy.

[0121] In one embodiment, step S32 includes but is not limited to at least any one of the following: S321, S322, S323.

[0122] S321, if the failure mode is not obtaining high-quality results, adjust the search parameters and optimize the keywords.

[0123] When high-quality results cannot be obtained after self-correction, the multi-agent retrieval enhancement generation system will dynamically adjust the retrieval parameters, optimize keywords, and combine user feedback data to continuously improve the accuracy of query information.

[0124] S322: If the failure mode is high-frequency failure, store the adjustment record and apply the verified optimization strategy when receiving similar query information.

[0125] For high-frequency failure modes, the multi-agent retrieval enhancement generation system stores adjustment records and automatically applies verified optimization strategies in subsequent similar queries to reduce repetitive errors.

[0126] S323, if the failure mode is reasoning failure, track the question-answering execution path based on the error analysis tool, identify the failure cause, and optimize the reasoning logic.

[0127] The multi-agent retrieval enhancement generation system features error analysis tools that track the execution path of question-answering, identify failure causes, and optimize reasoning logic. For example, in complex reasoning tasks, if missing data causes a break in the logic chain and reasoning failure, the multi-agent retrieval enhancement generation system uses error analysis tools to track the execution path of question-answering, identify the failure cause, and supplement the reasoning path to improve the accuracy of the final answer.

[0128] In the control experiment, a comparative test was conducted on the retrieval enhancement generation system in the prior art and the multi-agent retrieval enhancement generation system proposed in the embodiment of this application. Accuracy tests were conducted on four dimensions: factual question and answer, multi-hop reasoning question and answer, multi-round interactive question and answer, and noise query question and answer. The performance results of the two systems are shown in Table 1 below.

[0129] Table 1:

[0130]

[0131] According to the current test results, the multi-agent retrieval enhancement generation system proposed in the embodiment of the present application has increased the accuracy of the four aspects of factual question and answer, multi-hop reasoning question and answer, multi-round interactive question and answer, and noise query question and answer by 5.42, 14.84, 16.65 and 20.45 percentage points respectively compared with the retrieval enhancement generation system in the prior art. In terms of complexity, the technical solution of the multi-agent retrieval enhancement generation system proposed in the embodiment of the present application has been significantly improved in reasoning ability, context understanding ability and noise resistance. Especially in multi-hop reasoning question and answer and multi-round interactive question and answer tasks, this method can more effectively decompose query tasks, dynamically adjust retrieval strategies, and use agent collaboration to optimize answer quality, thereby significantly improving accuracy. In addition, in the noise query question and answer application scenario, the multi-agent retrieval enhancement generation system proposed in the embodiment of the present application can better understand incomplete or non-standardized input, and improve the accuracy of answers through an adaptive correction mechanism, further enhancing the robustness and practicality of the system.

[0132] The embodiments of this application provide an efficient and scalable multi-agent retrieval enhancement generation system question-answering method, enabling the intelligent question-answering system to adapt to complex query scenarios and achieve more accurate multi-source information fusion. Furthermore, through intelligent detection, adaptive correction, and continuous optimization, the multi-agent retrieval enhancement generation system can continuously improve its capabilities during the question-answering process, ensuring high accuracy and reliability of the question-answering results, thereby providing a more robust and intelligent user interaction experience.

[0133] The multi-agent retrieval enhancement generation system question-answering method provided in the embodiment of the present application introduces a multi-agent architecture, so that the multi-agent retrieval enhancement generation system has autonomous reasoning, dynamic retrieval, self-correction and task collaboration capabilities, thereby overcoming the limitations of the retrieval enhancement generation system in the prior art and improving the reliability and adaptability of intelligent question-answering in the power industry.

[0134] In terms of technical innovation, the question-answering method of the multi-agent retrieval enhancement generation system proposed in the embodiment of this application belongs to a new artificial intelligence paradigm optimized for the complex question-answering needs of the power industry. The multi-agent retrieval enhancement generation system can not only obtain information from external power databases, but also independently plan the reasoning process based on domain knowledge, dynamically optimize query strategies, and flexibly call specific analysis tools for the power industry. Unlike the traditional static "retrieval-reading-generation" model, the multi-agent retrieval enhancement generation system adopts multiple rounds of interaction and intelligent optimization mechanisms to dynamically evaluate retrieval results, adjust query methods, and call professional tools during the answer generation process to ensure high-quality intelligent question-answering services.

[0135] In terms of core technical characteristics, the multi-agent retrieval enhanced generation system question-answering method proposed in the embodiment of this application has the following four characteristics.

[0136] 1. Autonomous decision-making and power knowledge retrieval optimization. The multi-agent retrieval enhancement generation system can flexibly adjust the search scope and depth based on the complexity of the query content, prioritize highly relevant information, and improve the accuracy and efficiency of knowledge retrieval.

[0137] 2. Iterative optimization and self-correction mechanism: The multi-agent retrieval enhancement generation system improves the accuracy of question answering through query rewriting, retrieval method adjustment, and data completion.

[0138] 3. Multi-agent collaborative working mechanism. The multi-agent retrieval enhancement generation system adopts a task division model, where different agents collaborate to complete the question-answering task.

[0139] 4. Autonomous optimization of question-answering reasoning paths. The multi-agent retrieval enhancement generation system can dynamically adjust the reasoning order during the question-answering process, optimizing the answer generation path based on real-time changes in power data to ensure logical rigor and reasoning accuracy.

[0140] In terms of applicable scenarios and industry value, the multi-agent retrieval enhanced generation system question-answering method proposed in the embodiment of this application has significant advantages in the following power industry application scenarios.

[0141] 1. Power equipment fault diagnosis. Combining historical retrieval and real-time monitoring data, the multi-agent retrieval and enhanced generation system question-answering method proposed in the embodiments of this application can provide operation and maintenance personnel with accurate fault cause analysis and treatment suggestions, improving the accuracy of fault diagnosis and response efficiency.

[0142] 2. Professional technical document analysis: The multi-agent retrieval enhanced generation system question-answering method proposed in the embodiment of this application can automatically analyze technical documents in the power industry, such as equipment maintenance manuals and operating procedures, to help engineers quickly obtain key information and improve work efficiency.

[0143] 3. Standard query and policies and regulations. The multi-agent retrieval enhanced generation system question-answering method proposed in the embodiment of this application provides intelligent question-answering for power industry regulations and technical standards, supports compliance review, and ensures the accuracy and efficiency of policy interpretation.

[0144] The multi-agent retrieval enhanced generation system question-answering method proposed in the embodiment of the present application significantly improves the professionalism and practicality of the intelligent question-answering system in the power industry through a multi-agent architecture and dynamic optimization mechanism, and provides accurate and efficient technical support for key businesses such as equipment operation and maintenance, technical document management, and regulatory standards query.

[0145] Figure 4 It is a structural diagram of the multi-agent retrieval enhancement generation system provided in an embodiment of the present application.

[0146] like Figure 4 As shown, the multi-agent retrieval enhancement generation system includes:

[0147] The first determination module 41 is configured to, upon receiving query information submitted by a user, determine one or more query tasks corresponding to the query information based on whether the query information has an in-depth research requirement;

[0148] The routing decision agent 42 is configured to input the one or more query tasks into the tool usage agent according to the query requirements respectively corresponding to the one or more query tasks, and obtain one or more search results output by the tool usage agent;

[0149] A first evaluation module 43 is configured to perform quality evaluation and system optimization based on the one or more search results to obtain a processing result of good quality;

[0150] A second evaluation module 44 is configured to determine whether the processing result satisfies the query information submitted by the user;

[0151] A first summarizing module 45 is configured to summarize the processing results and generate a comprehensive answer if the processing results meet the query information submitted by the user;

[0152] The first providing module 46 is configured to provide the comprehensive answer to the user.

[0153] Another embodiment of the present invention discloses a multi-agent retrieval enhancement generation system. Figure 4 Based on the corresponding embodiment, the multi-agent retrieval enhancement generation system further includes a query planning agent, and the first determination module 41 is used to:

[0154] If the query information has an in-depth research requirement, the query planning agent determines one or more query tasks corresponding to the query information;

[0155] If the query information does not have an in-depth research requirement, the query information is determined as a query task.

[0156] Another embodiment of the present invention discloses a multi-agent retrieval enhancement generation system. Figure 4 Based on the corresponding embodiment, the multi-agent retrieval enhancement generation system further includes:

[0157] The first query module is used to return to the operation of executing the query planning agent to determine one or more query tasks corresponding to the query information if the processing result does not meet the query information submitted by the user.

[0158] Another embodiment of the present invention discloses a multi-agent retrieval enhancement generation system. Figure 4 Based on the corresponding embodiment, the routing decision agent 42 is used to:

[0159] Inputting the one or more query tasks into a tool-using agent, wherein the tool-using agent includes one or more retrieval agents and their corresponding retrieval tools;

[0160] The one or more retrieval agents respectively use corresponding retrieval tools to determine one or more retrieval results corresponding to the one or more query tasks.

[0161] Another embodiment of the present invention discloses a multi-agent retrieval enhancement generation system. Figure 4 Based on the corresponding embodiment, the multi-agent retrieval enhancement generation system further includes at least any one of the following:

[0162] A first correction module is used to correct the erroneous information when erroneous information is detected to obtain correct information;

[0163] The first optimization module is used to perform adaptive optimization according to the category of the failure mode.

[0164] Another embodiment of the present invention discloses a multi-agent retrieval enhancement generation system. Figure 4 Based on the corresponding embodiment, the first correction module is used for at least any one of the following:

[0165] When format error information is detected, correcting the format error information to obtain correct format information;

[0166] When it is detected that the retrieval results do not match the user's query intention, the retrieval process is updated to obtain updated retrieval results;

[0167] When it is detected that the data source is insufficient and the comprehensive answer is incomplete, the data source is updated to obtain an updated comprehensive answer;

[0168] When a task failure is detected, the abnormal task is recorded and manual review is requested.

[0169] Another embodiment of the present invention discloses a multi-agent retrieval enhancement generation system. Figure 4 Based on the corresponding embodiment, the first optimization module is used for at least any one of the following:

[0170] If the failure mode is not obtaining high-quality results, adjust the search parameters and optimize the keywords;

[0171] If the failure mode is high-frequency failure, store the adjustment record and apply the verified optimization strategy when receiving similar query information;

[0172] If the failure mode is reasoning failure, use error analysis tools to track the question-answering execution path, identify the cause of the failure, and optimize the reasoning logic.

[0173] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0174] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0175] The present application also provides an electronic device, such as Figure 5 As shown, the electronic device 5 includes: at least one processor 50, a memory 51, and a computer program 52 stored in the memory 51 and executable on the at least one processor 50. When the processor 50 executes the computer program 52, the steps in any of the above-mentioned method embodiments are implemented.

[0176] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned various method embodiments can be implemented.

[0177] An embodiment of the present application provides a computer program product. When the computer program product is run on a mobile terminal, the mobile terminal can implement the steps in the above-mentioned various method embodiments when executing the computer program product.

[0178] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / electronic device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0179] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0180] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0181] In the embodiments provided in the present application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0182] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0183] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A question-answering method for a multi-agent retrieval enhancement generation system, characterized in that: The multi-agent retrieval enhancement generation system includes a routing decision agent and a tool use agent, and the method includes: When receiving query information submitted by a user, determining one or more query tasks corresponding to the query information based on whether the query information has in-depth research requirements; The routing decision agent inputs the one or more query tasks into the tool use agent according to the query requirements respectively corresponding to the one or more query tasks, and obtains one or more search results output by the tool use agent; Performing quality assessment and system optimization based on the one or more search results to obtain high-quality processing results; Determining whether the processing result satisfies the query information submitted by the user; If the processing results satisfy the query information submitted by the user, summarizing the processing results to generate a comprehensive answer; The comprehensive answer is provided to the user.

2. The method according to claim 1, wherein The multi-agent retrieval enhancement generation system further includes a query planning agent, which determines one or more query tasks corresponding to the query information based on whether the query information has in-depth research requirements, including: If the query information has an in-depth research requirement, the query planning agent determines one or more query tasks corresponding to the query information; If the query information does not have an in-depth research requirement, the query information is determined as a query task.

3. The method according to claim 2, wherein After determining whether the processing result satisfies the query information submitted by the user, the method further includes: If the processing result does not satisfy the query information submitted by the user, the process returns to the step of the query planning agent determining one or more query tasks corresponding to the query information.

4. The method according to claim 1, wherein The step of inputting the one or more query tasks into a tool-using agent and obtaining one or more search results output by the tool-using agent comprises: Inputting the one or more query tasks into a tool-using agent, wherein the tool-using agent includes one or more retrieval agents and their corresponding retrieval tools; The one or more retrieval agents respectively use corresponding retrieval tools to determine one or more retrieval results corresponding to the one or more query tasks.

5. The method according to claim 1, wherein The method further comprises at least one of the following: When erroneous information is detected, the erroneous information is corrected to obtain correct information; Adaptive optimization is performed based on the category of failure mode.

6. The method according to claim 5, wherein When error information is detected, correcting the error information to obtain correct information includes at least one of the following: When format error information is detected, correcting the format error information to obtain correct format information; When it is detected that the retrieval results do not match the user's query intention, the retrieval process is updated to obtain updated retrieval results; When it is detected that the data source is insufficient and the comprehensive answer is incomplete, the data source is updated to obtain an updated comprehensive answer; When a task failure is detected, the abnormal task is recorded and manual review is requested.

7. The method according to claim 5, wherein The adaptive optimization according to the failure mode category includes at least one of the following: If the failure mode is not obtaining high-quality results, adjust the search parameters and optimize the keywords; If the failure mode is high-frequency failure, store the adjustment record and apply the verified optimization strategy when receiving similar query information; If the failure mode is reasoning failure, use error analysis tools to track the question-answering execution path, identify the cause of the failure, and optimize the reasoning logic.

8. A multi-agent retrieval enhancement generation system, characterized in that: The multi-agent retrieval enhancement generation system includes: A first determination module is configured to, upon receiving query information submitted by a user, determine one or more query tasks corresponding to the query information based on whether the query information has an in-depth research requirement; a routing decision agent, configured to input the one or more query tasks into a tool usage agent according to query requirements respectively corresponding to the one or more query tasks, and obtain one or more search results output by the tool usage agent; A first evaluation module is used to perform quality evaluation and system optimization based on the one or more search results to obtain a processing result of good quality; A second evaluation module is used to determine whether the processing result satisfies the query information submitted by the user; A first summarizing module is configured to summarize the processing results and generate a comprehensive answer if the processing results meet the query information submitted by the user; The first providing module is used to provide the comprehensive answer to the user.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The invention comprises a computer program, which, when being executed, enables the method according to any one of claims 1 to 7 to be performed.

Images