Multi-agent system

Abstract

The invention provides a multi-agent system, and relates to the technical field of agent task cooperative processing. Comprising a request execution subsystem and a service monitoring subsystem which are in communication connection through a message center module. The service monitoring subsystem is used for collecting target user behavior state data and environment state data and observing and analyzing the target user behavior state data and the environment state data to obtain an observation result and an analysis result; the request execution subsystem is used for processing a request message of a target user according to an observation result, an analysis result and context information of the request message, obtaining a processing result and feeding back the processing result to the target user through the message center module, the user request execution subsystem is of a hierarchical structure of a terminal agent, an intermediate agent and a central agent. According to the method, the problems of low task division efficiency, resource waste and insufficient risk identification capability in the prior art are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent agent task collaborative processing, and in particular to a multi-agent system. Background Art

[0002] Current multi-agent systems generally adopt a flat architecture or a simple hierarchical structure, which has the following significant drawbacks in task processing: Extensive task division and low processing efficiency: The existing system lacks a dynamic task allocation mechanism and cannot automatically match agents based on task complexity (for example, simple tasks still require the central agent to handle them), resulting in wasted resources; Agent collaboration lacks a standardized transfer mechanism, resulting in inefficient cross-domain collaboration and the existence of isolated agent islands: Task transfers between agents in existing systems rely on simple routing rules such as intent classification and lack dynamic matching capabilities based on metadata. When a task exceeds the processing scope of the current agent, it is unable to intelligently select the optimal transfer partner based on dimensions such as "responsibility matching" and "toolset compatibility", often leading to transfer failures or repeated transfers. When users raise cross-domain or complex requirements, the lack of an intermediate coordination mechanism prevents smooth transfers, resulting in the inability of agents to fully utilize their capabilities. Insufficient anomaly monitoring and risk identification capabilities: Traditional systems lack independent oversight systems and lack the ability to detect and adaptively adjust to changes in user and environment status data, process tracking (e.g., user requests that have expired and are not resolved), and other situations in real time. For example, when a terminal agent returns erroneous data, the system cannot promptly trigger a circuit breaker mechanism or dynamically switch to an alternative solution, which can easily lead to task interruption or a degraded user experience. Lack of context: The existing message transmission mechanism does not establish the association between user, session, and request, and cannot effectively track the complete context information of user requests during cross-task collaboration. Summary of the Invention

[0003] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a multi-agent system, which solves the problems of low efficiency in task division processing, waste of resources and insufficient risk identification capabilities in the prior art.

[0004] To achieve the above object, the present invention provides the following solutions: A multi-agent system comprising: Request execution subsystem and service monitoring subsystem that communicate through the message center module; The service monitoring subsystem is used to collect target user behavior status data and environmental status data and observe and analyze the target user behavior status data and environmental status data to obtain observation results and analysis results. The request execution subsystem is used to process the target user's request message based on the observation results, analysis results and context information of the request message, obtain the processing results and feed back the processing results to the target user through the message center module, wherein the user request execution subsystem is a hierarchical structure of terminal intelligent agents, intermediate intelligent agents and central intelligent agents.

[0005] Preferably, the status of the target user's request message includes: The user request has expired and is not resolved, the terminal agent returns error data, the user request is processed normally, and the user request is processed abnormally.

[0006] Preferably, the request message of the target user is multimodal data.

[0007] Preferably, the multimodal data includes: Text data, voice data, image data.

[0008] Preferably, the service monitoring subsystem includes: Observing agents and supervising agents; The observation agent is connected to the supervisory agent through a message center module; The observation agent is used to collect the behavior status data and environmental status data of the target user, and the supervision agent is used to observe and analyze the behavior status data and environmental status data of the target user to obtain observation results and analysis results.

[0009] Preferably, the request execution subsystem includes: Several terminal agents, several intermediate agents, and a central agent, all of which are connected to each other through a message center module; The terminal intelligent agent is used to process the target user's request message in combination with the observation result and the analysis result to obtain a first processing sub-result. If the first processing sub-result is that the processing is completed, a suggestion is generated based on the first processing sub-result and sent to the target user. If the first processing sub-result is not completed, a transfer judgment is performed to obtain a judgment result. If the judgment result is required, the terminal intelligent agent that can process the target user's request is identified based on the context of the request message and the metadata of each intelligent agent in the transfer address book. If the judgment result is not required, the current task is sent to the intermediate intelligent agent. The intermediate intelligent agent is used to reallocate the terminal intelligent agent for the current request message based on the first processing result, obtain multiple second processing sub-results and integrate them to obtain an integrated result. If the integrated result is that the task is completed, a corresponding suggestion is generated and sent to the target user. If the integrated result is not completed, the current request message is sent to the central intelligent agent. The central intelligent agent is used to process the current request message based on the integration result, observation result and analysis result to obtain a third processing sub-result and generate a corresponding suggestion and send it to the target user.

[0010] The present invention discloses the following technical effects: The present invention provides a multi-agent system, comprising: Request execution subsystem and service monitoring subsystem that communicate through the message center module; The service monitoring subsystem is used to collect the target user's behavior status data and environmental status data and observe and analyze the target user's behavior status data and environmental status data to obtain observation results and analysis results. The request execution subsystem is used to process the target user's request message according to the observation results, analysis results and context information of the request message, obtain the processing results and feed back the processing results to the target user through the message center module, wherein the user request execution subsystem is a hierarchical structure of terminal agents, intermediate agents and central agents. The present invention can give full play to the advantages of each agent and improve the overall performance and flexibility of the system through refined agent role division and intelligent collaboration mechanism; the independent supervision system and real-time monitoring capabilities enable the system to discover and handle abnormal situations more promptly, thereby enhancing the stability and reliability of the system; the complete context tracking mechanism ensures the consistency and accuracy of user requests in cross-task collaboration, further improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 A multi-agent data flow diagram provided by an embodiment of the present invention; Figure 2 An agent relationship diagram provided by an embodiment of the present invention; Figure 3 A message structure diagram provided for an embodiment of the present invention; Figure 4 A data flow diagram of a terminal agent directly processing simple tasks provided by an embodiment of the present invention; Figure 5 A flow chart of message processing provided by an agent in an embodiment of the present invention; Figure 6 A relationship diagram of the government affairs intelligent system provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0014] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] like Figure 1-5 As shown, the present invention provides a multi-agent system, comprising: Request execution subsystem and service monitoring subsystem that communicate through the message center module; The service monitoring subsystem is used to collect target user behavior status data and environmental status data and observe and analyze the target user behavior status data and environmental status data to obtain observation results and analysis results. The request execution subsystem is used to process the target user's request message based on the observation results, analysis results and context information of the request message, obtain the processing results and feed back the processing results to the target user through the message center module, wherein the user request execution subsystem is a hierarchical structure of terminal intelligent agents, intermediate intelligent agents and central intelligent agents.

[0016] Specifically, identifying user operating behaviors and environmental parameters includes: using a video image recognition model to identify whether there is a face in front of the camera, and identifying facial expressions and emotions; for example, using a voice model to recognize emotions based on voice; for example, identifying the flow of people in the surrounding environment.

[0017] After identifying the user's operating behavior and environmental parameters, the observation agent saves the observation results, parallel users, and timestamps to the "observation record" library.

[0018] Furthermore, the status of the target user's request message includes: The user request has expired and is not resolved, the terminal agent returns error data, the user request is processed normally, and the user request is processed abnormally.

[0019] Furthermore, the request message of the target user is multimodal data.

[0020] The multimodal data includes: Text data, voice data, image data.

[0021] Specifically, multimodal data generally includes text, voice, and image. Less common multimodal data include smell and taste.

[0022] Other integrated sensors that can provide multimodal data include: temperature, humidity, and air pressure.

[0023] Other multimodal data obtained through further image processing include: user emotions, identity, gender, age, actions, expressions, crowd conditions, surrounding environment, etc.

[0024] Other multimodal data obtained through further audio processing include noise level, voiceprint, etc.

[0025] Furthermore, the service monitoring subsystem includes: Observing agents and supervising agents; The observation agent is connected to the supervisory agent through a message center module; The observation agent is used to collect the behavior status data and environmental status data of the target user, and the supervision agent is used to observe and analyze the behavior status data and environmental status data of the target user to obtain observation results and analysis results.

[0026] Furthermore, the request execution subsystem includes: Several terminal agents, several intermediate agents, and a central agent, all of which are connected to each other through a message center module; The terminal intelligent agent is used to process the target user's request message in combination with the observation result and the analysis result to obtain a first processing sub-result. If the first processing sub-result is that the processing is completed, a suggestion is generated based on the first processing sub-result and sent to the target user. If the first processing sub-result is not completed, a transfer judgment is performed to obtain a judgment result. If the judgment result is required, the terminal intelligent agent that can process the target user's request is identified based on the context of the request message and the metadata of each intelligent agent in the transfer address book. If the judgment result is not required, the current task is sent to the intermediate intelligent agent. The intermediate intelligent agent is used to reallocate the terminal intelligent agent for the current request message based on the first processing result, obtain multiple second processing sub-results and integrate them to obtain an integrated result. If the integrated result is that the task is completed, a corresponding suggestion is generated and sent to the target user. If the integrated result is not completed, the current request message is sent to the central intelligent agent. The central intelligent agent is used to process the current request message based on the integration result, observation result and analysis result to obtain a third processing sub-result and generate a corresponding suggestion and send it to the target user.

[0027] Specifically, the multi-agent system processes user requests through dialogue. The initiation message of a user's request is usually a text or voice question from the user and accompanying multimodal information such as documents and images. A conversation (request information) between a user and a multi-agent system contains multiple requests. A request contains a message sent by the user to the terminal agent, zero to multiple messages between agents, and one to multiple messages sent by agents to the user. Messages between users and agents, messages between agents and users, and messages between agents are sent and received through the message center. Agent metadata includes: responsibilities, callable tool sets, superior agents, and the address book of transferable agents. Service monitoring subsystem The system is composed of an observation agent that collects and records user behavior status data and the status data of its environment; the supervision agent analyzes messages and observation records, and sends the analysis results to users or other agents through the message center; when terminal agents, intermediate agents, and central agents receive messages, they also obtain observation records related to the messages, and process the messages based on the messages and their context and observation records; when processing messages, terminal agents and intermediate agents process the messages according to their own responsibilities and callable tool sets: if they can handle the messages themselves, they send the processing results to the user through the message center; if they cannot handle the messages themselves, they send forwarding messages to other agents or superior agents through the message center.

[0028] Furthermore, the description of the agent through its metadata includes the following dimensions: responsibilities, available tools, superiors, and forwarding address book. Specifically: Responsibilities: Define the core functional scope and mission objectives of each agent. For example, a terminal agent might handle simple user inquiries, while an intermediate agent might coordinate cross-domain tasks. Clearly defining responsibilities ensures that tasks are accurately assigned to the appropriate agent. Toolset: Tools that agents can call to complete tasks, such as MCP services. Different types of agents have different toolsets to meet their needs for handling specific tasks. Superior: The direct superior of the intelligent agent, a higher-level intelligent agent, clarifies the hierarchical relationship of intelligent agents and facilitates the uploading and issuing of tasks. Transfer Directory: A list of agents that can be transferred to when a task exceeds the capabilities of the current agent or needs to be assigned to another agent, along with their applicable scenarios. For example, when a terminal agent encounters a complex task, it can select an appropriate intermediate agent based on the transfer directory.

[0029] Explanation of multi-agent data transmission in the attached figure: Figure 3 Where n represents the number of messages.

[0030] Ma1: The user submits a request message, which contains information such as the user's question and the recipient. This is the starting point of the task, and the user submits a request to the system through this message. Ma2: The terminal agent receives data, including the message sent to it and its context, and the initiating user's observation record. The terminal agent obtains complete information related to the task and prepares to process the task. Ma3: The terminal agent sends a message containing information such as processing results and recipients. If the terminal agent is able to process the task, it sends the processing results to the corresponding recipients, such as the user or the superior agent. Ma4: The intermediate agent receives data, including the message sent to itself and its context, and the initiating user's observation record. The intermediate agent obtains relevant information about the cross-domain task and performs coordination and processing. Ma5: The intermediate agent sends a message containing information such as processing results and recipients. The intermediate agent sends the processed results to the relevant agents or users. Ma6: The central agent receives data, including messages sent to it and their context, as well as observations from the initiating user. The central agent obtains global information and makes important decisions and formulates strategies. Ma7: The central agent sends a message containing information such as processing results and recipients. The central agent sends decisions and strategies to related agents to guide the operation of the system. Ma8: The user receives the message sent to him / her and obtains the task processing result. Mb1: The observing agent sends observation records of the user to provide user status information to the supervisory agent and the central agent. Mb2: Supervisory agents analyze messages and observation records to identify anomalies and risks. Mb3: The supervisory agent sends messages to specific recipients. For example, when a user request is processed abnormally, a notification is sent to the relevant agent or user.

[0031] Furthermore, the request message includes: User ID, session ID, request ID, sender ID, receiver ID, message time, and message body.

[0032] Furthermore, the message metadata is described as follows: User ID: Identifies the user who initiated the conversation and ensures that the system can accurately identify the user. Session ID: Identifies a single continuous interaction process, making it easier to track all user requests and responses in a continuous interaction. Request ID: identifies a single user request and specifies the uniqueness of each request. Sender ID: Identifies the source of the message, which can be a user ID or an agent ID, making it easier to determine the origin of the message. Receiver ID: Identifies the message target, which can be a user ID or agent ID to ensure accurate message delivery. Message time: The time when the message is sent, which records the timestamp of the message and is used for time series analysis and exception tracking. Message body: message content, including specific task information, processing results, etc.

[0033] A user can initiate one or more sessions, a session contains one or more requests, and a request consists of one or more messages. This hierarchical relationship establishes the complete context of user requests, ensuring that the complete information of user requests can be effectively tracked during cross-task collaboration.

[0034] Furthermore, the process of each agent processing messages is as follows: s1: Receive message: The agent receives messages from users or other agents. s2: Processing messages: The agent processes the message according to its own responsibilities and tool set. s3: Determine whether the processing is completed. (Yes) → s4: If the processing is completed, a response message is sent to the user to inform the user of the processing result.

[0035] (No) → s5: If the processing is not completed, enter the judgment and transfer stage. s5: Determine transfer: Determine whether transfer to other agents is necessary. (Yes) → s6: If transfer is required, send a transfer message to other agents based on the transfer address book and select the best transfer target. (No) → s7: If transfer is not required, proceed to the determination and transfer to the upper level. s7: Determine whether to transfer to the superior agent: Determine whether it needs to be transferred to the superior agent. (Yes) → s8: If a transfer to a higher level is required, a transfer message is sent to the higher level agent. (No) → s9: If no transfer to the upper level is required, generate a suggestion. s9: Generate suggestions: Generate suggestions for users based on the processing situation. s10: Send suggestion message to user: Send suggestion to user to provide further guidance.

[0036] Furthermore, Figure 6 As shown, the present invention also provides another embodiment of a government affairs agent system, wherein the terminal agent is a human resources and social security digital human screen, and the built-in model model is: Qwen 7B; the intermediate agents include: a human resources and social security expert agent and a policy expert agent, and the built-in model model of the intermediate agent is Qwen 32B. The observation agent is a digital human screen visual module, and the built-in model of the observation agent is Janus 7B. The central agent is a government affairs agent, and the built-in model model of the central agent is DeepSeek 671B.

[0037] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0038] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A multi-agent system, characterized in that: include: Request execution subsystem and service monitoring subsystem that communicate via the message center module; The service monitoring subsystem is used to collect the target user's behavior status and environmental status and observe and analyze the target user's behavior status and environmental status to obtain observation results and analysis results. The request execution subsystem is used to process the target user's request message based on the observation results, analysis results and context information of the request message, obtain the processing results and feed back the processing results to the target user through the message center module, wherein the user request execution subsystem is a hierarchical structure of terminal intelligent agents, intermediate intelligent agents and central intelligent agents.

2. A multi-agent system according to claim 1, characterized in that: The status of the target user's request message includes: The user request has expired and is not resolved, the terminal agent returns error data, the user request is processed normally, and the user request is processed abnormally.

3. A multi-agent system according to claim 1, characterized in that: The request message of the target user is multimodal data.

4. A multi-agent system according to claim 3, characterized in that: The multimodal data includes: Text data, voice data, image data.

5. A multi-agent system according to claim 1, characterized in that: The service monitoring subsystem includes: Observing agents and supervising agents; The observation agent is connected to the supervisory agent through a message center module; The observation agent is used to collect the behavior status and environmental status of the target user, and the supervision agent is used to observe and analyze the behavior status and environmental status of the target user to obtain observation results and analysis results.

6. A multi-agent system according to claim 5, characterized in that: The request execution subsystem includes: Several terminal agents, several intermediate agents, and a central agent, all of which are connected to each other through a message center module; The terminal intelligent agent is used to process the target user's request message in combination with the observation result and the analysis result to obtain a first processing sub-result. If the first processing sub-result is that the processing is completed, a suggestion is generated based on the first processing sub-result and sent to the target user. If the first processing sub-result is not completed, a transfer judgment is performed to obtain a judgment result. If the judgment result is necessary, the terminal intelligent agent that can process the target user's request is identified based on the context of the consultation task and the metadata of each intelligent agent in the transfer address book. If the judgment result is not necessary, the current task is sent to the intermediate intelligent agent. The intermediate intelligent agent is used to reallocate the terminal intelligent agent for the current request message based on the first processing sub-result, obtain multiple second processing sub-results and integrate them to obtain an integrated result. If the integrated result is that the task is completed, a corresponding suggestion is generated and sent to the target user. If the integrated result is not completed, the current request message is sent to the central intelligent agent. The central intelligent agent is used to process the current request message based on the integration result, observation result and analysis result to obtain a third processing sub-result and generate a corresponding suggestion and send it to the target user.

Images