Multi-agent cooperative task execution method, device and equipment and storage medium

Abstract

The invention relates to the field of business data distribution, financial science and technology and medical health, and discloses a multi-agent cooperative task execution method, which comprises the following steps: receiving a task instruction input by a user, and carrying out task step configuration on the task instruction to obtain a sub-task sequence; judging whether the task information of each subtask meets a preset execution logic standard, and if yes, sending the subtask to an intelligent agent according to an identifier of the intelligent agent corresponding to the subtask; and recording task data generated when each agent executes the subtask, extracting a second keyword in the task data to construct a decision-making dependency relationship graph among the agents, extracting collaborative data of each agent collaborative completion task instruction from the decision-making dependency relationship graph, and optimizing subsequent task allocation by utilizing the collaborative data. According to the method, the cooperation process among the agents is optimized, and continuous learning and improvement are carried out, so that the overall efficiency and reliability of the multiple agents are improved.

Description

Technical Field

[0001] The present invention relates to the fields of business data distribution, financial technology and medical health, and in particular to a method, device, equipment and storage medium for multiple intelligent agents to collaboratively perform tasks. Background Art

[0002] With the development of artificial intelligence and robotics technology, Multi-Agent System (MAS) has been widely used in many fields such as medical care, industrial automation, environmental monitoring, and financial services.

[0003] For multi-agent systems, it is not only about a single agent's understanding of task instructions, but also requires multiple agents to work together to complete complex tasks. Existing multi-agent systems have the following problems:

[0004] 1. In the actual business operations in the fields of financial technology and healthcare, it is easy to deal with task mismatch problems: for example, when a doctor asks "prepare breakfast for the patients in ward 304 and check their medicines", the existing multi-agent system may not be able to flexibly decompose this instruction into specific subtasks. For example, it is impossible to correctly identify the logical sequence of going to the kitchen to prepare breakfast, then going to the pharmacy to pick up medicines, and finally arriving at ward 304 to distribute food and medicines. This may lead to the wrong assignment of tasks, such as asking a robot responsible for moving and carrying items to handle medicines, while a robot with special equipment to handle medicines is assigned to do the handling work, resulting in resource waste and inefficiency. For another example, when there is a car insurance claim response processing, when a user initiates a car insurance claim application through an APP, the existing multi-agent system may not be able to flexibly decompose this instruction into specific subtasks, such as being unable to correctly assign tasks and intelligently advance processing at nodes such as identity recognition, insurance type determination, material upload guidance, and responsibility identification, resulting in resource waste and inefficiency.

[0005] 2. When executing the above tasks, if the coordination between the various agents is not tight enough, there may be a waste of resources or repeated execution of tasks. For example, two different robots with different tasks may go to the kitchen to prepare breakfast at the same time, causing congestion or resource conflicts; key steps may be omitted, such as forgetting to check the drug information and directly distributing it to the patient, which will obviously pose a threat to the patient's safety. For example, when responding to auto insurance claims, there may be a task conflict caused by the simultaneous activation of on-site inspection tasks and off-site inspection tasks.

[0006] Therefore, how to improve the accuracy of multi-agent collaborative task execution is a technical problem that needs to be solved urgently. Summary of the invention

[0007] In view of the above, it is necessary to provide a method for multi-agent collaborative task execution, the purpose of which is to be able to process complex instructions more accurately, optimize the collaborative process between agents, and continuously learn and improve to improve the overall efficiency and reliability of multi-agents.

[0008] In a first aspect, a method for multi-agent collaborative task execution is provided, comprising:

[0009] Receive a task instruction input by a user, analyze a first keyword contained in the task instruction, obtain a task intent, and obtain task route information associated with the task intent, online status information of multiple agents, and information about an object to be executed of the task instruction from a database;

[0010] According to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instruction, the task instruction is configured with task steps to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of an agent that executes each subtask;

[0011] Determine whether the task information of each subtask meets the preset execution logic standard. If yes, send the subtask to the corresponding agent according to the identifier of the agent corresponding to the subtask;

[0012] Record the task data generated by each intelligent agent when executing the subtask, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

[0013] In a second aspect, a multi-agent collaborative task execution device is provided, comprising:

[0014] A receiving module, configured to receive a task instruction input by a user, analyze a first keyword contained in the task instruction, obtain a task intent, and obtain task route information associated with the task intent, online status information of multiple agents, and information on an object to be executed of the task instruction from a database;

[0015] A configuration module, configured to configure the task steps of the task instruction according to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instruction, so as to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of an agent executing each subtask;

[0016] A judgment module is used to judge whether the task information of each subtask meets the preset execution logic standard. If yes, the subtask is sent to the corresponding agent according to the identifier of the agent corresponding to the subtask;

[0017] The optimization module is used to record the task data generated by each intelligent agent when executing subtasks, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

[0018] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method of multi-agent collaborative task execution when executing the computer program.

[0019] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method of multi-agent collaborative task execution are implemented.

[0020] Compared with the prior art, the present invention parses the task instructions input by the user, extracts the key intentions, and configures according to the information of the preset database to ensure that the task is correctly understood and decomposed into specific subtasks. According to the online status information of each intelligent agent and the needs of the executed object, the task is reasonably allocated to avoid resource waste and duplication of work, and improve resource utilization. Before dispatching the task, each subtask is strictly checked to see if it meets the preset execution logic standards, including the availability of resources to complete the subtask, the feasibility of the intelligent agent to execute the subtask in the physical space, and the compatibility with other task instructions, thereby reducing the risk of errors. The task execution data is recorded, the key information is analyzed, the decision dependency diagram is constructed, and the collaborative pattern is extracted from the historical data to optimize the future task allocation, forming a self-optimizing closed-loop system.

[0021] The problem that resource conflicts or task omissions may occur when multiple agents need to collaborate to complete a task in the prior art is solved. The present invention prevents these problems from occurring by pre-judging the execution logic standard, thereby ensuring efficient collaboration between agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of an application environment of a method for multi-agent collaborative task execution in one embodiment of the present invention;

[0023] Figure 2 A schematic diagram of a flow chart of a method for multi-agent collaborative task execution provided by an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of a module of a multi-agent collaborative task execution device provided by an embodiment of the present invention;

[0025] Figure 4 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;

[0026] Figure 5 It is another structural schematic diagram of a computer device in one embodiment of the present invention.

[0027] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] It should be noted that the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0030] The method for multi-agent collaborative task execution provided by the embodiment of the present invention can be applied in the following aspects: Figure 1In an application environment, the client communicates with the server of the multi-agent system through a network. The server receives the task instruction input by the user, analyzes the first keyword contained in the task instruction, obtains the task intent, obtains the task route information associated with the task intent, the online status information of multiple agents, and the executed object information of the task instruction from the database; configures the task steps of the task instruction according to the task route information, the online status information of multiple agents, and the executed object information of the task instruction, and obtains a subtask sequence, wherein the subtask sequence includes at least two subtasks and the identifier of the agent that executes each subtask; determines whether the task information of each subtask meets the preset execution logic standard, and if so, sends the subtask to the corresponding agent according to the identifier of the agent corresponding to the subtask; records the task data generated by each agent executing the subtask, extracts the second keyword in the task data to construct a decision dependency diagram between each agent, extracts the collaborative data of each agent collaborating to complete the task instruction from the decision dependency diagram, and optimizes the subsequent task allocation using the collaborative data. Among them, the client can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices. The server side can be implemented by an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0031] Reference Figure 2 FIG. 1 is a flow chart of a method for multi-agent collaborative task execution provided by an embodiment of the present invention. The method is executed by a device.

[0032] In this embodiment, the method for multiple agents to collaboratively perform tasks includes:

[0033] S1. Receive a task instruction input by a user, analyze the first keyword contained in the task instruction, obtain the task intent, and obtain the task route information associated with the task intent, the online status information of multiple agents, and the executed object information of the task instruction from a database.

[0034] In this embodiment, the user sends a task instruction to the multi-agent system (MAS) through natural language or other forms (such as input text). For example, in a medical environment, a doctor or nurse may say: "Please prepare breakfast for the patients in ward 304 and check their medications." Or, in a financial scenario, a financial account manager inputs through natural language: "Please buy 500 shares of Company A immediately and make sure the price does not exceed 150 yuan per share.".

[0035] Perform text preprocessing on the task instructions, perform word segmentation on the preprocessed task instructions to obtain a word segmentation list, identify the first keyword associated with the task instructions in the word segmentation list, use a preset intent rule engine to analyze the first keyword, and obtain the task intent.

[0036] The task route information associated with the task intent, the online status information of multiple agents, and the information of the executed object of the task instruction are obtained from the database. The task route information includes the location that each subtask needs to reach and its path planning. The agent online status information includes obtaining the status information of all available agents, such as location, current task, load capacity, etc. The executed object information includes clarifying the specific needs of the service object, such as the patient's ward number, dietary preferences, drug information, etc.

[0037] In other embodiments, the layout of the hospital (such as the location of wards) and any other information critical to completing the task (such as the specific location of the patient, where the medicine is stored, etc.) can be obtained.

[0038] The process of using the planning module of a multi-agent system to receive and parse the user's task instructions, and then prepare for the next task scheduling and execution based on the parsing results and existing data is an important part of realizing intelligent and automated task processing, ensuring that the system can accurately understand the user's request and make appropriate responses and action arrangements accordingly. The planning module is a key component in a multi-agent system (MAS), responsible for receiving task instructions and breaking them down into a series of smaller and more specific subtasks. Its core functions include task parsing, environmental perception, and task decomposition and allocation.

[0039] In one embodiment, analyzing the first keyword contained in the task instruction to obtain the task intent includes:

[0040] Performing text preprocessing on the task instructions, performing word segmentation processing on the preprocessed task instructions, and obtaining a word segmentation list;

[0041] Identify a first keyword associated with the task instruction in the word segmentation list, analyze the first keyword using a preset intention rule engine, and obtain the task intention.

[0042] The text preprocessing of the task instructions includes cleaning and formatting the input text, such as removing punctuation marks, converting to lowercase, standardizing abbreviations, etc.

[0043] The preprocessed task instructions are segmented into individual words or tokens to form a list of segmented words. For the above medical environment example, "prepare breakfast for the patients in ward 304 and check their medications" may be segmented into: ["for", "ward 304", "of", "patients", "prepare", "breakfast", "and", "check", "they", "of", "medicine"]. Or, for the financial scenario example, the task instructions of the financial account manager are analyzed to identify key elements such as "buy", "company A stock", "500 shares", "no more than 150 yuan per share". It obtains relevant information from the database, such as current market conditions, available funds, trading permissions, etc.

[0044] Natural language processing (NLP) technology and a rule engine are used to identify the first keyword associated with the task instruction in the word segmentation list. In the example, the first keyword may be "ward 304", "patient", "prepare breakfast", and "check medicine".

[0045] The first keyword is analyzed using a preset intention rule engine to infer the user's specific intention and obtain the task intention. According to the analyzed task intention, a preset database is accessed to extract necessary information (e.g., task route information, online status information of multiple agents, and information on the executed object of the task instruction) to support the execution of the task.

[0046] S2. Arrange the task steps of the task instruction according to the task route information, the online status information of multiple agents and the information of the executed object of the task instruction to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of the agent that executes each subtask.

[0047] In this embodiment, the complex task is decomposed into a series of specific and operable subtasks according to the analyzed task intention, for example, "go to the kitchen to prepare breakfast", "go to the pharmacy to get medicine", "deliver food and medicine to the patient in the ward 304".

[0048] For each subtask, the most appropriate agent is selected to perform the task based on the agent’s capabilities and current state. Different agents are responsible for different subtasks to optimize efficiency and avoid conflicts.

[0049] These subtasks are arranged in a logical order or time dependency to form an ordered execution sequence. Each subtask in this sequence is assigned to a specific agent, and the collaboration needs between agents are taken into account.

[0050] The final result is a sequence of at least two subtasks, each with a clear executor (i.e., a different agent) and a clear execution order, ensuring that the entire task can be completed efficiently and in a coordinated manner.

[0051] Organize these subtasks into an ordered sequence to ensure that they are executed in the correct order. In some cases, different subtasks can be performed simultaneously (parallel execution), while in other cases, they need to follow a specific order (serial execution). For example, for the above medical environment example, "go to the kitchen to prepare breakfast" must happen before "distribute food to patients". Each subtask will be assigned to the most suitable agent to perform according to its nature and requirements. For example, a robot specially designed for moving and carrying items may be responsible for preparing breakfast and distributing food; while another robot with special equipment to handle medicines will be responsible for picking up and distributing medicines. Or, for the financial scenario example, the customer's task instructions are broken down into several subtasks, including but not limited to checking account balances, setting limit orders, and submitting orders to exchanges. Each subtask is assigned to the agent that is most suitable to perform it - for example, an agent specializing in market data monitoring sets limit prices, and another agent responsible for executing trades submits orders.

[0052] By using the information stored in the database (such as task routes, agent status, and service object needs), the execution steps of the task are carefully arranged to ensure that the task can be broken down into multiple subtasks and executed in the correct order by the most suitable agent. This not only improves the accuracy and efficiency of task execution, but also enhances the flexibility and adaptability of the system, enabling the multi-agent system to operate effectively in a dynamic environment.

[0053] In one embodiment, the task instructions are arranged into task steps according to the task route information, the online status information of the plurality of agents and the information of the executed objects of the task instructions to obtain a subtask sequence, including:

[0054] Decomposing the task intention into a plurality of subtasks according to the task route information and the executed object information;

[0055] According to the online status information of each agent, select agents matching multiple subtasks from the multi-agent system;

[0056] The subtasks are sorted according to the logical order between the subtasks to obtain the subtask sequence.

[0057] According to the task route information (e.g., the path from the kitchen to the ward, the path from the pharmacy to the ward) and the specific needs of the object to be executed (e.g., the specific location of the patient, dietary requirements, medication information, etc.), the task instructions are broken down into smaller and more specific subtasks. For example: Subtask A: Go to the kitchen to prepare breakfast. Subtask B: Go to the pharmacy to get medicine. Subtask C: Deliver food and medicine to the patient in the ward 304.

[0058] Query the online status information of all agents, including their location, whether they are currently idle, special skills (such as the ability to handle medicines), load capacity, etc. Based on this online status information, select the agent that is most suitable for performing each subtask. For example, select an agent with cooking skills to prepare breakfast, and select another agent specializing in medicine handling to go to the pharmacy to pick up medicine.

[0059] Consider the dependencies and temporal order between subtasks. For example, food and medicine must be prepared before delivery, so subtasks A and B must be executed first, and then subtask C. Arrange the subtasks into an ordered sequence according to the above logical order, ensuring that each subtask is executed in the correct order. The final subtask sequence may be as follows: [Subtask B, Subtask A, Subtask C].

[0060] In other embodiments, each subtask step is further subdivided into smaller and more specific subtasks. Continuing with the above example, preparing breakfast can be subdivided into subtasks such as "going to the kitchen", "selecting ingredients", and "cooking food"; checking medicines can be subdivided into "going to the pharmacy", "getting a list of medicines", "checking medicines", etc. According to the logical order and time requirements of the tasks, these subtasks are sorted to form an ordered subtask sequence. Factors considered in sorting include dependencies between tasks (such as having to cook before delivering meals), estimated execution time, and path length. The subtask sequence clarifies the content of each subtask, which agent performs it, and the order of execution, ensuring that all agents can work in a coordinated manner to complete the user's task instructions efficiently and accurately.

[0061] The process of converting a complex task instruction (such as a doctor or nurse's request) into a specific execution plan through a series of steps covers the entire process from understanding the task intent, decomposing the task, selecting the appropriate executors to finally arranging the execution order. This ensures that the task is completed efficiently and accurately, while making full use of the system's resources and the capabilities of the agents, which not only improves the success rate of task execution, but also enhances the flexibility and adaptability of the system, enabling the multi-agent system to operate effectively in complex and dynamic environments.

[0062] S3. Determine whether the task information of each subtask meets the preset execution logic standard. If yes, send the subtask to the corresponding intelligent agent according to the identifier of the intelligent agent corresponding to the subtask.

[0063] In this embodiment, before assigning subtasks to each agent, the criticism module of the multi-agent system is used to verify whether the task information of each subtask meets the execution logic standard. The criticism module is a key component in the multi-agent system (MAS), which is responsible for verifying whether the task information of each subtask meets the execution logic standard before assigning subtasks to each agent. The role of the criticism module is similar to a quality control and optimization link, which is used to discover the shortcomings of the planning module and provide specific improvement suggestions so that the planning module can be quickly adjusted to ensure that each subtask is reasonable, feasible and can be executed correctly.

[0064] Determine whether the task information of each subtask meets the preset execution logic standard, including:

[0065] Resource availability to complete subtasks: Check whether the resources required to complete the subtasks are sufficient and available. For example, if the subtask is "prepare breakfast", you need to confirm that the kitchen has enough ingredients; if it is "get medicine", you need to make sure that the pharmacy has the medicine in stock.

[0066] Feasibility of the agent to perform the subtask in the physical space: Evaluate whether the agent can physically perform the subtask. This includes checking the location of the agent, whether the moving path is obstacle-free, whether the equipment is working properly, etc. For example, a robot responsible for delivering meals must be able to successfully reach ward 304 from the kitchen.

[0067] Compatibility with other task instructions: Make sure the current subtask does not conflict with other ongoing or planned tasks. For example, two robots should not try to use the same equipment or occupy the same channel at the same time.

[0068] Once all of the above logical criteria are met, confirming that the subtask can be performed safely and efficiently, the system will proceed to the next step.

[0069] Each agent has a unique identifier (ID) for identification and communication. The system will find the ID of the agent responsible for performing a specific subtask according to a predetermined allocation scheme, send the specific instructions of the subtask (such as target location, required operations, etc.) to the corresponding agent, and communicate directly through its ID to ensure that the task instructions are accurately conveyed to the correct executor.

[0070] An important verification step before task assignment in a multi-agent system is to ensure that each subtask has been strictly checked for logical standards. Only when a subtask meets the conditions of resource availability to complete the subtask, feasibility of the agent to perform the subtask in physical space, and compatibility with other task instructions will it be formally assigned to a specific agent for execution. This approach not only improves the success rate and efficiency of task execution, but also reduces resource waste and potential safety risks, ensuring the overall performance and reliability of the system.

[0071] In one embodiment, the method further comprises:

[0072] If the task information of a subtask does not meet the preset execution logic standard, the reason for not meeting the preset execution logic standard is analyzed to obtain the analysis result;

[0073] Reconfigure the task steps of the task instruction according to the analysis result.

[0074] Check the task information of each subtask one by one to determine whether it meets the execution logic standards mentioned above. This process can be achieved through technical means such as simulation execution and risk assessment.

[0075] For example, before actually deploying a subtask, a computer model or simulation tool is used to virtually run the subtask. The purpose is to predict problems that may arise during the execution of the task, verify the correctness of the task logic, and test the system response under different conditions.

[0076] Alternatively, risk assessment can be used to identify, analyze, and evaluate factors that may affect mission success, discover and manage risks that may lead to mission failure in advance, and ensure that the mission can proceed smoothly. Specifically, it includes: identifying potential risk points, such as insufficient resources, path conflicts, time constraints, etc. Conducting qualitative and quantitative analysis of each risk to determine its likelihood of occurrence and the extent of its impact. Formulating response strategies, such as preparing alternative plans, adding redundant resources, or setting buffer time. Regularly reviewing risk status and updating assessments based on actual conditions.

[0077] When it is determined that the planning information of a subtask fails to meet the execution logic criteria, the first thing to do is to determine which criteria are not met. For example, it may be insufficient resources, physical inaccessibility, logical errors, or safety violations.

[0078] The information on which criteria were not met is used to form a detailed analysis of why the planning of the subtask failed to meet the expected standards.

[0079] Make necessary adjustments to the task instructions based on the analysis results. This may involve modifying the task's goals, sequence, participating agents, or other relevant parameters to ensure that the new plan meets all execution logic criteria.

[0080] According to the adjusted task instructions, the task is decomposed and planned again to generate a new subtask sequence. This time the planning will take into account the previous problems and try to find a better solution. For example, if the power of an agent is low, you can consider arranging a charging cycle or selecting other agents with sufficient power to complete the task.

[0081] Through careful analysis and targeted adjustments to substandard planning, we can not only correct current problems, but also learn from them to improve the quality and success rate of future task planning. This reflects the progress of multi-agent systems in adaptability and robustness.

[0082] In step S3, it is ensured that each subtask is performed by the most suitable agent, and problems that may be caused by non-compliance with standards, such as task failure, resource waste, or safety hazards, are avoided.

[0083] S4. Record the task data generated by each intelligent agent when executing the subtask, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

[0084] In this embodiment, each agent generates a series of data when performing its assigned subtask, including but not limited to: Task execution results: such as whether the task is successfully completed, problems encountered, etc. Time information: start time, end time, time consumption of each stage. Interaction information: details of interaction with other agents or the environment. Position change information: moving path, stop points, etc. Status information: such as power level, load condition, etc.

[0085] These data are usually recorded in real time by preset programs on the agent, forming a detailed execution log.

[0086] Extract useful information from the execution log and organize it into a form that is easy to analyze. For example, classify information such as timestamps, event types, and participants. Convert task data into more natural language text descriptions that are easy for humans to understand and process. For example, "Robot A arrived in the kitchen at 8:00 and started preparing breakfast."

[0087] The converted text content is processed by natural language processing (NLP) to identify keywords related to the task instructions. These keywords include predetermined task actions (such as "prepare", "check"), objects (such as "breakfast", "medication"), locations (such as "ward 304"), and other terms that are important for understanding the task execution process.

[0088] The keywords can be used to identify dependencies between different agents. These dependencies can be sequential (one task must be completed before another), parallel (multiple tasks can be performed simultaneously but have a common goal or resource), or conditional (the execution of a task depends on the results of other tasks).

[0089] Based on the above dependencies, a decision dependency graph (Directed Acyclic Graph, DAG) is constructed to represent the decision dependencies between agents. In a directed acyclic graph, nodes represent subtasks or agents, and edges represent dependencies between agents. In this way, the collaboration mode and logical relationship between agents during the entire task execution process can be intuitively displayed.

[0090] In one embodiment, recording the task data generated by each agent performing a subtask includes:

[0091] Using the recording program installed in each intelligent agent, the initial data records generated by each intelligent agent when executing the subtask are recorded to obtain the execution log;

[0092] The unstructured or semi-structured log text of the execution log is converted into structured task data, wherein the task data includes task execution results, task completion time, interaction information, and location change information.

[0093] Each agent is pre-installed with one or more specially designed recording programs, which are responsible for monitoring and recording the various activities and state changes of the agent during the execution of subtasks.

[0094] As the agent begins to perform its assigned subtask, the logger keeps track of and records all relevant data points. These data points include, but are not limited to: Timestamp: The exact time each action occurred. Event type: For example, moving, picking up items, delivering items, etc. Interaction information: Interactions with other agents, devices, or the environment. Location information: The location of the agent and its movement path. Status updates: For example, battery level, load status, etc.

[0095] All the above information is organized into an execution log. The execution log is generally in the form of unstructured or semi-structured text, containing a lot of details about the agent's execution process, recording the agent's behavior throughout the task cycle.

[0096] Use preset algorithms or tools to process these unstructured or semi-structured log texts, extract valuable information, and organize it into a structured format.

[0097] The converted task data includes the following key aspects: Task execution results: whether the assigned subtasks were successfully completed, and if failed, what were the reasons. Time to complete the task: the specific time from the start to the end of the task, and the time spent on each stage. Interaction information: details of the interaction between the agent and other entities, such as communication content, collaboration, etc. Position change information: the trajectory of the position change of the agent during the entire task, which is very important for evaluating the effectiveness of path planning.

[0098] Through the preset recording program, the behavioral details of the intelligent agent can be captured and converted into structured task data to facilitate subsequent analysis and optimization. This helps to improve the transparency, traceability and self-improvement capabilities of the multi-agent system and ensure that the intelligent agent can complete tasks more efficiently and accurately.

[0099] In one embodiment, extracting the second keyword from the task data to construct a decision dependency graph between the agents includes:

[0100] Analyzing the task data to identify keywords associated with the task instructions;

[0101] Determine the dependency relationship between different agents according to the keywords, wherein the dependency relationship includes sequential dependency, parallel dependency and conditional dependency;

[0102] A decision dependency graph is constructed according to the dependency relationships, wherein the nodes of the decision dependency graph are represented as subtasks and / or agents, and the edges of the decision dependency graph are represented as dependency relationships between agents.

[0103] Identify secondary keywords related to the task instructions. These secondary keywords can help understand the specific requirements and steps of the task. For example, in a medical environment, words such as "prepare breakfast", "check medication", and "ward 304" are key information that specify the goal, object, and location of the task.

[0104] The second keyword may also include the agent's actions (such as "go to", "distribute"), status (such as "completed", "failed"), and other terms that are important for understanding the task execution process.

[0105] Identify the dependencies between different agents, including: Sequential dependencies: Certain subtasks must be completed in a predetermined order. For example, "prepare food" must precede "distribute food." This dependency ensures the logical correctness and temporal coherence of the task. Parallel dependencies: Multiple subtasks can be performed simultaneously but share the same goal or resources. For example, "prepare breakfast" and "get medicine" can be performed by different agents simultaneously, but both need to be delivered to the patient in "Ward 304" in the end. Conditional dependencies: The execution of one subtask depends on the result of another subtask. For example, the accuracy of "checking medicine" may determine whether "distributing medicine" can proceed.

[0106] In the constructed decision dependency graph (DAG), each node represents a subtask or an agent that performs the subtask, and it can be clearly seen which tasks are responsible for which agent.

[0107] The edges of the decision dependency graph represent the dependencies between agents. For example, if agent A needs to wait for agent B to complete a task before starting its own task, there will be an edge from B to A in the decision dependency graph, indicating this sequential dependency.

[0108] Through the definition of the above dependencies, a decision dependency graph (directed acyclic graph DAG) is constructed to graphically display the collaboration mode between agents, which not only reflects the execution process of the task, but also reveals the interaction and coordination between agents.

[0109] Extract the collaborative data of each agent working together to complete the task instructions from the decision dependency graph, and use the extracted collaborative data to evaluate the performance of the current task execution. For example, did all the agents complete their tasks as expected? Is there any waste of resources or duplication of work?

[0110] By evaluating the effectiveness of the current task execution, areas for improvement are identified, which may involve reallocating tasks, adjusting the configuration of the agent, or modifying the task planning strategy.

[0111] Based on the evaluation of collaborative data, the planning of future tasks can be optimized. For example, for subtasks that often become bottlenecks, you can consider scheduling or adding resources in advance; for tasks that can be processed in parallel, ensure that they are started at the same time to save time.

[0112] In one embodiment, extracting the collaborative data of each agent collaborating to complete the task instruction from the decision dependency graph includes:

[0113] Analyze the correlation between each node of the decision dependency graph to obtain first analysis data;

[0114] Analyze the dependency relationship between the edges of the decision dependency graph to obtain second analysis data;

[0115] A critical path between the first analysis data and the second analysis data is calculated as the collaboration data.

[0116] In the decision dependency graph (DAG), each node represents a subtask or an agent that performs the subtask. The associations between these nodes are analyzed in detail, including the following aspects: Subtask type: Identify the specific task type represented by each node, such as "prepare breakfast", "get medicine", etc. Task execution results: Evaluate whether each subtask is completed successfully, and if it fails, understand the reason for the failure. Time to complete the task: Record the start time, end time, and total time of each subtask to evaluate the efficiency of the task. Position change information: Track the position changes of the agent during the execution of the subtask, which is very important for evaluating the effectiveness and safety of the mobile path. The above analysis generates a set of data about the associations between nodes, namely the first analysis data. This set of first analysis data provides a comprehensive understanding of each subtask and its execution status, and is the basis for optimizing task planning.

[0117] The edges in the decision dependency graph represent dependencies between agents, such as sequential dependencies, parallel dependencies, and conditional dependencies. Identify these edges to determine how different agents or subtasks interact with each other. Sequential dependencies: Certain tasks must be completed in a specific order. Parallel dependencies: Multiple tasks can proceed simultaneously but share the same goal or resources. Conditional dependencies: The execution of one task depends on the outcome of another task.

[0118] By analyzing the dependency relationships of the edges, another set of data, the second analysis data, is generated. This set of second analysis data reveals the collaboration patterns and logical relationships between agents, which is crucial to ensuring the smooth progress of the task.

[0119] The critical path refers to the longest path among all nodes from the start to the end of a task, which determines the shortest completion time of the entire task. Calculating the critical path requires comprehensive consideration of the first analysis data and the second analysis data, for example, considering the execution time and other relevant attributes of each subtask, understanding the dependencies between different agents, and ensuring that tasks are executed in the correct order and logic.

[0120] The calculated critical path is used as collaborative data to optimize future task allocation and execution strategies. The critical path not only helps identify potential bottlenecks and improvement points, but also provides a basis for improving the overall efficiency of the multi-agent system.

[0121] In one embodiment, the optimizing subsequent task allocation by using the collaborative data includes:

[0122] Generate a reward function according to the task instruction and the collaborative data, and score the agent that completes the subtask using the reward function;

[0123] A first positive value is assigned to an agent whose score is greater than or equal to a threshold value as feedback, and a second negative value is assigned to an agent whose score is less than the threshold value as feedback, so as to optimize subsequent task allocation.

[0124] A reward function is generated based on task instructions and operating environment data, and is used to score and provide feedback to agents that complete subtasks. This can motivate agents to perform tasks efficiently and accurately, while promoting self-optimization of multi-agent systems.

[0125] The reward function defines the reward value that the agent should receive after completing the predetermined subtask. This value can be positive (reward) or negative (penalty), depending on whether the agent's performance meets expectations.

[0126] The design of the reward function includes multiple dimensions, such as completion speed, accuracy, collaboration efficiency, etc. Each dimension has a corresponding weight to ensure a comprehensive evaluation of the agent's performance.

[0127] When the agent completes a subtask, a score is calculated based on the preset reward function. This score reflects the agent's performance in this subtask. The scoring process not only considers the result of the task (such as whether it is successfully completed), but may also involve some details of the execution process (such as whether safety regulations are followed, whether resources are used effectively, etc.).

[0128] A score threshold is set in advance to distinguish between agents that perform well and agents that need improvement. For agents whose scores are greater than or equal to the threshold, a first positive value is given as feedback. This is usually a positive number, indicating recognition and encouragement for the agent.

[0129] For agents whose scores are less than the threshold, a second negative value is given as feedback, which means criticism of the agent or a hint for improvement.

[0130] By continuously adjusting the feedback mechanism, the multi-agent system can gradually improve its problem-solving ability and efficiency in practice.

[0131] The whole process of the present invention is described in detail by using a multi-agent collaborative task execution (automated securities trading and risk control) in a specific financial scenario:

[0132] The account manager inputs in natural language: "Please buy 500 shares of Company A immediately, and make sure the price does not exceed 150 yuan per share." The multi-agent system receives and analyzes the customer's task instructions, identifying key elements such as "buy", "Company A stock", "500 shares", and "no more than 150 yuan per share". It obtains relevant information from the database, such as current market conditions, available funds, trading permissions, etc.

[0133] Based on the parsing results, this task is broken down into several subtasks, including but not limited to checking account balances, setting limit orders, and submitting orders to exchanges. Each subtask is assigned to the agent that is best suited to perform it - for example, an agent dedicated to market data monitoring sets the limit price, while another agent responsible for executing transactions submits the order. Before assigning, check whether all necessary conditions are met, such as whether there is enough funds to cover the purchase and whether the internal risk management policy is followed. Once confirmed, send specific execution instructions to the corresponding agent.

[0134] The agents perform their respective subtasks and record all the operation details. These records are used to build a decision dependency graph that shows the collaboration pattern between the agents. The system uses this data to evaluate the performance during the transaction and adjust the future task allocation strategy accordingly to improve efficiency and reduce the error rate. Through the above process, the multi-agent system not only achieves the task of completing securities trading efficiently and accurately, but also enhances the real-time monitoring and control capabilities of potential risks, thereby providing a safer and more reliable financial service experience.

[0135] In step S4, collaborative data is used for self-learning, gradually accumulating experience so as to better cope with similar tasks in the future. This continuous learning and adaptation mechanism enables the multi-agent system to remain efficient and flexible in a constantly changing environment.

[0136] In steps S1-S4, the present invention parses the task instructions input by the user, extracts the key intentions, and configures according to the information of the preset database to ensure that the task is correctly understood and decomposed into specific subtasks. According to the online status information of each intelligent agent and the needs of the executed object, the task is reasonably allocated to avoid resource waste and duplication of work, and improve resource utilization. Before dispatching the task, each subtask is strictly checked to see if it meets the preset execution logic standards, including the availability of resources to complete the subtask, the feasibility of the intelligent agent to execute the subtask in the physical space, and the compatibility with other task instructions, thereby reducing the risk of errors. The task execution data is recorded, the key information therein is analyzed, a decision dependency diagram is constructed, and the collaborative mode is extracted from the historical data to optimize the future task allocation, forming a self-optimizing closed-loop system.

[0137] The problem that resource conflicts or task omissions may occur when multiple agents need to collaborate to complete a task in the prior art is solved. The present invention prevents these problems from occurring by pre-judging the execution logic standard, thereby ensuring efficient collaboration between agents.

[0138] like Figure 3 , which is a schematic diagram of a module of a multi-agent collaborative task execution device provided by an embodiment of the present invention.

[0139] The multi-agent collaborative task execution device 100 of the present invention can be installed in a device. According to the functions to be implemented, the multi-agent collaborative task execution device 100 can include a receiving module 110, a configuration module 120, a judgment module 130 and an optimization module 140. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by a device processor and can complete fixed functions, which are stored in the memory of the device.

[0140] In this embodiment, the functions of each module / unit are as follows:

[0141] The receiving module 110 is used to receive a task instruction input by a user, analyze a first keyword contained in the task instruction, obtain a task intention, and obtain task route information associated with the task intention, online status information of multiple agents, and information on an object to be executed of the task instruction from a database;

[0142] A configuration module 120, configured to configure the task steps of the task instruction according to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instruction, to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of an agent that executes each subtask;

[0143] The judging module 130 is used to judge whether the task information of each subtask meets the preset execution logic standard. If it does, the subtask is sent to the corresponding agent according to the identifier of the agent corresponding to the subtask;

[0144] The optimization module 140 is used to record the task data generated by each intelligent agent when executing a subtask, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

[0145] In one embodiment, the receiving module 110 is specifically configured to:

[0146] Performing text preprocessing on the task instructions, performing word segmentation processing on the preprocessed task instructions, and obtaining a word segmentation list;

[0147] Identify a first keyword associated with the task instruction in the word segmentation list, analyze the first keyword using a preset intention rule engine, and obtain the task intention.

[0148] In one embodiment, the configuration module 120 is specifically configured to:

[0149] Decomposing the task intention into a plurality of subtasks according to the task route information and the executed object information;

[0150] According to the online status information of each agent, select agents matching multiple subtasks from the multi-agent system;

[0151] The subtasks are sorted according to the logical order between the subtasks to obtain the subtask sequence.

[0152] In one embodiment, the determination module 130 is specifically configured to:

[0153] If the task information of a subtask does not meet the preset execution logic standard, the reason for not meeting the preset execution logic standard is analyzed to obtain the analysis result;

[0154] Reconfigure the task steps of the task instruction according to the analysis result.

[0155] In one embodiment, the optimization module 140 is specifically configured to:

[0156] Using the recording program installed in each intelligent agent, the initial data records generated by each intelligent agent when executing the subtask are recorded to obtain the execution log;

[0157] The unstructured or semi-structured log text of the execution log is converted into structured task data, wherein the task data includes task execution results, task completion time, interaction information, and location change information.

[0158] In one embodiment, the optimization module 140 is specifically configured to:

[0159] Analyze the correlation between each node of the decision dependency graph to obtain first analysis data;

[0160] Analyze the dependency relationship between the edges of the decision dependency graph to obtain second analysis data;

[0161] A critical path between the first analysis data and the second analysis data is calculated as the collaboration data.

[0162] In one embodiment, the optimization module 140 is specifically configured to:

[0163] Generate a reward function according to the task instruction and the collaborative data, and score the agent that completes the subtask using the reward function;

[0164] A first positive value is assigned to an agent whose score is greater than or equal to a threshold value as feedback, and a second negative value is assigned to an agent whose score is less than the threshold value as feedback, so as to optimize subsequent task allocation.

[0165] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, it realizes the functions or steps of the service side of a method for multi-agent collaborative execution of tasks.

[0166] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, the functions or steps on the client side of a method for multi-agent collaborative execution of tasks are implemented.

[0167] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0168] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0169] Receive a task instruction input by a user, analyze a first keyword contained in the task instruction, obtain a task intent, and obtain task route information associated with the task intent, online status information of multiple agents, and information about an object to be executed of the task instruction from a database;

[0170] According to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instruction, the task instruction is configured with task steps to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of an agent that executes each subtask;

[0171] Determine whether the task information of each subtask meets the preset execution logic standard. If yes, send the subtask to the corresponding agent according to the identifier of the agent corresponding to the subtask;

[0172] Record the task data generated by each intelligent agent when executing the subtask, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

[0173] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0174] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0175] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. 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.

[0176] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit it. If software tools or components other than those of the company appear in the application embodiments, they are only used for illustrative purposes and do not represent actual use. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for multi-agent collaborative task execution, characterized in that: The method comprises: Receive a task instruction input by a user, analyze a first keyword contained in the task instruction, obtain a task intent, and obtain task route information associated with the task intent, online status information of multiple agents, and information about an object to be executed of the task instruction from a database; According to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instruction, the task instruction is configured with task steps to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of an agent that executes each subtask; Determine whether the task information of each subtask meets the preset execution logic standard. If yes, send the subtask to the corresponding agent according to the identifier of the agent corresponding to the subtask; Record the task data generated by each intelligent agent when executing the subtask, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

2. The method for multi-agent collaborative task execution as claimed in claim 1, characterized in that: The analyzing the first keyword contained in the task instruction to obtain the task intention includes: Performing text preprocessing on the task instructions, performing word segmentation processing on the preprocessed task instructions, and obtaining a word segmentation list; Identify a first keyword associated with the task instruction in the word segmentation list, analyze the first keyword using a preset intention rule engine, and obtain the task intention.

3. The method for multi-agent collaborative task execution as claimed in claim 1, characterized in that: The step of configuring the task instructions according to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instructions to obtain a subtask sequence includes: Decomposing the task intention into a plurality of subtasks according to the task route information and the executed object information; According to the online status information of each agent, select agents matching multiple subtasks from the multi-agent system; The subtasks are sorted according to the logical order between the subtasks to obtain the subtask sequence.

4. The method for multi-agent collaborative task execution as claimed in claim 1, characterized in that: The method further comprises: If the task information of a subtask does not meet the preset execution logic standard, the reason for not meeting the preset execution logic standard is analyzed to obtain the analysis result; Reconfigure the task steps of the task instruction according to the analysis result.

5. The method for multi-agent collaborative task execution as claimed in claim 1, characterized in that: The task data generated by each agent performing a subtask includes: Using the recording program installed in each intelligent agent, the initial data records generated by each intelligent agent when executing the subtask are recorded to obtain the execution log; The unstructured or semi-structured log text of the execution log is converted into structured task data, wherein the task data includes task execution results, task completion time, interaction information, and location change information.

6. The method for multi-agent collaborative task execution as claimed in claim 1, characterized in that: The step of extracting collaborative data of each agent cooperating to complete the task instruction from the decision dependency graph includes: Analyze the correlation between each node of the decision dependency graph to obtain first analysis data; Analyze the dependency relationship between the edges of the decision dependency graph to obtain second analysis data; A critical path between the first analysis data and the second analysis data is calculated as the collaboration data.

7. The method for multi-agent collaborative task execution as claimed in claim 1, characterized in that: The optimizing subsequent task allocation by utilizing the collaborative data includes: Generate a reward function according to the task instruction and the collaborative data, and score the agent that completes the subtask using the reward function; A first positive value is assigned to an agent whose score is greater than or equal to a threshold value as feedback, and a second negative value is assigned to an agent whose score is less than the threshold value as feedback, so as to optimize subsequent task allocation.

8. A multi-agent collaborative task execution device, characterized in that: The device comprises: A receiving module, configured to receive a task instruction input by a user, analyze a first keyword contained in the task instruction, obtain a task intent, and obtain task route information associated with the task intent, online status information of multiple agents, and information on an object to be executed of the task instruction from a database; A configuration module, configured to configure the task steps of the task instruction according to the task route information, the online status information of the plurality of agents and the information of the executed object of the task instruction, so as to obtain a subtask sequence, wherein the subtask sequence includes at least two subtasks and an identifier of an agent executing each subtask; A judgment module is used to judge whether the task information of each subtask meets the preset execution logic standard. If yes, the subtask is sent to the corresponding agent according to the identifier of the agent corresponding to the subtask; The optimization module is used to record the task data generated by each intelligent agent when executing subtasks, extract the second keyword in the task data to construct a decision dependency graph between the intelligent agents, extract the collaborative data of each intelligent agent collaborating to complete the task instructions from the decision dependency graph, and use the collaborative data to optimize subsequent task allocation.

9. A device, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a multi-agent collaborative task execution program that can be executed by the at least one processor, and the multi-agent collaborative task execution program is executed by the at least one processor so that the at least one processor can execute the method for multi-agent collaborative task execution as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a multi-agent collaborative task execution program, which can be executed by one or more processors to implement the multi-agent collaborative task execution method as described in any one of claims 1 to 7.

Images