Multi-agent-based task solving method and system and multi-agent control equipment

Abstract

The invention discloses a multi-agent-based task solving method and system and multi-agent control equipment, and belongs to the technical field of artificial intelligence, and the method comprises the steps: screening target task knowledge related to target task features in a shared knowledge base, and determining candidate agents based on the target task knowledge; each candidate agent calculates the corresponding task fitness based on the current environment state and the target task knowledge, the candidate agent with the maximum task fitness value is determined as the target agent, and the target agent executes the target task; according to the method, the candidate agents capable of executing the target task are determined by taking the knowledge as guidance, and then the specific and appropriate target agent is determined according to the actual situation, so that the knowledge is effectively utilized, and the task execution efficiency is ensured.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a task-solving method, system and multi-agent control device based on multi-agent. Background Art

[0002] Multi-Agent Reinforcement Learning (MARL) is an important branch in the field of reinforcement learning, which focuses on studying how multiple agents optimize their respective decisions through learning in a shared environment. Multi-agent reinforcement learning means that there are multiple agents in an agent system, and these agents learn through interaction with the environment and achieve common goals or solve competitive tasks through mutual interaction. Different from traditional single-agent reinforcement learning, multi-agent reinforcement learning faces more complex problems, such as cooperation and competition, confrontation and collaboration, information sharing and privacy, etc.

[0003] Currently, the efficiency of knowledge transfer is mainly improved by enhancing the learning ability of agents. For example, FLAME (a multi-modal large language model-driven navigation agent) adapts to complex environments by fine-tuning the model. However, when facing new tasks, most existing systems still need to train agents from scratch. That is to say, the existing agents have low efficiency in reusing knowledge, resulting in low efficiency in task execution.

[0004] Therefore, in the process of task execution by agents in the prior art, there is a problem that the task execution efficiency is low due to the low knowledge reuse rate. Summary of the Invention

[0005] In view of this, it is necessary to provide a task-solving method, system and multi-agent control device based on multi-agent to solve the problem of low modeling efficiency in the process of constructing a hydropower station model in the prior art.

[0006] To solve the above problems, the present invention provides a task-solving method based on multi-agent, including: Screening target task knowledge related to the target task characteristics in the shared knowledge base, and determining candidate agents based on the target task knowledge; Each candidate agent calculates the corresponding task fitness based on its current environmental state and the target task knowledge, determines the candidate agent with the maximum task fitness value as the target agent, and the target agent executes the target task.

[0007] In a possible implementation, the task characteristics include task type, goal, and required skills for execution; the task knowledge includes historical tasks, execution environment, and agent capabilities related to the task characteristics; screening for target task knowledge related to the target task characteristics in the shared knowledge base and determining candidate agents based on the target task knowledge includes: Selecting historical tasks in the shared knowledge base with a similarity degree of task characteristics to the target task within a preset range, and determining the agents that executed the historical tasks as candidate agents; Among them, the similarity degree of task characteristics includes the task type similarity coefficient, the goal similarity coefficient, and the matching degree of required skills for execution.

[0008] In a possible implementation, each candidate agent calculates the corresponding task fitness based on its current environmental state and target task knowledge, including: Comparing and determining the fitness coefficient of each candidate agent with the target task according to the fitness parameters, and determining the feasibility coefficient for executing the target task based on the current environmental state of the candidate agent; Determining the task fitness of the candidate agent according to the fitness coefficient and the feasibility coefficient; Among them, the fitness parameters include historical success rate, task completion time, and resource utilization efficiency.

[0009] In a possible implementation, determining the candidate agent with the maximum task fitness value as the target agent and having the target agent execute the target task further includes: Obtaining the environmental feedback data of the target agent in real time during the execution of the target task, and dynamically updating the feasibility coefficient according to the environmental feedback data to obtain the updated feasibility coefficient; Updating the task fitness according to the updated feasibility coefficient.

[0010] In a possible implementation, after the target agent finishes executing the target task, it further includes: Uploading the target task, the target agent, and the environmental feedback data to the shared knowledge base to update the shared knowledge base.

[0011] In a possible implementation, when the shared knowledge base includes historical tasks with exactly the same task characteristics as the target task, preferably selecting the agent that successfully executed the historical task as the target agent; When the agent that successfully executed the historical task cannot be used as the target agent, retrieving the successful experience data from the shared knowledge base to guide the target agent to execute the target task.

[0012] In a possible implementation, after determining the candidate agent with the maximum task fitness value as the target agent and having the target agent execute the target task, it further includes: When the target agent encounters an abnormal situation during the execution of the target task, a second target agent is reselected to execute the target task; Among them, the abnormal situations include that the target agent fails and the current environmental state changes significantly.

[0013] To solve the above problems, the present invention also provides a task-solving system based on multiple agents, including: A candidate agent determination module, configured to screen target task knowledge related to the target task characteristics in the shared knowledge base, and determine candidate agents based on the target task knowledge; A target task execution module, configured to calculate the corresponding task fitness for each candidate agent based on its current environmental state and target task knowledge, determine the candidate agent with the maximum task fitness value as the target agent, and the target agent executes the target task.

[0014] To solve the above problems, the present invention also provides a multi-agent control device, including a memory and a processor, wherein, The memory is used to store programs; The processor is coupled to the memory and is configured to execute the programs stored in the memory to implement the steps in the above-mentioned task-solving method based on multiple agents.

[0015] To solve the above problems, the present invention also provides a computer-readable storage medium for storing computer-readable programs or instructions, and when the programs or instructions are executed by a processor, they can implement the steps in the above-mentioned task-solving method based on multiple agents.

[0016] The beneficial effects of adopting the above embodiments are as follows: The present invention provides a task-solving method based on multiple agents. By taking the target task characteristics as the guide and the shared knowledge base as the query scope, candidate agents related to the target task are obtained, and the candidate agents are used as the candidate subjects for executing the target task, which greatly reduces the scope of subsequent data comparison. Also, because the candidate agents are related to the target task, it is ensured that the relevant knowledge possessed by the candidate agents themselves can be utilized, improving the knowledge reuse rate; further, by comparing the current environmental state and target task knowledge of each candidate agent to determine the task fitness of the candidate agent, that is, to determine the rationality of the candidate agent executing the target task, thereby ensuring the execution efficiency; in summary, in this embodiment, candidate agents capable of executing the target task are first determined based on knowledge, and then specific suitable target agents are determined based on the actual situation, which not only effectively utilizes knowledge but also ensures the execution efficiency of the task. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1Schematic flowchart of an embodiment of the multi-agent-based task solving method provided by the present invention; Figure 2 Schematic flowchart of an embodiment of calculating task fitness provided by the present invention; Figure 3 Schematic flowchart of an embodiment of solving tasks and updating the knowledge base provided by the present invention; Figure 4 Block diagram of an embodiment of the multi-agent-based task solving system provided by the present invention; Figure 5 Block diagram of another embodiment of the multi-agent-based task solving system provided by the present invention; Figure 6 Block diagram of an embodiment of the multi-agent control device provided by the present invention. Detailed implementation manners

[0018] The preferred embodiments of the present invention will be specifically described below with reference to the accompanying drawings. The accompanying drawings form a part of the present invention and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0019] Before presenting the embodiments, the definitions of agents and edge servers are given first: An agent is an intelligent system with perception, decision-making, and execution capabilities. It can sense environmental information through sensors, analyze and make decisions using artificial intelligence algorithms, and intervene and control the environment through actuators. Its characteristics are as follows: 1. Autonomous learning and adaptation: An agent can learn from experience and adjust its behavior according to new information or environmental changes. 2. Problem-solving ability: An agent can identify problems and adopt appropriate strategies to solve them. 3. High efficiency and accuracy: An agent can process a large amount of information in a short time, improve work efficiency, and continuously optimize the algorithm through deep learning to improve the accuracy of recognition and prediction. 4. Personalized service: An agent can provide customized services according to user needs and behavior habits.

[0020] An edge server, usually referred to as an edge server, is a server deployed at the edge of the network. The edge server is located at the edge of the network, that is, close to the user side or the data source. Compared with traditional central servers, the edge server is closer to the user and can provide faster and more reliable services.

[0021] In order to solve the problem in the prior art that the task execution efficiency is low due to the low knowledge reuse rate during the process of task execution by agents, the present invention provides a multi-agent-based task solving method, system, and multi-agent control device, which will be described in detail below respectively.

[0022] AsFigure 1 As shown Figure 1 FIG. is a schematic flowchart of an embodiment of a multi-agent based task solving method provided by the present invention. The multi-agent based task solving method includes: S101: Screen target task knowledge related to target task characteristics in the shared knowledge base, and determine candidate agents based on the target task knowledge; S102: Each candidate agent calculates a corresponding task fitness based on its current environmental state and the target task knowledge, determines the candidate agent with the maximum task fitness value as the target agent, and the target agent executes the target task.

[0023] It should be noted that the shared knowledge base refers to a collection of knowledge about a specific field, which takes experts, paper documents, database data, and network information resources within a specific discipline or topic as the knowledge source, uses knowledge units as the basic storage objects, and is expressed, stored, and managed by a computer.

[0024] In this embodiment, by taking the target task characteristics as the guidance and the shared knowledge base as the query scope, candidate agents related to the target task are obtained, and the candidate agents are used as the candidate subjects for executing the target task, which greatly reduces the scope of subsequent data comparison. Also, because the candidate agents are related to the target task, it is ensured that the relevant knowledge possessed by the candidate agents themselves can be utilized, improving the knowledge reuse rate. Further, by comparing the current environmental state and the target task knowledge of each candidate agent, the task fitness of the candidate agent is determined, that is, the rationality of the candidate agent executing the target task is determined, thus ensuring the execution efficiency. In summary, in this embodiment, first, candidate agents capable of executing the target task are determined based on knowledge, and then specific suitable target agents are determined based on the actual situation, which not only effectively utilizes knowledge but also ensures the execution efficiency of the task.

[0025] It should be noted that the execution entity for obtaining the shared knowledge base in this application can be a storage device, such as: server hard disk, network attached storage (NAS), storage area network (SAN); it can be a processing device, such as: server processor, memory; it can be a network device, such as: router, switch, firewall; it can be an access device, such as: client computer, mobile device; it can also be other auxiliary hardware, such as: backup device, load balancer, etc., which are not limited here.

[0026] The multi-agent based task solving method provided by the embodiments of this application can be applied to a multi-agent based task solving system. The multi-agent based task solving system can be such as MetaGPT, AutoGen, and XAgent, etc., which are not limited here.

[0027] As a preferred embodiment, in S101, the task features include task type, target, and required skills for execution; the task knowledge includes historical tasks, execution environments, and agent capabilities related to the task features; in order to screen for target task knowledge related to the target task features in the shared knowledge base and determine candidate agents based on the target task knowledge, specifically, select historical tasks in the shared knowledge base whose degree of similarity to the task features of the target task is within a preset range, and determine the agents that executed the historical tasks as candidate agents; Among them, the degree of similarity of task features includes a task type similarity coefficient, a target similarity coefficient, and a matching degree of required skills for execution.

[0028] Specifically, first, determine the task type similarity coefficient through the Jaccard similarity coefficient. For task types, the type label of each task can be regarded as a set. For the type label sets A and B of any two tasks, the task type similarity coefficient is calculated by the formula:

[0029] where represents the size of the intersection of set A and set B, that is, the number of common elements in the two sets; represents the size of the union of set A and set B, that is, the number of all different elements in the two sets.

[0030] In addition, it should be noted that the value of the task type similarity coefficient is between 0 and 1, where 1 means the task types are exactly the same, and 0 means the task types are completely different.

[0031] Second, determine the target similarity coefficient through cosine similarity. For task targets, the target of each task can be regarded as a vector. For the targets X and Y of any two tasks, the formula for calculating their target similarity coefficient is:

[0032] where represents the dot product of vector X and vector Y, and represent the norms of vector X and vector Y respectively.

[0033] In addition, it should be noted that the value of the target similarity coefficient is between -1 and 1, where 1 means exactly the same, 0 means no similarity, and -1 means exactly the opposite.

[0034] Third, determine the matching degree of required skills for execution through weighted skill matching. The specific calculation formula is:

[0035] whereT i is the requirement program of the target task on the i th skill, H i is the requirement program of the historical task on the i th skill, W i is the importance weight of the i th skill, refers to the weighted sum of the minimum values of the requirements of two tasks on each skill, refers to the weighted sum of the maximum values of the requirements of two tasks on each skill.

[0036] In addition, it should be noted that the value of the required skill matching degree for execution ranges from 0 to 1, where 1 indicates a perfect match and 0 indicates a complete mismatch.

[0037] In this embodiment, first, based on the task characteristics, relevant historical tasks are screened out in the shared knowledge base. Then, based on the execution environment and the capabilities of the execution agent when executing the historical task, a secondary screening is performed on the agents corresponding to the historical tasks to obtain candidate agents that are more matched with the target task in terms of the execution environment and capabilities, thus better ensuring the quality of the candidate agents.

[0038] It should be noted that there are various types of tasks, which can be divided according to different classification criteria. For example, classified by business scenarios, they can be divided into: design tasks, R & D tasks, marketing tasks, customer service tasks, administrative management tasks, etc. Specifically, they are classified according to the needs and specific circumstances of the target task, and then the target task is classified according to the classification criteria. The classification criteria are not limited here and should be appropriate to meet the actual needs.

[0039] The specific bases for judging whether task goals are similar include: result orientation (directly comparing the expected final results of the tasks), Figure 1 consistency (analyzing the intentions or motives behind the tasks), effectiveness evaluation criteria (considering the criteria used to evaluate the effectiveness of the tasks), stakeholder needs (analyzing the stakeholders served by the tasks and their needs), etc. They can be adaptively set according to specific circumstances and are not limited here.

[0040] Generally, the required skills usually refer to a series of abilities or professional knowledge that an individual or a team must possess to complete a certain task or job. In this embodiment, the required skills for execution refer to the capabilities of the agent to complete the task. In a specific embodiment, the basis for judging the similarity of the required skills for execution can be the remaining battery power of the agent, which is not limited here.

[0041] Further, in S102, in order to calculate the task fitness of each candidate agent, as Figure 2 shown, Figure 2 is a schematic flowchart of an embodiment for calculating task fitness provided by the present invention, including: S201: Determine the fitness coefficient of each candidate agent with respect to the target task according to the fitness parameters comparison, and determine the feasibility coefficient for executing the target task based on the current environmental state of the candidate agent; S202: Determine the task fitness of the candidate agent according to the fitness coefficient and the feasibility coefficient; Among them, the fitness parameters include historical success rate, task completion time, and resource utilization efficiency.

[0042] In this embodiment, first, taking the fitness parameters as a benchmark, calculate the matching relationship between each candidate agent and the target task, such as: by uniformly scoring the historical success rate, task completion time, and resource utilization efficiency of all candidate agents, obtaining the corresponding fitness coefficient; then, based on the current environmental state of each candidate agent, analyze the feasibility of the candidate agent executing the target task, such as: taking the time when each candidate agent reaches the starting point of the target task as the basis for judging feasibility, sorting according to the time size, and obtaining the corresponding feasibility coefficient; finally, superimpose the fitness coefficient and the feasibility coefficient to comprehensively determine the task fitness of the candidate agent.

[0043] Further, since all agents are in a dynamically changing state, therefore, in order to improve the efficiency of task solving, after obtaining the task fitness of each candidate agent, it is also necessary to real-time obtain the environmental feedback data of the target agent when executing the target task, and dynamically update the feasibility coefficient according to the environmental feedback data to obtain the updated feasibility coefficient; Update the task fitness according to the updated feasibility coefficient.

[0044] In this embodiment, by monitoring the environmental feedback data, the feasibility coefficient of the candidate agent is adaptively updated and revised, and then the task fitness of the candidate agent is recalculated, realizing the dynamic adjustment of the task fitness, effectively improving the reliability of the task fitness of the candidate agent, and further effectively improving the reliability of the determined target agent.

[0045] As a preferred embodiment, when determining the target agent, in order to improve the reliability of task completion, when the shared knowledge base includes a historical task that is exactly the same as the task characteristics of the target task, preferably select the agent that successfully executed the historical task as the target agent.

[0046] However, there may be a situation where an agent that has successfully executed a historical task with task characteristics exactly the same as the target task cannot be used as the target agent due to too low a feasibility coefficient. For example, the agent is unavailable, or the agent cannot reach the specified location, etc. To effectively improve the reliability of the target agent in executing the target task, successful experience data is retrieved from the shared knowledge base to the selected target agent to guide the target agent in executing the target task.

[0047] As a preferred embodiment, after determining the target agent, to ensure the execution effect of the task, it is also necessary to monitor the process of the target agent executing the target task, and when the target agent encounters an abnormal situation during the execution of the target task, re-select a second target agent to execute the target task; Among them, the abnormal situation includes that the target agent fails or the current environmental state changes significantly.

[0048] In this embodiment, by monitoring the task execution situation of the target agent, and by repeating the steps of seeking the target agent to select a second target agent, adaptive processing is performed on abnormal situations that may affect the execution effect, thereby effectively ensuring the execution effect of the target task.

[0049] As a preferred embodiment, after the target agent finishes executing the target task, to increase the knowledge volume in the shared knowledge base, it is also necessary to upload the target task, the target agent, and the environmental feedback data to the shared knowledge base to update the shared knowledge base.

[0050] In a specific embodiment, as Figure 3 shown, Figure 3 is a schematic flowchart of an embodiment for solving tasks and updating the knowledge base provided by the present invention. By analyzing the task requirements, targeted task allocation is achieved, and the knowledge and feedback situations involved in the task execution process are uniformly sent to the shared knowledge base.

[0051] Among them, the ATA (Adaptive Task Allocation) algorithm refers to the adaptive task allocation algorithm, which is a method of automatically adjusting the task allocation strategy according to factors such as task requirements, system resources, and constraint conditions in a dynamic environment.

[0052] In the above manner, by performing data UI configuration on the initial hydropower station model, a visualized hydropower station model is obtained, greatly reducing the difficulty of model review and facilitating users to access details such as the structure and component data of the model. As a result, when subsequently adjusting the equipment and parameters of the visualized hydropower station model, the equipment and parameters in the model can be modified quickly and efficiently, improving the efficiency of generating the target three-dimensional hydropower station model. Additionally, by performing three-dimensional data configuration on the model, the intuitiveness, integrity, and authenticity of the model are enhanced, facilitating users to use the model.

[0053] In a specific embodiment, in an intelligent transportation system, the intelligent agents are autonomous vehicles. Multiple autonomous vehicles optimize urban traffic flow through a multi-agent based task-solving method, combining real-time data processing and intelligent collaboration to achieve dynamic traffic management.

[0054] Specifically, the target task is to assign the optimal path and target point to each vehicle; by utilizing real-time traffic data (such as traffic jams and signal changes), the optimal path and target point are assigned to each vehicle, and the route is quickly responded to and adjusted during the process of the vehicle executing the task.

[0055] It should be noted that when an emergency occurs (such as an accident or road closure), the autonomous vehicle calls on previous experience to adjust the path to avoid delays, and traffic information and task status are shared among all autonomous vehicles to ensure the coordination and efficiency of the overall traffic system.

[0056] In this embodiment, by performing capacity scheduling on the autonomous vehicles, the operation of the intelligent transportation system is automatically adjusted, thus ensuring the execution efficiency of the vehicle operation tasks.

[0057] In another specific embodiment, in a complex production workshop, the intelligent agents are industrial robots. The industrial robots utilize adaptive task allocation and knowledge reuse to flexibly respond to changes in production requirements and ensure the continuous operation of the production line.

[0058] Specifically, the target task is to dynamically assign tasks to the industrial robots according to real-time production requirements.

[0059] The industrial robots extract information from the shared knowledge base and quickly respond to changes in the production process. For example, adjust the operation parameters to adapt to new production steps.

[0060] It should be noted that the shared knowledge base includes all the relevant data of the historical tasks executed by the industrial robots.

[0061] In this embodiment, the industrial robots collaborate through a sharing mechanism to avoid work overlap and improve production efficiency.

[0062] In an embodiment of the autonomous navigation of an intelligent UAV swarm in search and rescue missions, the agents are self-driving UAVs. In post-disaster rescue missions, the UAV swarm achieves autonomous navigation and efficient rescue through adaptive task allocation and knowledge reuse.

[0063] The target task is as follows: According to the disaster area map and the locations of trapped people, dynamically allocate the search areas of UAVs, and optimize the paths using historical rescue experience.

[0064] Specifically, the UAVs adjust their flight strategies according to real-time environmental changes (such as obstacles or weather), and use solutions for similar situations to avoid task interruption.

[0065] It should be noted that the shared knowledge base refers to the relevant data for all self-driving UAVs to perform tasks. During the process of optimizing task allocation, multiple UAVs also avoid duplicate searches and improve rescue efficiency by sharing real-time search areas and target information.

[0066] In the above embodiment, by scheduling and controlling a variety of agents, the established tasks are completed to achieve the effective operation of the system.

[0067] To solve the above problems, the present invention also provides a multi-agent-based task solving system, as Figure 4 shown, Figure 4 is a structural block diagram of an embodiment of the multi-agent-based task solving system provided by the present invention. The multi-agent-based task solving system 400 includes: A candidate agent determination module 401, configured to screen target task knowledge related to the target task characteristics in the shared knowledge base, and determine candidate agents based on the target task knowledge; A target task execution module 402, configured to calculate corresponding task fitness for each of the candidate agents based on its current environmental state and the target task knowledge, determine the candidate agent with the maximum task fitness value as the target agent, and have the target agent execute the target task.

[0068] In a specific embodiment, to more specifically illustrate the actual operation of the multi-agent-based task solving system, the multi-agent-based task solving system specifically includes multiple agents and edge servers; Among them, the edge servers are respectively connected to each agent by signals, and the edge servers are also used to dynamically adjust the parameters of the communication protocol with the agents according to the target task execution progress and environmental feedback data.

[0069] In this embodiment, multiple agents transmit data signals through an edge server. Since the edge server directly transmits signals to the agents, by dynamically adjusting the parameters of the communication protocol with the agents according to the progress of the target task execution and the environmental feedback data, the efficiency of the target agent receiving signals can be improved, thereby improving the execution efficiency of the target task.

[0070] In a specific embodiment, to illustrate the data flow process between the agent and the edge server, as Figure 5 shown, Figure 5 FIG. is a structural block diagram of another embodiment of the task solving system based on multiple agents provided by the present invention. By elaborating on the information flow between the agent and the edge server and determining the latter agent B based on the adaptive task allocation algorithm, the process of determining the target agent is realized. Then, the relevant knowledge and experience and the knowledge reuse algorithm used are updated to the shared knowledge base, increasing the amount of knowledge and ensuring the execution efficiency of the task.

[0071] As Figure 6 shown, the present invention also correspondingly provides a multi-agent control device 600. The multi-agent control device 600 includes a processor 601, a memory 602, and a display 603. Figure 6 Only some components of the multi-agent control device 600 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0072] The processor 601 can be a central processing unit (CPU), a microprocessor, or other data processing chips in some embodiments, and is used to run the program code stored in the memory 602 or process data, such as the multi-agent based task solving method in the present invention.

[0073] In some embodiments, the processor 601 can be a single server or a server group. The server group can be centralized or distributed. In some embodiments, the processor 601 can be local or remote. In some embodiments, the processor 601 can be implemented on a cloud platform. In one embodiment, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination of the above.

[0074] The memory 602 can be an internal storage unit of the multi-agent control device 600 in some embodiments, such as the hard disk or memory of the multi-agent control device 600. The memory 602 can also be an external storage device of the multi-agent control device 600 in other embodiments, such as a plug-in hard disk equipped on the multi-agent control device 600, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc.

[0075] Furthermore, the memory 602 can also include both the internal storage unit of the multi-agent control device 600 and the external storage device. The memory 602 is used to store the application software and various types of data for installing the multi-agent control device 600.

[0076] The display 603 can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. in some embodiments. The display 603 is used to display the information of the multi-agent control device 600 and to display a visual user interface. The components 601 - 603 of the multi-agent control device 600 communicate with each other through the system device bus.

[0077] In one embodiment, when the processor 601 executes the multi-agent based task solving program in the memory 602, the following steps can be implemented: Screen the target task knowledge related to the target task characteristics in the shared knowledge base, and determine candidate agents based on the target task knowledge; Each candidate agent calculates the corresponding task fitness respectively based on its current environmental state and the target task knowledge, determines the candidate agent with the maximum task fitness value as the target agent, and the target agent executes the target task.

[0078] It should be understood that when the processor 601 executes the multi-agent based task solving program in the memory 602, in addition to the above functions, other functions can also be implemented. For specific details, please refer to the description of the corresponding method embodiments above.

[0079] Furthermore, the embodiments of the present invention do not specifically limit the type of the multi-agent control device 600 mentioned. The multi-agent control device 600 may be a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop, or other portable multi-agent control devices. Exemplary embodiments of the portable multi-agent control device include, but are not limited to, portable multi-agent control devices equipped with IOS, android, microsoft, or other operating system devices. The above-mentioned portable multi-agent control device may also be other portable multi-agent control devices, such as a laptop with a touch-sensitive surface (such as a touch panel). It should also be understood that in some other embodiments of the present invention, the multi-agent control device 600 may not be a portable multi-agent control device, but a desktop computer with a touch-sensitive surface (such as a touch panel).

[0080] Correspondingly, the embodiments of the present invention also provide a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the task-solving method based on multi-agents provided by the above-mentioned method embodiments can be realized.

[0081] Those skilled in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0082] The above has introduced in detail the task-solving method, system, and multi-agent control device based on multi-agents provided by the present invention. Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A multi-agent task solution method, characterized in that: include: Screening target task knowledge related to target task features in a shared knowledge base, and determining candidate agents based on the target task knowledge; Each candidate agent calculates the corresponding task fitness based on its current environment state and the target task knowledge, determines the candidate agent with the largest task fitness value as the target agent, and the target agent performs the target task.

2. The multi-agent task-solving method according to claim 1, characterized in that: Task characteristics include task type, goal, and skills required for execution; task knowledge includes historical tasks related to task characteristics, execution environment, and execution agent capabilities; The step of screening target task knowledge related to the target task features in the shared knowledge base and determining candidate agents based on the target task knowledge includes: Selecting the historical tasks whose task feature similarity with the target task is within a preset range in the shared knowledge base, and determining the agent that executes the historical tasks as the candidate agent; The task feature similarity includes a task type similarity coefficient, a target similarity coefficient and a matching degree of skills required for execution.

3. The multi-agent task-solving method according to claim 1, characterized in that: Each of the candidate agents calculates the corresponding task fitness based on its current environment state and the target task knowledge, including: Determine the fitness coefficient of each candidate agent and the target task according to the fitness parameter comparison, and determine the feasibility coefficient of executing the target task based on the current environmental state of the candidate agent; Determining the task fitness of the candidate agent according to the fitness coefficient and the feasibility coefficient; The fitness parameters include historical success rate, task completion time and resource utilization efficiency.

4. The multi-agent task-solving method according to claim 3, characterized in that: The step of determining the candidate agent with the largest task fitness value as the target agent, and having the target agent perform the target task, further includes: Acquire environmental feedback data of the target agent when performing the target task in real time, and dynamically update the feasibility coefficient according to the environmental feedback data to obtain an updated feasibility coefficient; The task fitness is updated according to the updated feasibility coefficient.

5. The multi-agent task-solving method according to claim 4, characterized in that: After the target agent completes the target task, the method further includes: The target task, the target agent and the environmental feedback data are uploaded to the shared knowledge base to update the shared knowledge base.

6. The multi-agent task-solving method according to claim 1, characterized in that: When the shared knowledge base includes the historical tasks that are completely consistent with the task features of the target task, an agent that successfully executes the historical tasks is preferably used as the target agent; When the agent that successfully executes the historical task cannot serve as the target agent, successful experience data is retrieved from the shared knowledge base to guide the target agent to execute the target task.

7. The multi-agent task-solving method according to claim 1, characterized in that: After determining that the candidate agent with the largest task fitness value is the target agent, and the target agent performs the target task, the method further includes: When the target agent encounters an abnormal situation during the execution of the target task, reselecting a second target agent to execute the target task; The abnormal situation includes a failure of the target intelligent agent and a significant change in the current environmental state.

8. A multi-agent based task solving system, characterized in that: include: A candidate agent determination module is used to screen target task knowledge related to target task characteristics in a shared knowledge base, and determine candidate agents based on the target task knowledge; The target task execution module is used to calculate the corresponding task fitness of each candidate agent based on its current environment state and the target task knowledge, determine the candidate agent with the largest task fitness value as the target agent, and have the target agent execute the target task.

9. A multi-agent control device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the multi-agent based task solution method described in any one of claims 1 to 7 above.

10. A computer-readable storage medium, characterized in that: Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps of the multi-agent-based task solution method described in any one of claims 1 to 7 above.

Images