Multi-agent based decision method, apparatus, device, and medium

Abstract

The application provides a multi-agent-based decision method, device, equipment and medium, which comprises the following steps: collecting data of multiple data sources, and constructing multiple knowledge bases of different knowledge fields; for different task requirements contained in a target social measure, a corresponding decision model is adopted to construct multiple sub-agents with the knowledge bases of the corresponding knowledge fields; the multiple sub-agents are controlled to formulate respective preliminary social measure implementation schemes for the target social measure, and the multiple preliminary social measure implementation schemes are debated to obtain a final social measure implementation scheme. The application can form multi-field and multi-dimensional insights by integrating multiple sub-agents representing different perspectives to make decisions, and then fuse the social measure implementation schemes of the multiple sub-agents through the debating mechanism of the multiple sub-agents, so that the sub-agents can produce more comprehensive and diversified decision results, and effectively cope with complex and changeable decision problems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-agent modeling, and in particular to a multi-agent based decision-making method, device, equipment and medium. BACKGROUND

[0002] Multi-agent reasoning has shown great potential in simulating human behavior. Multi-agent systems can simulate complex social phenomena, such as social network interactions, individual and group behavior, financial market dynamics, etc. By simulating behavior in different situations, these systems can reveal the underlying mechanisms of social phenomena and provide a basis for human decision-making. Large language models have strong knowledge representation and natural language generation capabilities, and can perform reasoning and learning to some extent, making it possible to build more realistic and credible human-like agents.

[0003] In the past few years, many works have built agents based on large models. For example, researchers created 25 agents in a virtual town and observed their behavior and interactions. Agents can perform daily activities such as making breakfast, going to work, and creating art, as well as complex social interactions such as planning and attending parties, remembering and reflecting on past events, and making future plans. Each agent has a perception system that can capture dynamic changes in the surrounding environment, such as the behavior of other agents, time, and environmental changes. This information is stored in the agent's memory, including short-term memory and long-term memory. Short-term memory stores recent events and information, and long-term memory stores long-term experiences and historical data. Through a feedback loop mechanism, agents can retrieve relevant information from memory to make decisions in the current situation.

[0004] Researchers have proposed the RecAgent framework to study user behavior patterns by simulating user behavior in recommendation systems to improve the performance of recommendation systems. The agent construction part of this framework is divided into three modules: profile module, memory module, and behavior module. The profile module describes the basic characteristics of the user, such as gender, age, personality, interests, etc. The memory module designs three types of memory: perception memory records the user's immediate perception and short-term reaction, short-term memory saves recent events and user behavior, and long-term memory stores the user's long-term preferences and historical behavior. The action module is responsible for simulating various behaviors of users in the recommendation system, such as browsing recommended content, interacting with other users, posting messages on social media, etc. By generating these behaviors through large language models, user behavior simulation can be made more realistic.

[0005] However, existing methods model a target through a single agent, which has the following shortcomings:

[0006] (1) Perspective limitation: Single-agent models are often limited by their inherent perspective when simulating complex real-world scenarios. While large models have powerful capabilities, they may also carry biases or fail to fully consider the multi-domain, multi-dimensional elements and diverse perspectives involved in a problem. This single-perspective modeling approach cannot accurately reproduce the multi-angle thinking and comprehensive decision-making process exhibited by humans when facing complex problems, thereby limiting the effectiveness and universality of the model in solving complex decision-making problems.

[0007] (2) Illusions and errors: Single-agent decision-making is easily influenced by the inherent illusions and false assumptions within the large model, i.e., the agent generates seemingly reasonable but actually incorrect conclusions in the absence of sufficient evidence. Such errors not only reduce the reliability of the agent in reasoning, but also may mislead the decision-making process.

[0008] (3) Poor interpretability: The decision-making process of complex problems exhibits highly nonlinear and invisible internal mechanisms, lacking transparency and being difficult to understand and explain. SUMMARY

[0009] The present application provides a multi-agent-based decision-making method, device, equipment and medium to solve the problem that a single agent is limited by its inherent perspective when simulating complex real-world scenarios in related technologies, limiting the effectiveness and universality of the agent in solving complex decision-making problems.

[0010] The present application provides a multi-agent-based decision-making method, device, equipment and medium to solve the problem that a single agent is limited by its inherent perspective when simulating complex real-world scenarios in related technologies, limiting the effectiveness and universality of the agent in solving complex decision-making problems.

[0011] Collecting data from multiple data sources to build multiple knowledge bases in different knowledge domains;

[0012] For different task requirements contained in the target social measures, a corresponding decision-making model is adopted to build multiple sub-agents with the knowledge base of the corresponding knowledge domain;

[0013] Control multiple sub-agents to develop their own preliminary social measures implementation plans, and debate multiple preliminary social measures implementation plans to obtain the final social measures implementation plan.

[0014] According to the multi-agent-based decision-making method provided by the present application, data from multiple data sources is collected to build multiple knowledge bases in different knowledge domains, including:

[0015] Connect multiple different types of data sources;

[0016] According to the data stored in the multiple different types of data sources, divide the multiple knowledge domains;

[0017] For each knowledge domain, use the data in the corresponding knowledge domain to build a knowledge base.

[0018] According to the multi-agent based decision-making method provided by the application, for different task requirements and different knowledge fields contained in the target social measure, a plurality of sub-agents are constructed by using corresponding decision-making models and knowledge bases of corresponding knowledge fields, including:

[0019] According to the target social measure, shared knowledge is extracted from a plurality of knowledge bases;

[0020] For each task requirement, a decision-making model corresponding to the task requirement is determined;

[0021] For each task requirement, the perspective features, domain knowledge and thinking methods required to meet the task requirement are extracted from the knowledge base of the corresponding knowledge field to generate independent knowledge corresponding to the task requirement;

[0022] Based on the shared knowledge, for each task requirement, a sub-agent is constructed by using the decision-making model corresponding to the task requirement and the corresponding independent knowledge.

[0023] According to the multi-agent based decision-making method provided by the application, a plurality of sub-agents are controlled to formulate respective preliminary social measure implementation schemes for the target social measure, and a final social measure implementation scheme is obtained by debating a plurality of preliminary social measure implementation schemes, including:

[0024] Control a plurality of sub-agents to formulate preliminary social measure implementation schemes that meet respective corresponding task requirements for the target social measure;

[0025] Control each sub-agent to demonstrate the preliminary social measure implementation scheme formulated by itself to obtain a demonstration result, and the demonstration includes determining the decision-making basis, reasoning process, expected effect and potential risks of the preliminary social measure implementation scheme;

[0026] For each sub-agent, control the sub-agent to refute the preliminary social measure implementation schemes formulated by other sub-agents to obtain a refutation result, and the refutation includes vulnerability detection and refutation scoring of the preliminary social measure implementation scheme;

[0027] Control each sub-agent to adjust the preliminary social measure implementation scheme according to the demonstration result and the refutation result of the preliminary social measure implementation scheme formulated by itself to obtain a revised implementation scheme formulated by each sub-agent;

[0028] Repeat the steps of demonstration, refutation and revision until the demonstration result and / or refutation result of the revised implementation scheme meets the preset condition, and then determine the final social measure implementation scheme according to the revised implementation scheme currently obtained by each sub-agent.

[0029] According to the method, the sub-agent is controlled to demonstrate the preliminary social measure implementation scheme formulated by each sub-agent, and a demonstration result is obtained, including:

[0030] The sub-agent is controlled to analyze the preliminary social measure implementation scheme according to the decision basis and reasoning process recorded when the preliminary social measure implementation scheme is generated;

[0031] The sub-agent is controlled to score the preliminary social measure implementation scheme, and the expected effect and potential risks of the preliminary social measure implementation scheme are determined according to the score of the preliminary social measure implementation scheme.

[0032] According to the method, the sub-agent is controlled to demonstrate the preliminary social measure implementation scheme formulated by each sub-agent, and a demonstration result is obtained, including:

[0033] The sub-agent is controlled to analyze the preliminary social measure implementation scheme formulated by other sub-agents, and the vulnerabilities and rebuttal scores of the preliminary social measure implementation scheme formulated by other sub-agents are determined according to the analysis result.

[0034] According to the method, the sub-agent is controlled to demonstrate the preliminary social measure implementation scheme formulated by each sub-agent, and a demonstration result is obtained, including:

[0035] The feedback information of the user on the final social measure implementation scheme is received;

[0036] The sub-agent is adjusted according to the feedback information, and the adjustment content at least includes one of the number of sub-agents, the type of sub-agent, shared knowledge, independent knowledge, the type of decision model, the demonstration mode and the rebuttal mode.

[0037] The application further provides a decision device based on multiple agents, comprising the following modules:

[0038] The data acquisition module is used for acquiring data of multiple data sources and constructing multiple knowledge bases of different knowledge fields;

[0039] The sub-agent generation module is used for constructing multiple sub-agents by adopting corresponding decision models and knowledge bases of corresponding knowledge fields according to different task requirements contained in the target social measure;

[0040] The debate module is used for controlling multiple sub-agents to formulate respective preliminary social measure implementation schemes for the target social measure, and performing debate on the multiple preliminary social measure implementation schemes to obtain a final social measure implementation scheme.

[0041] According to the decision device based on multiple agents, the data acquisition module is specifically used for:

[0042] connecting a plurality of different types of data sources;

[0043] dividing a plurality of knowledge fields according to data stored in the plurality of different types of data sources;

[0044] constructing a knowledge base for each knowledge field using data belonging to the knowledge field.

[0045] According to the multi-agent based decision device provided by the application, the sub-agent generation module is specifically used for:

[0046] extracting shared knowledge from the plurality of knowledge bases according to the target social measure;

[0047] determining a decision model corresponding to each task requirement;

[0048] extracting perspective features, domain knowledge and thinking modes required to meet the task requirement from the knowledge base of the corresponding knowledge field for each task requirement, and generating independent knowledge corresponding to the task requirement;

[0049] using the decision model corresponding to each task requirement and the corresponding independent knowledge to construct a sub-agent based on the shared knowledge.

[0050] According to the multi-agent based decision device provided by the application, the debate module is specifically used for:

[0051] controlling a plurality of sub-agents to develop a preliminary social measure implementation scheme meeting respective corresponding task requirements for the target social measure;

[0052] controlling each sub-agent to demonstrate the preliminary social measure implementation scheme developed by itself to obtain a demonstration result, and the demonstration includes determining the decision basis, reasoning process, expected effect and potential risks of the preliminary social measure implementation scheme;

[0053] controlling each sub-agent to refute the preliminary social measure implementation scheme developed by other sub-agents to obtain a refutation result, and the refutation includes vulnerability detection and refutation score of the preliminary social measure implementation scheme;

[0054] controlling each sub-agent to adjust the preliminary social measure implementation scheme according to the demonstration result and the refutation result of the preliminary social measure implementation scheme developed by itself to obtain a revised implementation scheme developed by each sub-agent;

[0055] repeating the steps of demonstration, refutation and revision until the demonstration result and / or refutation result of the revised implementation scheme meet a preset condition, and then determining a final social measure implementation scheme according to the revised implementation scheme currently obtained by each sub-agent.

[0056] According to the application, a multi-agent-based decision device is provided, wherein the argumentation module controls each sub-agent to argue for a preliminary social measure implementation scheme formulated by the sub-agent, and when an argumentation result is obtained, the argumentation module is specifically configured to:

[0057] control the sub-agent to analyze the preliminary social measure implementation scheme according to a decision basis and a reasoning process recorded when the preliminary social measure implementation scheme is generated;

[0058] control the sub-agent to score the preliminary social measure implementation scheme, and determine an expected effect and a potential risk of the preliminary social measure implementation scheme according to the score of the preliminary social measure implementation scheme.

[0059] According to the application, a multi-agent-based decision device is provided, wherein the argumentation module controls each sub-agent to argue for a preliminary social measure implementation scheme formulated by the sub-agent, and when an argumentation result is obtained, the argumentation module is specifically configured to:

[0060] for each sub-agent, the argumentation module controls the sub-agent to analyze the preliminary social measure implementation scheme formulated by other sub-agents, and determines a loophole and a rebuttal score of the preliminary social measure implementation scheme formulated by other sub-agents according to an analysis result.

[0061] According to the application, a multi-agent-based decision device is provided, wherein the device further comprises a feedback module, and the feedback module is specifically configured to:

[0062] receive feedback information of a user on a final social measure implementation scheme;

[0063] adjust the sub-agents according to the feedback information, and the adjustment content at least includes one of a number of sub-agents, a type of sub-agent, shared knowledge, independent knowledge, a type of decision model, an argumentation manner and a rebuttal manner.

[0064] The application further provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements any of the above multi-agent-based decision methods when executing the program.

[0065] The application further provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement any of the above multi-agent-based decision methods.

[0066] The application further provides a computer program product, which comprises a computer program, and the computer program is executable on a processor to implement any of the above multi-agent-based decision methods.

[0067] The method for making decisions based on multiple agents provided by the application first collects data of multiple data sources, constructs multiple knowledge bases of different knowledge fields, then constructs multiple sub-agents by corresponding decision models and knowledge bases of corresponding knowledge fields according to different task requirements contained in the target social measures, and finally controls the multiple sub-agents to make respective preliminary social measure implementation schemes for the target social measures, and debates the multiple preliminary social measure implementation schemes to obtain a final social measure implementation scheme. By using the method provided by the embodiment of the application, multiple sub-agents can be integrated, and since each sub-agent is constructed by a knowledge base of a different knowledge field and a decision model corresponding to a task requirement, each sub-agent can represent a different perspective, and the decision by the integrated multiple sub-agents can form multi-field and multi-dimensional insights, and then the social measure implementation schemes of the multiple sub-agents are fused through the debating mechanism of the multiple sub-agents, so that the agents can produce more comprehensive and diversified decision results to effectively cope with complex and changeable decision problems. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0069] Figure 1 is a flowchart of the method for making decisions based on multiple agents provided by the application.

[0070] Figure 2 is a structural diagram of the multiple agents provided by the application.

[0071] Figure 3 is a structural diagram of the decision device based on multiple agents provided by the application.

[0072] Figure 4 is a structural diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0073] In order to make the objects, technical solutions and advantages of the application clearer, the technical solutions in the application will be described clearly and completely below with reference to the drawings in the application. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0074] Multi-agent reasoning has shown great potential in simulating human behavior. Multi-agent systems can simulate complex social phenomena, such as social network interactions, individual and group behavior, financial market dynamics, and more. By simulating behavior in different situations, these systems can reveal the underlying mechanisms of social phenomena and provide a basis for human decision-making. Large language models have strong knowledge representation and natural language generation capabilities, enabling a certain degree of reasoning and learning, which makes it possible to build more realistic and credible human-like agents.

[0075] In the past few years, many works have built agents based on large models. For example, researchers created 25 agents in a virtual town and observed their behavior and interactions. Agents can perform daily activities such as making breakfast, going to work, and creating art, as well as complex social interactions such as planning and attending parties, remembering and reflecting on past events, and making future plans. Each agent has a perception system that can capture dynamic changes in the surrounding environment, such as the behavior of other agents, time, and environmental changes. This information is stored in the agent's memory, including short-term memory and long-term memory. Short-term memory stores recent events and information, while long-term memory stores long-term experiences and historical data. Through a feedback loop mechanism, agents can retrieve relevant information from memory to make decisions in the current situation.

[0076] Researchers et al. proposed the RecAgent framework to study user behavior patterns by simulating user behavior in recommendation systems to improve the performance of recommendation systems. The agent construction part of this framework is divided into three modules: profile module, memory module, and behavior module. The profile module describes the basic characteristics of the user, such as gender, age, personality, interests, and other information; the memory module designs three types of memory: perception memory records the user's immediate perception and short-term reaction, short-term memory saves recent events and user behavior, and long-term memory stores the user's long-term preferences and historical behavior; the action module is responsible for simulating various behaviors of users in the recommendation system, such as browsing recommended content, interacting with other users, and posting messages on social media. By generating these behaviors through large language models, user behavior simulation can be made more realistic.

[0077] However, existing methods are based on a single agent modeling a target, which has the following shortcomings:

[0078] (1) Perspective limitation: Single-agent models are often limited by their inherent perspective when simulating complex real-world scenarios. Although large models have strong capabilities, they may also have biases or fail to fully consider the multi-domain, multi-dimensional elements and diverse perspectives involved in the problem. This single-perspective modeling approach cannot accurately reproduce the multi-angle thinking and comprehensive decision-making process that humans exhibit when faced with complex problems, limiting the effectiveness and universality of the model in solving complex decision-making problems.

[0079] (2) Illusions and errors: when a single agent makes decisions, it is easily affected by the inherent illusions and error assumptions of the large model, that is, the agent generates seemingly reasonable but actually incorrect conclusions in the absence of sufficient basis. Such errors not only reduce the reliability of the agent in deduction, but also may have misleading effects on decision-making.

[0080] (3) Poor interpretability: the decision-making process of complex problems exhibits highly nonlinear and invisible internal mechanisms, lacks transparency, and is difficult to understand and explain.

[0081] Therefore, there is an urgent need for a multi-agent-based decision-making method to solve the above problems.

[0082] The following will be described in conjunction with Figures 1-4 The multi-agent-based decision-making method, device, equipment and medium of the present application are described.

[0083] Figure 1 is one of the flowcharts of the multi-agent-based decision-making method provided by the present application, as Figure 1 shown, the method comprises the following steps:

[0084] Step 101, collect data of multiple data sources, and construct multiple knowledge bases of different knowledge fields.

[0085] This step can be implemented by a data collection module, which is designed in a modular manner, supports flexible configuration and extension, and can connect and access multiple types of data sources. These data sources include but are not limited to real-time environmental monitoring systems, historical databases, domain knowledge bases, social media platforms, etc., covering a wide range of information fields from physical environment to social culture. The data in these data sources provides the necessary basic information and background knowledge for decision-making, providing necessary data support for subsequent agent generation and decision-making.

[0086] In some embodiments, collecting data of multiple data sources and constructing multiple knowledge bases of different knowledge fields comprises: connecting multiple data sources of different types; dividing multiple knowledge fields according to the data stored in the multiple data sources of different types; for each knowledge field, constructing a knowledge base using the data of the knowledge field.

[0087] First, identify and determine the data sources that need to be connected. These data sources can include structured data, semi-structured data, unstructured data, and real-time data, etc. The data types in these data sources can include pictures, videos, text data, etc. Design appropriate data interfaces for each type of data source, and use data integration platforms or tools to integrate data from different sources, and convert the data of different data sources into a unified format for processing and analysis.

[0088] By analyzing the data content, the theme and field of the data are identified. For example, social media data can involve sentiment analysis, topic classification, etc., while sensor data can involve environmental monitoring, device status, etc. According to the theme and content of the data, a knowledge field model is defined. A knowledge field can be a specific discipline or business field, such as medical health, financial market, user behavior, etc. Then use clustering algorithms such as K-means clustering, hierarchical clustering, etc. Group the data, each group represents a knowledge field, and then apply topic modeling techniques to identify the topics in the data, which helps to divide the knowledge field.

[0089] Finally, a knowledge base is designed and built for each knowledge field, relevant data is extracted and loaded into the knowledge base, and maintained.

[0090] The embodiments of the present disclosure can effectively integrate and utilize information from different data sources, create useful knowledge bases, and support subsequent decision-making processes.

[0091] Step 102, for different task requirements contained in the target social measures, a plurality of sub-agents are constructed by corresponding decision models and knowledge bases of corresponding knowledge fields.

[0092] This step can be implemented by a sub-agent generation module, which divides, designs and constructs a plurality of sub-agents according to task requirements and knowledge field division. Each sub-agent is given a specific perspective, knowledge base and reasoning rule, and can independently complete preliminary analysis and evaluation of the target problem. The reasoning rule can be implemented using a decision model, and the perspective information of the sub-agent is determined by the task requirements in combination with the knowledge base.

[0093] Taking the decision task of driving path planning as an example, the target decision measure of this task is to plan a reasonable driving path for the user. This target decision measure can contain many different task requirements, such as the shortest driving path, the shortest expected red light duration in the driving path, the smallest probability of congestion in the driving path, etc. For the above three path planning task requirements, the sub-agent generation module can select the knowledge base required to meet these three task requirements, for example, the shortest driving path task requirement requires a path length knowledge base, the shortest expected red light duration in the driving path task requirement requires a red light time knowledge base, and the smallest probability of congestion in the driving path task requirement requires a historical congestion road condition knowledge base. It is not difficult to understand that the sub-agent plans the path for these task requirements, and these task requirements contain the perspective information of the sub-agent, which is to plan for the shortest path, to plan for the shortest expected red light duration in the driving path, and to plan for the smallest probability of congestion.

[0094] The corresponding decision model can adopt a machine learning model. For example, the task requirement of the shortest path can adopt a machine learning model trained with the target output of the shortest path planning path, the task requirement of the shortest red light time can adopt a machine learning model trained with the target output of the shortest red light time planning path, and the task requirement of the minimum congestion probability can adopt a machine learning model trained with the target output of the minimum congestion probability planning path.

[0095] The sub-agent 210 can be constructed in combination with the task requirement of the shortest path, the knowledge base containing the path distance, and the decision model 211 with the target output of the shortest path planning path. The sub-agent 220 can be constructed in combination with the task requirement of the shortest red light time, the knowledge base containing the red light time, and the decision model 221 with the target output of the shortest red light time planning path. The sub-agent 230 can be constructed in combination with the task requirement of the minimum congestion probability, the knowledge base containing the historical congestion road condition, and the decision model 231 with the target output of the minimum congestion probability planning path.

[0096] In some embodiments, for different task requirements and different knowledge fields contained in the target social measure, a plurality of sub-agents are constructed by adopting corresponding decision models and knowledge bases of corresponding knowledge fields, including: extracting shared knowledge from the plurality of knowledge bases according to the target social measure; determining a decision model corresponding to each task requirement; extracting perspective features, domain knowledge, and thinking modes required to meet the task requirement from the knowledge base of the corresponding knowledge field for each task requirement to generate independent knowledge corresponding to the task requirement; and constructing a sub-agent for each task requirement using the corresponding decision model and the corresponding independent knowledge of the task requirement based on the shared knowledge.

[0097] The shared knowledge is the common basis of all sub-agents, ensuring their basic consistency and collaboration. For example, in the above-mentioned path planning task scenario, the map information can be used as the shared knowledge 201 of all sub-agents.

[0098] On the basis of the shared knowledge, each sub-agent has independent knowledge, which contains the perspective features, domain knowledge, and thinking modes specific to the sub-agent. For example, the independent knowledge 212 of the sub-agent 210 can be the domain knowledge in the knowledge base containing the path distance, the independent knowledge 222 of the sub-agent 220 can be the domain knowledge in the knowledge base containing the red light time, and the independent knowledge 232 of the sub-agent 230 can be the domain knowledge in the knowledge base containing the historical congestion road condition. Thus, the target agent 200 generated by the sub-agent generation module for the driving path planning task is as shown in Figure 2 .

[0099] Each sub-agent is controlled by a respective decision model to make decisions, so that each sub-agent can understand the problem, analyze situational information and make decisions in a unique way, thereby promoting diversity and innovation within the agent system.

[0100] The embodiments of the present disclosure ensure that the generated sub-agents meet the overall needs of the task and have unique personalities and abilities by generating sub-agents with different characteristics.

[0101] Step 103, control multiple sub-agents to formulate respective preliminary social measure implementation schemes for the target social measure, and debate the multiple preliminary social measure implementation schemes to obtain a final social measure implementation scheme.

[0102] This step can be implemented by a debate module, which controls each sub-agent to formulate a preliminary social measure implementation scheme according to its unique perspective and decision-making method, and then allows multiple sub-agents to simulate the human decision-making process to debate. Through simulating the complex collision of thoughts, it is ensured that the social measure implementation schemes obtained from different perspectives and reasoning can be fully discussed and coordinated, so that the final social measure implementation scheme is obtained by fusing multiple preliminary social measure implementation schemes.

[0103] In some embodiments, controlling multiple sub-agents to formulate respective preliminary social measure implementation schemes for the target social measure, and debating the multiple preliminary social measure implementation schemes to obtain a final social measure implementation scheme, includes: controlling multiple sub-agents to formulate preliminary social measure implementation schemes that meet the respective corresponding task requirements; controlling each sub-agent to demonstrate the preliminary social measure implementation scheme formulated by itself to obtain a demonstration result, the demonstration including specifying the decision basis, reasoning process, expected effect and potential risks of the preliminary social measure implementation scheme; for each sub-agent, controlling the sub-agent to refute the preliminary social measure implementation schemes formulated by other sub-agents to obtain a refutation result, the refutation including vulnerability detection and refutation score of the preliminary social measure implementation scheme; controlling each sub-agent to adjust the preliminary social measure implementation scheme according to the demonstration result and the refutation result of the preliminary social measure implementation scheme formulated by itself to obtain a revised implementation scheme formulated by each sub-agent; repeating the steps of demonstration, refutation and revision until the demonstration result and / or refutation result of the revised implementation scheme meet a preset condition, then determining the final social measure implementation scheme according to the revised implementation scheme currently obtained by each sub-agent.

[0104] The debate module aims to simulate the complex collision of thoughts in the brain during the human decision-making process, and through the four stages of proposal, demonstration, refutation and reflection revision, to promote deep communication and mutual error correction among sub-agents, thereby enhancing the scientificity, rationality and explainability of the decision.

[0105] In the proposal phase, each sub-agent independently proposes a preliminary social measure implementation plan based on its unique perspective and decision-making approach. This process serves as the starting point for decision-making, requiring sub-agents to fully utilize their unique characteristics to form innovative and feasible preliminary ideas.

[0106] In the argumentation phase, sub-agents need to elaborate and argue for their proposed social measure implementation plans, including clearly stating the decision basis, reasoning process, expected effects, and potential risks. Through this process, not only does it enhance the credibility and explainability of the plan, but it also provides clear targets for subsequent refutation and evaluation.

[0107] The refutation phase is the core part of the debate mechanism. In the refutation phase, sub-agents need to question and refute other agents' plans, pointing out their shortcomings or potential problems. This process simulates the intense confrontation and collision between different viewpoints in the human decision-making process, helping to discover flaws and shortcomings in the plan and promoting continuous improvement and optimization of the plan. The comprehensive evaluation phase is a key step in reflecting and modifying the plan based on the debate situation.

[0108] In the reflection and modification phase, sub-agents adjust and improve their own plans based on feedback, integrating the insights of different agents to form more mature and feasible social measure implementation plans.

[0109] Repeat the argumentation, refutation, and reflection and modification processes until the argumentation results and / or refutation results of the modified social measure implementation plans meet the preset conditions, such as reaching a consensus among sub-agents or achieving a certain balance in the modified social measure implementation plans, to obtain the final social measure implementation plan.

[0110] Among them, control each sub-agent to argue for the preliminary social measure implementation plan they have developed, and obtain argumentation results, including: controlling the sub-agent to analyze the preliminary social measure implementation plan based on the decision basis and reasoning process recorded when generating the preliminary social measure implementation plan; controlling the sub-agent to score the preliminary social measure implementation plan, and determining the expected effect and potential risks of the preliminary social measure implementation plan based on the score of the preliminary social measure implementation plan.

[0111] For example, the shortest path implementation scheme made by the sub-agent 210 can also generate a decision basis and reasoning process, and the decision model 211 can analyze the shortest path implementation scheme, the decision basis and the reasoning process, and then score them according to the analysis results. For example, if the score of the shortest path implementation scheme is 7, and the full score of the implementation scheme is 10, then the potential risk degree of the shortest path implementation scheme can be determined as 3. Similarly, the decision model 221 of the sub-agent 220 can analyze the shortest red light duration implementation scheme made by the sub-agent 220 to obtain the argumentation result of the scheme. The decision model 231 of the sub-agent 230 can analyze the minimum congestion probability implementation scheme made by the sub-agent 230 to obtain the argumentation result of the scheme.

[0112] For each sub-agent, the control sub-agent analyzes the preliminary implementation scheme made by the other sub-agent, and determines the vulnerabilities and the refutation score of the preliminary implementation scheme made by the other sub-agent according to the analysis results.

[0113] For example, the decision model 221 of the sub-agent 210 analyzes the shortest red light duration implementation scheme made by the sub-agent 220 and the minimum congestion probability implementation scheme made by the sub-agent 230 from the perspective of planning the shortest path, and obtains the vulnerabilities and the scheme score of the shortest red light duration implementation scheme and the minimum congestion probability implementation scheme detected by the sub-agent 210. Similarly, the sub-agent 220 can obtain the vulnerabilities and the scheme score of the shortest path implementation scheme and the minimum congestion probability implementation scheme detected by the sub-agent 220, and the sub-agent 230 can obtain the vulnerabilities and the scheme score of the shortest path implementation scheme and the shortest red light duration implementation scheme detected by the sub-agent 230.

[0114] Then, the sub-agent 210 corrects the preliminary shortest path implementation scheme according to the vulnerabilities and the scheme score of the shortest path implementation scheme detected by the sub-agent 220 and the sub-agent 230, and obtains the corrected shortest path implementation scheme. The sub-agent 220 corrects the preliminary shortest red light duration implementation scheme according to the vulnerabilities and the scheme score of the shortest red light duration implementation scheme detected by the sub-agent 210 and the sub-agent 230, and obtains the corrected shortest red light duration implementation scheme. The sub-agent 230 corrects the preliminary minimum congestion probability implementation scheme according to the vulnerabilities and the scheme score of the minimum congestion probability implementation scheme detected by the sub-agent 210 and the sub-agent 220, and obtains the corrected minimum congestion probability implementation scheme.

[0115] If the modified shortest path implementation scheme, the modified shortest red light duration implementation scheme and the modified minimum congestion probability implementation scheme meet the preset condition, a final implementation scheme is determined according to the three modified implementation schemes. For example, the average score of the shortest path implementation scheme is calculated according to the scheme score of the modified shortest path implementation scheme of the sub-agent 210, the sub-agent 220 and the sub-agent 230, the average score of the shortest red light duration implementation scheme is calculated according to the scheme score of the modified shortest red light duration implementation scheme of the sub-agent 210, the sub-agent 220 and the sub-agent 230, the average score of the minimum congestion probability implementation scheme is calculated according to the scheme score of the modified minimum congestion probability implementation scheme of the sub-agent 210, the sub-agent 220 and the sub-agent 230, if the difference between the average scores of the three modified implementation schemes is less than a preset value, the modified implementation scheme with the highest average score is taken as the final implementation scheme. Or when the average scores of the three modified implementation schemes no longer change, the modified implementation scheme with the highest average score is taken as the final implementation scheme.

[0116] The embodiment of the present disclosure sets up a self-argumentation and debate mechanism among multiple sub-agents, requires each sub-agent to fully demonstrate its point of view and reason, and accepts the interrogation and discussion of other sub-agents. This process not only increases the depth of communication within the system, but also enables external observers to clearly see the whole process of decision-making, thereby improving the transparency and explainability of the decision-making. At the same time, multiple sub-agents can examine and supplement each other, forming a self-correcting and perfecting cycle. In the debate process, sub-agents can discover and correct each other's biases and errors, thereby improving the overall fault tolerance. This mechanism significantly improves the accuracy and reliability of system decision-making, making the decision-making result more robust and credible.

[0117] By using the method provided by the embodiment of the present disclosure, multiple sub-agents can be integrated. Since each sub-agent is constructed by a knowledge base in different knowledge fields and a decision-making model corresponding to task requirements, each sub-agent can represent a different perspective. Integrating multiple sub-agents for decision-making can form multi-field and multi-dimensional insights, and then the social measure implementation schemes of multiple sub-agents are fused through the debate mechanism of multiple sub-agents, so that the agent can produce more comprehensive and diversified decision-making results, effectively dealing with complex and variable decision-making problems.

[0118] In some embodiments, the method further comprises: receiving feedback information of the user on the final social measure implementation scheme; and adjusting the sub-agents according to the feedback information, the adjustment content at least including one of the number of sub-agents, the type of sub-agents, shared knowledge, independent knowledge, the type of decision-making model, the argumentation manner and the refutation manner.

[0119] The embodiments of the present disclosure can be implemented by a feedback module, which serves as a bridge connecting humans and intelligent agents, and is responsible for fine-tuning the core components of the intelligent agent, i.e., the sub-agent generation module and the debate module, according to real-time or cumulative human feedback information.

[0120] Specifically, the feedback mechanism first captures and analyzes direct evaluations, preference indications from users or experts, and indirect feedback from task execution results. By deeply analyzing human feedback, the feedback module can identify possible deficiencies in the intelligent agent structure design, and then guide the sub-agent generation module and the debate module to adaptively adjust. This includes but is not limited to the number of sub-agents, the type of sub-agents, shared knowledge and independent knowledge, the type of large models, argumentation logic, and refutation strength, etc.

[0121] This process not only improves the adaptability of decision-making strategies to specific task environments, but also promotes the generalization ability of intelligent agents in unknown or dynamic environments. Through continuous iteration and optimization, intelligent agents can gradually learn to follow human expectations and norms, making more reasonable and efficient decisions.

[0122] In addition, intelligent agents have great application potential in the decision-making of measures implementation schemes, especially in complex environments, they can help decision-makers simulate human cognition and behavior, thereby improving the scientificity and accuracy of measures implementation scheme formulation and evaluation.

[0123] For example, intelligent agents can be used to evaluate measures implementation schemes. Intelligent agents can simulate the behavior of social members with different backgrounds and individual characteristics. Each intelligent agent can contain unique attributes such as age, occupation, income level, social status, cultural background, education level, etc. Through these attributes, intelligent agents can exhibit diverse behavior patterns. In the impact simulation of measures implementation schemes, different individuals will have different reactions to social measures, and human-like intelligent agents can generate diverse behavioral responses based on these differences, thereby more comprehensively reflecting the reactions of various groups of people in society to social measures. In addition, human-like intelligent agents can also simulate the interaction between individuals after the implementation of social measures. Since the actual effect of social measures depends not only on the independent behavior of individuals, but also on the interaction of relationships in social networks, it is necessary to simulate the interaction between individuals to more accurately evaluate the impact of social measures.

[0124] For example, in evaluating a social measure that encourages environmental protection, intelligent agents can simulate how changes in attitudes towards environmental protection in a person's social circle after receiving the social measure affect the decision-making of others, thereby predicting the promotion of environmental protection behavior in the entire society by the social measure.

[0125] For example, in emergency management, agents can be used to simulate the emergency response and social recovery process of disasters, helping some place managers to make emergency plans and safety measures more effectively. Simulate the behavior patterns of users in crisis to assist in designing more flexible response measures.

[0126] Based on the social simulation application of the humanoid agent, the implementer of the embodiment can more comprehensively understand the complex social dynamics, predict the effect after the implementation of the social measure by simulating human behavior, foresee potential problems and optimize adjustment, thereby improving the scientificity, rationality and execution effect of the social measure, and reducing the risk of implementation failure. The following describes the multi-agent based decision device provided by the present application, and the multi-agent based decision device described below can be correspondingly referred to the multi-agent based decision method described above.

[0127] As shown in Figure 3 The multi-agent based decision device 300 provided by the present application includes the following modules:

[0128] The data acquisition module 301 is configured to acquire data of a plurality of data sources and construct a plurality of knowledge bases of different knowledge fields.

[0129] The sub-agent generation module 302 is configured to construct a plurality of sub-agents by using corresponding decision models and knowledge bases of corresponding knowledge fields according to different task requirements contained in the target social measure.

[0130] The debate module 303 is configured to control the plurality of sub-agents to formulate respective preliminary social measure implementation schemes for the target social measure, and to debate the plurality of preliminary social measure implementation schemes to obtain a final social measure implementation scheme.

[0131] According to the multi-agent based decision device provided by the present application, the data acquisition module 301 is specifically configured to:

[0132] Connect a plurality of different types of data sources;

[0133] According to the data stored in the plurality of different types of data sources, divide a plurality of knowledge fields;

[0134] For each knowledge field, construct a knowledge base by using the data of the knowledge field.

[0135] According to the multi-agent based decision device provided by the present application, the sub-agent generation module 302 is specifically configured to:

[0136] Extract shared knowledge from the plurality of knowledge bases according to the target social measure;

[0137] For each task requirement, determine a decision model corresponding to the task requirement;

[0138] For each task requirement, the perspective features, domain knowledge and thinking modes required to meet the task requirement are extracted from the knowledge base of the corresponding knowledge field, and independent knowledge corresponding to the task requirement is generated;

[0139] Based on shared knowledge, a sub-agent is constructed for each task requirement using the decision model corresponding to the task requirement and the corresponding independent knowledge.

[0140] According to the multi-agent based decision device provided by the application, the argumentation module 303 is specifically used for:

[0141] Controlling a plurality of sub-agents to develop a preliminary social measure implementation scheme that meets the respective corresponding task requirements for the target social measure;

[0142] Controlling each sub-agent to demonstrate the preliminary social measure implementation scheme developed by itself to obtain a demonstration result, the demonstration including the decision basis, reasoning process, expected effect and potential risks of the preliminary social measure implementation scheme;

[0143] For each sub-agent, the sub-agent is controlled to refute the preliminary social measure implementation scheme developed by other sub-agents to obtain a refutation result, the refutation including vulnerability detection and refutation scoring of the preliminary social measure implementation scheme;

[0144] Controlling each sub-agent to adjust the preliminary social measure implementation scheme according to the demonstration result and the refutation result of the preliminary social measure implementation scheme developed by itself to obtain a revised implementation scheme developed by each sub-agent;

[0145] Repeat the steps of the above demonstration, refutation and revision until the demonstration result and / or refutation result of the revised implementation scheme meet the preset conditions, then determine the final social measure implementation scheme according to the revised implementation scheme currently obtained by each sub-agent.

[0146] According to the multi-agent based decision device provided by the application, when the argumentation module 303 controls each sub-agent to demonstrate the preliminary social measure implementation scheme developed by itself to obtain a demonstration result, it is specifically used for:

[0147] Controlling the sub-agent to analyze the preliminary social measure implementation scheme according to the decision basis and reasoning process recorded when the preliminary social measure implementation scheme is generated;

[0148] Controlling the sub-agent to score the preliminary social measure implementation scheme, and determining the expected effect and potential risks of the preliminary social measure implementation scheme according to the score of the preliminary social measure implementation scheme.

[0149] According to a multi-agent decision-making device provided by the present invention, the debate module 303, when controlling each sub-agent to refute the preliminary social measures implementation plans formulated by other sub-agents and obtaining the refutation results, is specifically used for:

[0150] For each sub-agent, the control sub-agent analyzes the preliminary social measures implementation plans formulated by other sub-agents, and determines the vulnerabilities and rebuttal scores of the preliminary social measures implementation plans formulated by other sub-agents based on the analysis results.

[0151] According to the present invention, a multi-agent-based decision-making device is provided, the device further comprising a feedback module 304, the feedback module 304 being specifically used for:

[0152] Receive user feedback on the final implementation plan of social measures;

[0153] Adjustments are made to the sub-agents based on feedback information. The adjustments include at least one of the following: number of sub-agents, type of sub-agents, shared knowledge, independent knowledge, decision model type, argumentation method, and refutation method.

[0154] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other via the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a multi-agent-based decision-making method, which includes:

[0155] Collect data from multiple data sources and build multiple knowledge bases for different knowledge domains;

[0156] To address the different task requirements of the target social measures, multiple sub-agents are constructed using corresponding decision-making models and knowledge bases in relevant knowledge domains.

[0157] Multiple sub-agents are controlled to formulate their own preliminary social measure implementation plans for the target social measure, and debate the multiple preliminary social measure implementation plans to obtain the final social measure implementation plan.

[0158] In addition, the logic instructions in the memory 1130 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0159] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the multi-agent based decision-making method provided by the above-mentioned methods, which comprises:

[0160] Collecting data of a plurality of data sources to construct a plurality of knowledge bases of different knowledge fields;

[0161] For different task requirements contained in the target social measure, a plurality of sub-agents are constructed by using corresponding decision-making models and knowledge bases of corresponding knowledge fields;

[0162] Controlling the plurality of sub-agents to formulate respective preliminary social measure implementation schemes for the target social measure, and debating the plurality of preliminary social measure implementation schemes to obtain a final social measure implementation scheme.

[0163] In another aspect, the present application also provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the multi-agent based decision-making method provided by the above-mentioned methods, which comprises:

[0164] Collecting data of a plurality of data sources to construct a plurality of knowledge bases of different knowledge fields;

[0165] For different task requirements contained in the target social measure, a plurality of sub-agents are constructed by using corresponding decision-making models and knowledge bases of corresponding knowledge fields;

[0166] Controlling the plurality of sub-agents to formulate respective preliminary social measure implementation schemes for the target social measure, and debating the plurality of preliminary social measure implementation schemes to obtain a final social measure implementation scheme.

[0167] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0168] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0169] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A decision-making method based on multi-agent agents, characterized in that, The method includes: Collect driving route planning data from multiple data sources and construct multiple driving route planning knowledge bases in different knowledge domains; To address the different task requirements of the target driving route planning task, multiple driving route planning sub-agents are constructed by adopting corresponding driving route planning models and driving route planning knowledge bases of corresponding knowledge domains. The system controls the multiple driving route planning sub-agents to formulate their respective preliminary driving route planning schemes for the target driving route planning task, and debates the multiple preliminary driving route planning schemes to obtain the final driving route. To address the different task requirements of the target driving route planning task, multiple driving route planning sub-agents are constructed using corresponding driving route planning models and driving route planning knowledge bases of relevant knowledge domains, including: Based on the target driving route planning task, shared knowledge is extracted from the multiple knowledge bases; For each task requirement, determine the corresponding driving route planning model; For each task requirement, extract the perspective features, domain knowledge, and thinking methods required to meet the task requirement from the knowledge base of the corresponding knowledge domain, and generate independent knowledge corresponding to the task requirement. Based on the shared knowledge, a driving route planning sub-agent is constructed for each task requirement using the driving route planning model corresponding to the task requirement and the corresponding independent knowledge; the shared knowledge is map information. The system controls the multiple driving route planning sub-agents to formulate their respective preliminary driving route planning schemes for the target driving route planning task, and debates the multiple preliminary driving route planning schemes to obtain the final driving route, including: Control the multiple driving route planning sub-agents to formulate preliminary driving route planning schemes that meet the corresponding task requirements for the target driving route planning task; Each driving route planning sub-agent is controlled to demonstrate its own preliminary driving route planning scheme and obtain the demonstration result. The demonstration includes clarifying the driving route planning basis, reasoning process, expected effect and potential risks of the preliminary driving route planning scheme. For each driving route planning sub-agent, the driving route planning sub-agent is controlled to refute the preliminary driving route planning scheme formulated by other driving route planning sub-agents, and the refutation results are obtained. The refutation includes the detection of loopholes in the preliminary driving route planning scheme and the refutation score. The preliminary driving route planning scheme includes the shortest route implementation scheme, the shortest red light duration implementation scheme, and the minimum congestion probability implementation scheme. The driving route planning sub-agent includes a route sub-agent, a red light duration sub-agent, and a congestion probability sub-agent; the route sub-agent is used to plan the shortest route, obtaining the shortest route implementation scheme; the red light duration sub-agent is used to plan the route with the fewest red light durations, obtaining the shortest red light duration implementation scheme; the congestion probability sub-agent is used to plan the route with the minimum congestion probability, obtaining the minimum congestion probability implementation scheme. The vulnerability detection and rebuttal scoring includes the vulnerability and scheme scores detected by the route sub-agent for the shortest red light duration implementation scheme and the minimum congestion probability implementation scheme, the vulnerability and scheme scores detected by the red light duration sub-agent for the shortest route implementation scheme and the minimum congestion probability implementation scheme, and the vulnerability and scheme scores detected by the congestion probability sub-agent for the shortest route implementation scheme and the shortest red light duration implementation scheme. Each driving route planning sub-agent adjusts the initial driving route planning scheme based on the demonstration and rebuttal results of its own initial driving route planning scheme, thereby obtaining the revised implementation scheme formulated by each driving route planning sub-agent; Repeat the above steps of argumentation, refutation, and correction until the argumentation and / or refutation results of the corrected implementation scheme meet the preset conditions. Then, determine the final driving route based on the corrected implementation scheme currently obtained by each driving route planning sub-agent.

2. The decision-making method based on multi-agent systems according to claim 1, characterized in that, The process involves collecting driving route planning data from multiple data sources to construct multiple driving route planning knowledge bases across different knowledge domains, including: Connect to multiple data sources of different types; Based on the driving route planning data stored in the multiple different types of data sources, multiple knowledge domains are defined; For each knowledge domain, a driving route planning knowledge base is constructed using data from that knowledge domain.

3. The decision-making method based on multi-agent systems according to claim 1, characterized in that, The process involves controlling each driving route planning sub-agent to validate its respective preliminary driving route planning scheme and obtain validation results, including: The driving route planning sub-agent is controlled to analyze the preliminary driving route planning scheme based on the driving route planning basis and reasoning process recorded when the preliminary driving route planning scheme is generated; The driving route planning sub-agent is controlled to score the preliminary driving route planning scheme, and the expected effect and potential risks of the preliminary driving route planning scheme are determined based on the score.

4. The decision-making method based on multi-agent systems according to claim 1, characterized in that, For each driving route planning sub-agent, the system controls the driving route planning sub-agent to refute the preliminary driving route planning schemes formulated by other driving route planning sub-agents, and obtains the refutation results, including: For each driving route planning sub-agent, the system controls the driving route planning sub-agent to analyze the preliminary driving route planning schemes formulated by other driving route planning sub-agents, and determines the loopholes and rebuttal scores of the preliminary driving route planning schemes formulated by other driving route planning sub-agents based on the analysis results.

5. The decision-making method based on multi-agent systems according to claim 1, characterized in that, The method further includes: Receive user feedback on the final driving route; The driving route planning sub-agent is adjusted based on the feedback information. The adjustment includes at least one of the following: the number of driving route planning sub-agents, the type of driving route planning sub-agents, shared knowledge, independent knowledge, driving route planning model type, argumentation method, and rebuttal method.

6. A decision-making device based on multi-agent systems, characterized in that, include: The data acquisition module is used to collect driving route planning data from multiple data sources and build multiple driving route planning knowledge bases in different knowledge domains; The sub-agent generation module is used to construct multiple driving route planning sub-agents based on the different task requirements included in the target driving route planning task, using the corresponding driving route planning model and the driving route planning knowledge base of the corresponding knowledge domain. The debate module is used to control the multiple driving route planning sub-agents to formulate their own preliminary driving route planning schemes for the target driving route planning task, and to debate the multiple preliminary driving route planning schemes to obtain the final driving route. To address the different task requirements of the target driving route planning task, multiple driving route planning sub-agents are constructed using corresponding driving route planning models and driving route planning knowledge bases of relevant knowledge domains, including: Based on the target driving route planning task, shared knowledge is extracted from the multiple knowledge bases; For each task requirement, determine the corresponding driving route planning model; For each task requirement, extract the perspective features, domain knowledge, and thinking methods required to meet the task requirement from the knowledge base of the corresponding knowledge domain, and generate independent knowledge corresponding to the task requirement. Based on the shared knowledge, a driving route planning sub-agent is constructed for each task requirement using the driving route planning model corresponding to the task requirement and the corresponding independent knowledge; the shared knowledge is map information. The system controls the multiple driving route planning sub-agents to formulate their respective preliminary driving route planning schemes for the target driving route planning task, and debates the multiple preliminary driving route planning schemes to obtain the final driving route, including: Control the multiple driving route planning sub-agents to formulate preliminary driving route planning schemes that meet the corresponding task requirements for the target driving route planning task; Each driving route planning sub-agent is controlled to demonstrate its own preliminary driving route planning scheme and obtain the demonstration result. The demonstration includes clarifying the driving route planning basis, reasoning process, expected effect and potential risks of the preliminary driving route planning scheme. For each driving route planning sub-agent, the driving route planning sub-agent is controlled to refute the preliminary driving route planning scheme formulated by other driving route planning sub-agents, and the refutation results are obtained. The refutation includes the detection of loopholes in the preliminary driving route planning scheme and the refutation score. The preliminary driving route planning scheme includes the shortest route implementation scheme, the shortest red light duration implementation scheme, and the minimum congestion probability implementation scheme. The driving route planning sub-agent includes a route sub-agent, a red light duration sub-agent, and a congestion probability sub-agent; the route sub-agent is used to plan the shortest route, obtaining the shortest route implementation scheme; the red light duration sub-agent is used to plan the route with the fewest red light durations, obtaining the shortest red light duration implementation scheme; the congestion probability sub-agent is used to plan the route with the minimum congestion probability, obtaining the minimum congestion probability implementation scheme. The vulnerability detection and rebuttal scoring includes the vulnerability and scheme scores detected by the route sub-agent for the shortest red light duration implementation scheme and the minimum congestion probability implementation scheme, the vulnerability and scheme scores detected by the red light duration sub-agent for the shortest route implementation scheme and the minimum congestion probability implementation scheme, and the vulnerability and scheme scores detected by the congestion probability sub-agent for the shortest route implementation scheme and the shortest red light duration implementation scheme. Each driving route planning sub-agent adjusts the initial driving route planning scheme based on the demonstration and rebuttal results of its own initial driving route planning scheme, thereby obtaining the revised implementation scheme formulated by each driving route planning sub-agent; Repeat the above steps of argumentation, refutation, and correction until the argumentation and / or refutation results of the corrected implementation scheme meet the preset conditions. Then, determine the final driving route based on the corrected implementation scheme currently obtained by each driving route planning sub-agent.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multi-agent-based decision-making method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-agent-based decision-making method as described in any one of claims 1 to 5.

Images