Research On The Emergence Of Cooperation Norms In Multi-agent System

As a method for analyzing complex systems,multi-agent system is of great significance in managing and coordinating communication and cooperation among agents.However,due to the selfishness and independent decision-making ability of the participants,how to quickly and effectively emerge the cooperation norms in complex scenarios has become a major challenge in the process of the system development.Using computer technology to reproduce the agents in the real system and proposing a highly efficient and universal theoretical model to allow agents to continuously optimize their own behavior by using local information in the interaction process to emerge a cooperation specification.In addition,to learn,understand and even control the multi-agent system by analyzing the simulation results has gradually become the core issue in the field of artificial intelligence and artificial society.Based on the background of multi-agent system,this paper abstractly describes the problem scenario in its development process as general dilemma models,and proposes corresponding mechanism for norm emergence.The main contributions are as follows:(1)The existing work on the study of strategy learning rules is too singular,mainly relying on simple information such as payoff and reputation to design interactive protocols,failing to take full account of the agent's cognitive ability and social attributes,which makes the proposed theoretical model difficult to accurately describe the complex behavior characteristics of agents in real life.To solve this problem,the paper proposes a learning rule with punishment mechanism based on individual influence,in which the decision-making behavior of the agent is mainly determined by individual benefit and individual influence.Individual influence is mainly controlled by the adjustment of sensitivity and the duration of the current strategy and adjusts according to the learning behavior of the individual.The experimental results show that the learning method greatly improves the cooperation level of the system.And through micro-analysis,it is found that the difference of individual influence between agents is the dominating reason for the improvement of cooperation.Finally,this paper verifies experimental results in complex networks such as small-world networks and scale-free networks,and finds that the learning rule can solve the dilemma in different complex network scenarios.(2)Reinforcement learning has been proven to fail to promote the emergence of cooperation norms in the traditional prisoner's dilemma model.Recent research on strategic exploration based on historical information has been proven to achieve the desired cooperation effects in the multi-strategy game model.In response to this achievement,the paper proposes the reinforcement learning method to promote the emergence of cooperation norms in the scenario of multi-strategy game with exploitation strategy.The experimental results show that the reinforcement learning method greatly improves the level of cooperation in the system and its ability to enhance cooperation is stronger than other learning methods that have been proposed.At the same time,the paper analyzes the reasons why reinforcement learning can promote cooperation and reveals the role of different kinds of agents in the process of system development.