Research And Verification Of Trust Measurement Of Multi-agent Based On Reinforcement Learning

In a smart city,an agent refers to an entity with the basic characteristics of autonomy,sociality,responsiveness and proactiveness,and is an important part of a smart city.The Multi-Agent System(MAS)is composed of a series of interacting agents,and the internal agents complete complex tasks through mutual cooperation and competition.However,due to limited resources and services,there will be vicious competition between agents.If you only consider your own tasks without considering the interests of other agents,the final choice may lead to the loss of the overall interests of the smart city.Existing algorithms have little effect on solving the problems of competition and cooperation in a multi-agent environment.How to promote the harmonious development of society on the premise of maximizing the overall benefits is a scientific problem that needs to be broken through.To solve this problem,we introduce trust:if an agent can establish trusted relationships with other agents and produce cooperative behavior,it will improve the overall benefit.In order to effectively solve the vicious competition in smart cities and make agents adopt a cooperative strategy,this paper proposes a new trust-based logical framework and trust measurement model.The main work is as follows:First,the Trust Computation Logic,TCL),which defines trustworthiness in a multi-agent environment,gives the definition of trustworthy relationships,uses trustworthy computing logic to give some reasonable assumptions,and gives strict proofs for these assumptions.Secondly,for multi-agent systems,this paper proposes an algorithm for constructing trusted behaviors,forms a trusted behavior library,and uses the A3C algorithm to use DeepMind’s Sequential Social Dilemma(SSD)multi-agent game theory environment.Experiments verify the effectiveness of the trusted behavior construction algorithm.Through the above work,this paper solves the current vicious competition among multi-agents,and its main innovations and advantages are as follows:First,the innovation of trusted formal expression in multiagent systems,and the effective expression of credibility is the credibility of agents.The basic problem in measurement and behavior modeling,this paper innovatively builds credibility in multi-agent systems based on CTL logic language and interpretation system,and introduces TCL logic,which can compare the predictions of other agents with actual observations,inferred to be credible.We innovatively use a representation of a public matrix,which enables each agent to judge whether other agents are trustworthy,so that the interaction target can be quickly determined when exchanging information.This paper regards trustworthiness as the criterion for agents to select interactive objects,and gives a breakthrough in the precise definition of trustworthiness in a multi-agent environment.Corresponding reasonable assumptions and rigorous proofs are required to demonstrate the validity of the formal expression.The second is the innovation of the trust measurement model in the multi-agent system.Based on the prediction and observation of other agents,this paper explores new ideas and methods of trust measurement supported by machine learning,and successfully introduces and defines the intelligent measurement model.A new method for the behavior of individuals as a basis for judging their credibility.This paper is verified in the game environment,and the algorithm works well according to prior knowledge and experience judgment.