The Research On Multi-Agent System Socialized Behavior Based On Evolutionary Game

With the mature of deep learning technology,artificial intelligence is gradually from cutting-edge technology to daily life,therefore,a technology which for the further analysis and development of artificial intelligence—group intelligence,has begun to be studied by a large number of scholars,and social behavior of swarm intelligence,such as cooperation behavior and malicious behavior,is a difficulty and hot spot of current research.Evolutionary game theory provides a powerful theoretical model for this problem.In evolutionary game,we can construct a generic ecological game model.In this model,we can easily observe the socialization behaviors of each agent in the intelligent group,and timely comprehend the impact of corresponding behavioral changes on the whole system.In such a macroscopic and microscopic comprehensive research,how to find out the strategy equilibrium of the multi-agent system,and how to adjust the influencing factors and design a reasonable mechanism to guide the agent behavior in the game,so as to achieve the self-adaptive multi-agent system,which is the key point of this paper.In order to solve the above problems,this paper based on game theory,mainly from two aspects to deeply research cooperative behavior and malicious behavior,respectively.On the one hand,in the prisoner’s dilemma,by a learning range enhancement mechanism to research cooperation in multi-agent system.On the other hand,based on the background of network security,combined with detection mechanism and belief algorithm,several new models are proposed to study the game in network offense and defense.The main research work of this paper is as follows:First,we discussed the basic framework of game theory,and the operating principle of core modules in ecological game model,deeply analyzed the population dynamics of evolutionary game theory:replicator dynamics and evolutionary stable strategy.At the same time,we also introduced the common dilemma game model and the game evolution in different spatial network structures.In the research of socialized cooperative behavior,this paper proposes a learning range enhancement mechanism based on the framework of evolutionary game theory,and conducts a series of experimental researches on the prisoner game model,It mainly explores the following three aspects:the influence of learning range on the system cooperation rate,the relationship between learning range and size of game network,and the relationship between learning range and the spatial network structure.The result found it’s wrong about the intuitive understanding of the learning ability:the higher cooperation level,the higher promotion to the system.but it has strong dependence on the system environment,meanwhile,there is a threshold in the influence of the system cooperation rate.In the research of socialized malicious behavior,this paper migrates the game background to the field of network security,analysis and studies the characteristics in network security game:identity hiding,information asymmetry,harmfulness and so on.meanwhile,based on the bayesian belief model and abstract the detection mechanism function,this paper proposes a relative historical profit model with a certain innovation significance,which makes up the weaknesses of bayesian model and finally combines the advantages of the two models,an innovative model of improved Bayesian is proposed,from theoretical deduction to experimental verification,multiple aspects to explore the comprehensive effect of new model,in order to construct a flexible security network.