Research On Cooperative Evolutionary Model In Distributed Multi-agent Systems

A distributed multi-agent system is a computing system with various autonomous agents to achieve a common goal.Such systems are widely used in file sharing,online social networks,wireless sensing,etc.As agents in these systems are selfish and strategic,they tend to maximize their own profits:exploit other agents;profits and not to share resources.If noncooperation(not to share)spreads out,the effective of these systems greatly decreases.Incentivizing agents to promote performance is a heated research field in multi-agent systems which draws lots of'atteintions.In this dissertation,we deeply study existing cooperation models and find some drawbacks:existing studies do not consider ubiquitous agents' altruism;existing studies focus on promoting cooperation but neglect considering cost of implementing mechanisms;interactions among strategic agents are not properly depicted in reputation-based mechanisms.In this dissertation,we deeply study incentive mechanisms based on evolutionary game and propose solutions from three aspects:(1)depict agents and interactions among them;(2)model instinct of agents and(3)reduce system costs.The contributions of this dissertation are as followings:(1)To solve the problem of neglecting altruism,a cooperation evolutionary model with altruistic agents is proposed.Altruism of each agent focuses on fitness calculation.Neighbors'payoffs are introduced into fitness calculation:fitness of each agent not only relates to its own payoff but also agents' around it.The importance of other agents' payoffs is weighted by a fitness parameter.Through extensive theoretical analysis and simulations,we prove that altruism of agent can speed up evolution process.Simulations in structured networks and real topology show that with altruistic agents,evolutionary results can be changed.(2)A dynamic reward based incentive mechanism is proposed.In this model,the reward value is a function to the system environment:reward more when cooperation at disadvantage and reward less when cooperation at advantage.We analyze the evolutionary stable strategy under different scenarios by replicator dynamics,and quantitatively prove that dynamic reward mechanism can reduce system costs.Simulation results show that our proposed mechanism can promote cooperation and reduce system cost.(3)An EigcinTrust-based cooperation evolution model is proposed by introducing evolutionary game.In this model,we fully consider agent types and interactions among them which can better model agentsr behaviors in distributed multi-agent systems.By introducing evolutionary game theory to depict rational agents and interactions,we propose a more realistic evolutionary model.Simulation and analysis results show that our proposed model can effectively promote cooperation,reduce free-riding behaviors and reduce inauthentic file downloading rate.