Global Coordination Of Multi-Agents System Based On Belief Model

With the evolution of AI, especially for the application of distributed AI in large scale MAS, the macroscopical colony phenomenon comes out more and more offen.The traditional study of agent structure and theory can not fit the situation any more. So the research of Artificial Social System emerged. This method is used to control the balance between the complexities of central control and the conflicts of distributed control. It has already been through top-down off-line design to bottom-up on-line design to adjust to the dynamic MAS. Now the emergence of colony intelligence in on-line designed MAS has been the hotspot in human culture and technologies.This thesis researches the emergence of social convention in multi-agent system with bottom-up on-line design, builds the formal model of multi-agent system and analyzed the point of the system that will influence the emergence. The main works of the thesis are summarized as follows:1. According to the evolution of the emergence phenomenon in MAS, we choose proper mathematics model to formalize the artificial system, and decide to use game theory and belief degree as the emergence mechanisms.2. In previous research, agents usually change their strategies according to the feedback of their external environment such as, resources an agent hold, geography locations and social position. This thesis tries to use the belief among agents as the strategy update function which we call HCB to change their strategies. And we use game theory matrix as the payoff of belief feedback. Then,we build the MAS model based on game theory and belief degree. In the experiment part we verified the validation of emergence with HCB strategy and compared the efficiency of HCB and HCR.3. According to the summarizations of the influence factors of emergence, we've picked some key points and made a detail research, which including: a) the key K & P properties to the efficiency of emergence. b) The validation of emergence on B-A scale free network and the efficiency compare of the emergence on W-S model small world and B-A scale free network. c) We try to combine the belief and position factors together ith different emphasis as the parameter of strategy update function and verify its validation and efficiency on the emergence.4. After the emergence has come out, the system comes to a balance state. We just change 20% Agents'strategies to observe the state of the system.