Research Of Probabilistic Framework For Multi-Agent System Model

Bayesian technology and Bayesian networks are used to process uncertainty problems of Artificial Intelligence domain. Because of the incorporation between Bayesian technology and Agent technology, a very promising new domain which possesses a distant spectacle is formed.The representation of Multi —Agent system is the key of a series of operations on Multi-Agent system, such as reasoning,learning and coorperation. Based on the introduction and discussion of the theory of Agent technology and Bayesian Networks,the thesis states the representation framwork of multi-agents model on large knowledge domain—Multiply Sectioned Bayesian Networks Model. The thesis studied the realization of Linked Junction Forest which leads to facility of reasoning and introduces the representation of systematic belief and the specific operation used to realise reasoning. The main works are as follows:(l)The thesis proposed an 1(1/2)-slice junction tree algorithm as well as analyzed dynamic bayesian network stucture. The algorithm has better characteristic property, such as the simple realization, without limiting the sequence of expurgation, only one extension; and by experiment, the thesis has proved that the algorithm are superior to ordinary extended junction tree algorithm and 2-slice junction tree algorithm in respect of run-time and space.(2) Owing to the locality of large knowledge domain, the dissertation has represented multi-agent system on large knowledge domain by means of multiply sectioned bayesian network, and then combined the idea of junction tree, the dissertation has studied the realization of linked junction forest and the effective improvement of UpdateBelief.