Research On Cooperation And Approximate Reasoning Mechanism Among Multi-Agents Under The Environment Of Abdidss

It often happens to large -scale complex non-structure decision problems in the process of modern enterprise's production, business and management. To solve these problems, many kinds of professional knowledge are required and many qualitative problems are dealt with. It is required objectively that decision support systems should satisfy distribution of decision organization, complexity of decision problems, cooperation of decision systems, uncertainty of decision knowledge and reasoning, adaptability of decision environments and parallelism of decision processes while complex problems are decided on.But traditional central-type decision support systems can not satisfy theses requirements. ABDIDSS integrates agent technology with distributed intelligence decision support systems and is applied to higher decision levels and more complex decision environments. So it is a feasible model which satisfies complex problem decision. In order to show mechanism to improve ability and knowledge to solve complex problems under the environment of ABDIDSS, research on cooperation and reasoning among multi-agents has important theory meanings.In order to solve dynamic, complicated and uncertain problems, multi-agents based distributed intelligence decision support systems, description and decomposition of complicated decision tasks, and cooperation and reasoning among multi-agents are elaborated in this paper. The details are given as follows. 1. Agent and ABDIDSS modelsIn allusion to the current problems of research on agents theory models, payoff and probability factors are introduced on the basis of mental state models of classical BDI of agent and joint BDI of multi-agents in this paper, mental state models of individual agents and joint metal state models of group agents based on payoff and probability factors are built up. Introducing payoff and probability factors satisfies requirements of uncertainty and self-benefit of agents to the environments. In the meantime, agent structure models corresponding to mental state models and multi-agents - based DIDSS model are built up.2. Description and decomposition of decision tasksFormal description methods of complex decision task based on BN(Bayesian network) are put forward; it is shown that decomposition problem of decision tasks is equivalent to decomposition problem of Bayesian network; Decomposition properties of decision agent tasks, and decomposition properties and decomposition methods of corresponding Bayesian network are put forward.Complexity of BN reasoning systems relies on probability table-scale corresponding to SBN(sub- Bayesian network) decomposed, namely attribute variable-scale of probability table and value states. Therefore an optimal decomposition method exists, which minimizes reasoningcomplexity in the SBN. Optimal decomposition properties and methods are put forward in the paper.In the meantime, decomposition problem of BN is optimized in genetic algorithm, general principle, algorithm designing and realizing of Bayesian network decomposition based on genetic algorithm are put forward.3. Basic mechanisms of cooperation among multi-agentsAccording to knowledge domain of decision tasks involved and knowledge property of individual agent, cooperation manners to solve decision tasks for multi-agents include tasks-share cooperation manner and results-share cooperation manner. Tasks assumption and assignation are managed in contract net to tasks-share cooperation manner; decision information and decision results are shared among multi-agents in partial global planning to results-share cooperation manner; cooperation solving processes of tasks are put forward under the environment of ABDIDSS.4. Game theory- based cooperation mechanisms among multi-agentsFirst, Multi-agents cooperation model based on game theory is put forward, cooperation scene based on game theory is described in multi-agents influence diagrams, therefore cooperation among multi-agents is equivalent to solving game equilibrium solutions described in multi-agents influence diagrams. Princip...