一个基于Bayesian学习的协商模型

在Multi-Agent系统(MAS)中，每一个Agent都有不同的目标。通常只拥有对方的不完全信息。Agent需要具有解决在实现各自目标过程中所产生的各种矛盾的能力。协商是解决这些矛盾的一种有效途径。本文提出了一个基于Bayesian学习的协商模型NMBL：在每一轮协商中，Agent通过Bayesian学习获取协商对手的信息，更新对协商对手的信念，然后根据基于冲突点和不妥协度的协商策略提出下一轮的协商提议。NMBL把整个协商过程看成一个动态的交互过程，体现了Multi-Agent系统的动态特性，同时NMBL具有较强的学习能力。试验证明，该模型具有较好的协商性能。

A Negotiation Model Based on Bayesian Learning

In Multi-Agent systems where each Agent has a different goal, Agent must be able to solve conflicts aris- ing in the process of achieving its goal, with incomplete knowledge about other Agents. Negotiation is an effective ap- proach to solve these problems. This paper introduces a negotiation model based on Bayesian learning, called NMBL. Agent gets information of the negotiation opponents in every iteration by means of Bayesian learning, updates the pri- or knowledge of the negotiation opponents and then brings forward the offer of the next iteration according to negotia- tion strategies based on the conflicting point and un-compromising degree. NMBL regards the whole negotiation pro- cess as a dynamic interaction conduct, which reveals the dynamic characteristic of Multi-Agent systems' NMBL also has a relatively strong learning ability. The experiments show that this model has good negotiation performance.