【作者】 姚宏亮；

【导师】 王浩；

【作者基本信息】 合肥工业大学 ， 计算机软件与理论， 2003， 硕士

【摘要】 贝叶斯技术和贝叶斯网络是人工智能中不确定问题处理的一种工具。贝叶斯技术和Agent技术的融合形成一个具有广阔前景的新兴研究领域。本文对贝叶斯理论和贝叶斯模型进行介绍和讨论。基于数据驱动的贝叶斯网络结构学习是当前的一个研究热点。本文提出了一种基于遗传算法和MDL原则的贝叶斯网络结构优化算法(GAMDL)，优点在于能有效地避免过分拟合，能解决NP问题，且学习优化的过程具有自适应性，算法具有较快的学习速度且可得到近似全局最优的结果。本文在动态贝叶斯网络的基础上引入了用于处理决策问题的影响图模型，另外还在动态贝叶斯网络和Agent技术介绍的基础上，给出一种基于分解和合并的耦合隐含马尔可夫模型，解决了三个以上的Agent组成的多Agent系统在计算上不可处理的难题。更多还原

【Abstract】 Bayesian technology and Bayesian networks are used to process uncertain problems of Artificial Intelligence domain. Bayesian technology is incorporated in Agent technology, so a very promising new domain is formed.Bayesian theory and Bayesian models are system introduced in the thesis. It is a hotspot that Learning Bayesian network structures based on data driven. The GAMDL Algorithms is developed which is based on Genetic Algorithms and Minimum Description Length principle. The advantages of this learning algorithm are that it can efficiently void over-fitting, solve NP-hard problem, and learning process has adaptability. The algorithm has efficient learning performance.Dynamic Bayesian Networks (DBNs) are a powerful methodology for representing and computing with uncertain problem of stochastic processes. Influence diagrams are a useful tool that it is used to deal with decision problem. Actions of three or more Agents are modeled by combining Agent technology with Bayesian theory. An approach of decomposition and incorporation is developed to resolve the problem which is intractable for exact calculations about multi-agent system based on Dynamic Bayesian networks. The approach improves ability of model representing.更多还原