基于贝叶斯优化构建DBN结构优化算法

针对动态贝叶斯网络(DBN)结构学习问题,提出了一种基于贝叶斯优化(BOA)的DBN结构寻优算法。首先,从传统进化优化机制的基本理论和基本操作入手,刻划了基于概率模型进化算法的基本思想。其次,通过描述基于概率模型进化算法的构图基础,引出了DBN结构学习机制,即基于BOA的DBN结构寻优算法。BOA算法的关键是根据优良解集学习得到DBN,以及根据DBN推理生成新个体,前者更为重要,依据基于贪婪机理的遗传算法解决动态网络学习,再应用DBN前向模拟完成后一步。仿真结果表明了该算法的可行性。

Constructing DBN structure based on BOA

An optimal algorithm for dynamic Bayesian networks(DBN) based on Bayesian optimal algorithm(BOA) is developed for learning and constructing DBN structure.Firstly,some basic theories and concepts of the probability model evolutionary algorithm are introduced.Secondly,the basic mode for constructing DBN diagram are described and the mechanism of DBN structure learning based on BOA is clarified.The BOA includes two parts of main technique,one is to gain the structure and parameter of DBN in term of good solutions,the another is to produce new group according to DBN.The learning of DBN is done by genetic algorithm based on the greed mechanism.The inference of DBN is done by a forward-simulation algorithm. Matlab simulation result demonstrates the proposed algorithm is effective.