基于遗传算法的动态Bayesian网结构学习的研究

动态Bayesian网是复杂随机过程的图形表示形式,从数据中学习建造动态Bayesian网是目前的研究热点问题.本文针对该问题提出了一种遗传算法.文中设计了结合数学期望的适应度函数,该函数利用进化过程中的最好动态Bayesian网把不完备数据转换成完备数据,使动态Bayesian网的学习分解为两个Bayesian网(初始网和转换网)的学习,简化了学习的复杂度.此外,文中给出了网络结构的编码方案,设计了相应的遗传算子.模拟实验结果表明,该算法能有效地从不完备数据序列中学习动态Bayesian网,并且实验结果说明了隐藏变量的作用和遗传控制参数对结果模型的影响.

Research on Learning Dynamic Bayesian Networks by Genetic Algorithms

Dynamic Bayesian networks are a representation for complex stochastic processes.How to learn structure of Dynamic Bayesian networks from data is a hot problem of research.An evolutionary algorithm is proposed.Fitness function based on expectation is presented to convert incomplete data to complete data utilizing current best dynamic Bayesian network of evolutionary process.Thus dynamic Bayesian networks can be learned by using two Bayesian networks,prior network and transition network,to reduce the computational complexity.Encoding is given,and genetic operators are designed which provides guarantee of convergence.Experimental results not only show this algorithm can be effectively used to learn Dynamic Bayesian networks structure from incomplete data sequences,but also illustrate the role of hidden variables and the influence of genetic control parameters on learned model.