基于MDL原理与混合遗传算法的Bayesian网络结构学习

Learning Bayesian Network Structures Based on MDL and Hybrid Genetic Algorithms

从大型数据库中学习Bayesian网络结构是Bayesian网络应用的难点之一。在分析标准遗传算法与爬山算法各自优点与不足的基础上,将这两种算法相结合,以最小描述长度为评价函数,得到一种混合遗传算法,实现了它们的优势互补。文章给出了混合遗传算法的计算步骤,并通过对ALARM数据库学习得到的Bayesian网络结构。

One of the difficulties of the application of Bayesian Networks is that when the data arise, it is very hard to learn the structures of Bayesian Networks from large databases. Both Standard Genetic Algorithms and Hill-Climbing can be used in structure learning, but none of them can get proper result easily. The combination of the two algorithms can have better effect. The ALARM Network is learned, Hybrid Genetic Algorithms is usedthen Bayesian Network structure is got as the result after the Minimum Description Length is selected as the fitness function.