摘要
针对贝叶斯网络结构学习的复杂问题,首先在深入研究网络结构特点基础上建立了1-步依赖系数BN邻接矩阵,将网络结构学习问题转化为邻接矩阵边的关系处理;然后,采用爬山算法进行局部寻优,再用遗传算法进行全局优化的思想进行贝叶斯网络结构求解。算例分析表明:所提算法可行,计算效率高于遗传算法。
Aiming at the complex problems of Bayesian network structure learning, an in-depth study of network structure characteristics is conducted, and a 1-step dependent coefficient adjacency matrix is established by which the problem of network structure learning is transformed into the relationship of adjacency matrix. And then the genetic algorithm is used to solve the problem of the global optimization whereas the mountain climbing algorithm is used to solve the problem of the local optimization. Confirmatory calculations prove the feasibility of the proposed algorithm. Comparison of the proposed algorithm with genetic algorithm shows that it works more efficiently than the latter.
作者
李东亮
许伟
吴迪
邢广笑
LI Dong-liang;XU Wei;WU Di;XING Guang-xiao(College of Power Engineering,Naval Univ.of Engineering,Wuhan 430033,China;Dept.of Marine Replenishment,Naval Logistics Academy,Tianjin 300450,China;Unit No.92001,Qingdao,266000,China)
出处
《海军工程大学学报》
CAS
北大核心
2019年第6期61-64,共4页
Journal of Naval University of Engineering
基金
湖北省自然科学基金资助项目(2013CFB440)
关键词
贝叶斯网络
结构学习
优化
Bayesian network
structure learning
optimization
