基于蚁群节点寻优的贝叶斯网络结构算法研究  

Study on Bayesian network structure learning algorithm based on ant colony node order optimization

作　　者：刘浩然[1] 孙美婷[1] 李雷[2] 刘永记 刘彬[1] 

机构地区：[1]燕山大学信息科学与工程学院,秦皇岛066004 [2]燕山大学电气工程学院,秦皇岛066004

出　　处：《仪器仪表学报》2017年第1期143-150,共8页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(51641609);河北省自然科学基金(F2016203354)项目资助

摘　　要：K2算法是学习贝叶斯网络结构的经典算法。针对K2算法依赖最大父节点数和节点序的不足,以及蚁群算法搜索空间庞大的问题,提出了一种新的贝叶斯结构学习算法-MWST-ACO-K2算法。该算法通过计算互信息建立最大支撑树(MWST),得到最大父节点数;然后利用蚁群算法(ACO)搜索最大支撑树,获得节点顺序;最后结合K2算法得到最优的贝叶斯网络结构。仿真实验结果表明,该方法不仅解决了K2算法依赖先验知识的问题,而且减少了蚁群算法的搜索空间,简化了搜索机制,得到较好的贝叶斯结构。最后将该算法应用到冀东水泥回转窑的实际数据中,构建水泥回转窑的贝叶斯网络结构,提高了故障诊断的准确率。K2 algorithm is the classical learning algorithm of Bayesian network structure. Aiming at the problems that K2 algorithm depends on the maximum number of parent nodes node order and ant colony optimization algorithm has large search space,this paper proposes a new Bayesian structure learning algorithm- MWST-ACO-K2 algorithm. Firstly,through calculating the mutual information,the algorithm establishes the Most Weight Supported Tree（ MWST） and obtain the maximum number of parent nodes. Secondly,ant colony optimization algorithm is adopted to search the Most Weight Supported Tree and obtain the node order. Finally,combining with K2 algorithm,the proposed algorithm can obtain the optimal Bayesian network structure. The simulation experiment results show that the proposed algorithm not only solves the problem that K2 algorithm relies on prior knowledge,but also reduces the search space of ant colony algorithm,simplifies the search mechanism and obtains good Bayesian structure. The proposed algorithm was applied to the operation data of the cement rotary kiln in Jidong Cement Company,established the Bayesian network structure model of the cement rotary kiln and achieved precise and rapid fault diagnosis.

关 键 词：互信息 蚁群优化 K2算法 贝叶斯网络结构学习 水泥回转窑 

分 类 号：TH165.3[机械工程—机械制造及自动化]