贝叶斯网络分类器的基于改进粒子群参数学习方法  

A parameter learning method of Bayesian network classification based on improved particle swarm

作　　者：丁晓彬 刘久富[1] 郑锐 王彪[1] 刘海洋[2] 杨忠[1] 王志胜[1] DING Xiaobin;LIU Jiufu;ZHENG Rui;WANG Biao;LIU Haiyang;YANG Zhong;WANG Zhisheng(School of Automation,Nanjing University of Aeronautics and Astronautics,Nanjing,210016,China;School of Electronic Science and Engineering,Southeast University,Nanjing,211189,China)

机构地区：[1]南京航空航天大学自动化学院,江苏南京210016 [2]东南大学电子信息工程学院,江苏南京211189

出　　处：《应用科技》2019年第4期32-36,41,共6页Applied Science and Technology

基　　金：国家自然科学基金项目(61473144)

摘　　要：研究了贝叶斯网络分类器的高效参数学习方法。生成方法解决联合分布的参数估计问题,而判别方法解决后验分布的参数估计问题。对判别参数学习方法的研究,首先通过建立类条件贝叶斯网络模型;在此基础上,对该模型以对数形式参数化,得到判别类条件贝叶斯网络模型;最后,通过改进粒子群算法对该模型进行最优化求解,得到各节点的概率。将贝叶斯网络分类器的判别参数学习方法与TAN分类器相结合,可用于对液体火箭发动机的故障诊断与分类中。针对某型号火箭的两次仿真数据进行故障诊断与分类,与其他方法相比,改进的分类器需要的数据量小,准确率和学习效率更高。The efficient parameter learning method of Bayesian network classifiers is analyzed.Generative methods address the estimation of the parameters of the joint distribution.However,discriminative methods address the esti?mation of the parameters of the posterior distribution.A discriminative parameter learning method is proposed.First?ly,the class-conditional Bayesian network model is established.On this basis,the model is parameterized by loga?rithmic form and we obtain discriminative class-conditional Bayesian network model(CCBN).Finally,the solution of the model is obtained by improved particle swarm optimization algorithm,and the probability of each node is got?ten.The discriminant parameter learning method of Bayesian network classifier combined with the TAN classifier can be used for fault diagnosis and classification of liquid rocket engine.The fault diagnosis and classifications on two simulation data sets of a certain type of rocket are carried out.Compared with other methods,the improved clas?sifier needs less data and has higher accuracy and learning efficiency.

关 键 词：贝叶斯网络 判别参数学习 改进粒子群 故障诊断 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]