基于预测能力的连续贝叶斯网络结构学习  

Learning of continuous Bayesian networks structure from data set based on forecasting ability

作　　者：董立岩[1] 苑森淼[1] 刘光远[2] 李永丽[3] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]吉林大学通信工程学院,长春130012 [3]东北师范大学计算机学院,长春130024

出　　处：《计算机工程与应用》2007年第9期23-24,48,共3页Computer Engineering and Applications

基　　金：国家自然科学基金(the National Natural Science Foundation of China under Grant No.60275026)。

摘　　要：通过对连续随机变量之间预测能力及其计算方法的讨论,提出基于预测能力的连续贝叶斯网络结构学习方法。该方法包括两个步骤,每个步骤都伴随环路检验。首先建立初始贝叶斯网络结构,其次调整初始贝叶斯网络结构,包括增加丢失的弧、删除多余的弧及调整弧的方向,并使用模拟数据进行了对比实验,结果表明该方法非常有致。In this paper,the definition of forecasting ability and its calculational method are presented between two continuous variables.A method of learning continuous Bayesian networks structure from data set based on forecasting ability is developed. This method is made up of two parts.Each part is combined with checking a cyclic route in a directed graph.Firstly,an elementary Bayesian network structure is set up.Secondly,this elementary Bayesian network structure is regulated,including to increase the losed arcs,to delete superfluous ares and to regulate direction of arcs.The experiment is made by using simulant data and the experimental results are shown by the means of contrasting.

关 键 词：连续贝叶斯网络 预测能力 最小切割集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]