采用排序和打分-搜索的多阶转换网学习方法  

Learning Multi-order Transition Networks Method Using Nodes Ranking and Local Search and Scoring

作　　者：冷翠平[1] 王双成[1,2] 王辉[3] 

机构地区：[1]上海立信会计学院数学与信息学院,上海201620 [2]上海立信会计学院开放经济与贸易研究中心,上海201620 [3]中央民族大学信息工程学院,北京100081

出　　处：《烟台大学学报（自然科学与工程版）》2010年第1期42-46,共5页Journal of Yantai University(Natural Science and Engineering Edition)

基　　金：国家自然科学基金资助项目(60675036);上海市教委重点学科(第五期)"国际贸易"和上海市教委重点资助项目(09zz202);上海高校选拔培养优秀青年教师科研专项基金资助项目(slx-07010)

摘　　要：针对现有学习方法主要用于先验网和一阶转换网,且对于多变量复杂转换网其效率和可靠性难以得到保障等问题,建立了一种从向量时间序列数据中发现多阶转换网的方法.在给出多阶数据集构造方法的基础上,通过条件相对平均熵计算建立完全有向无环图,并基于完全有向无环图排序结点,在结点顺序的基础上,通过局部打分-搜索建立转换网.这种方法将更加高效、可靠和实用.At present, learning dynamic Bayesian network are mainly used in building prior network and oneorder transition network. Problems such as inefficiency and unreliability exist in learning transition network with multi variables and complex structure. In this paper, a method of learning multi-order transition network by means of vector time series data is presented. First the method of setting up multi-order data set is given. The perfect directed acyelic graph is built through calculating the conditional relative average entropy. Then a transition network can be obtained by nodes ranking based on perfect directed aeyclic graph and local search and scoring. The method will be more efficient, reliable and practical.

关 键 词：动态贝叶斯网络 多阶数据集 多阶转换网 结点排序 局部打分-搜索 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]