混合高阶隐马氏模型的EM算法  

EM ALGORITHM OF MIXTURE HIGH-ORDER HIDDEN MARKOV MODEL

作　　者：叶飞[1,2] 王翼飞[2] 

机构地区：[1]铜陵学院数学与计算机科学系,铜陵244000 [2]上海大学数学系,上海200444

出　　处：《系统科学与数学》2013年第3期334-350,共17页Journal of Systems Science and Mathematical Sciences

基　　金：国家自然科学基金(30871341);上海市重点学科建设项目(S30104);上海市教委重点学科建设项目(J50101);科技部重大科技专项基金项目(2009ZX09103-686)

摘　　要：提出了一类混合高阶隐马氏模型.基于模型降阶法,通过Hadar等价变换方法将混合高阶隐马氏模型分别转换为与之等价的两个混合一阶隐马氏模型,然后利用混合一阶隐马氏模型的EM算法建立了混合高阶隐马氏模型的EM算法,给出了混合高阶隐马氏模型的参数重估公式,并使用一阶隐马氏模型的标准技术来计算这些参数重估公式.该研究推广了Rabiner,Hadar和杜世平等人的结果.A class of mixture high-order hidden Markov model is proposed. Based on the model reduction method, mixture high-order hidden Markov model is trans- formed into two equivalent mixture first-order hidden Markov models respectivelyby using Hadar＇s equivalent transformation method, and then the EM algorithm of mixture high-order hidden Markov model is developed and derived by means of the EM algorithm of an equivalent mixture first-order hidden Markov model. Also, the parameter reestimation formula of mixture high-order hidden Markov model can be expressed by using standard techniques applicable to the first-order hidden Markov model. Some results in the literatures published by Rabiner~ Hadar, and Du etc. are extended.

关 键 词：一阶隐马氏模型 混合一阶隐马氏模型 高阶隐马氏模型 混合高阶隐马氏模型 EM算法 

分 类 号：O211.62[理学—概率论与数理统计]