Abstract: To solve the slow convergence problems of Phase-type(PH) distribution data fitting EM algorithm, this paper presents a numerical acceleration EM algorithm. The algorithm achieves the purpose of acceleration by increasing the parameters change of every EM iterative. To illustrate the effectiveness of the acceleration fitting method, the method is compared with the standard EM algorithm by using four benchmark examples. Comparison results show that the acceleration EM algorithm is simple and practical, it can guarantee the convergence of the algorithm, and effectively raise the convergence speed of the fitting algorithm.

Key words: Phase-type(PH) distribution, EM algorithm, data fitting, numerical acceleration

摘要： 摘 要：针对Phase-type(PH)分布数据拟合EM算法收敛速度慢的问题，提出一种数值加速EM算法，通过增加每一步EM迭代的参数变化量达到加速的目的。用4个拟合实例与标准EM算法拟合进行对比，结果表明，该加速EM算法简单实用，保证了算法的收敛性，有效提高了PH分布数据拟合EM算法的收敛速度。

关键词: PH分布, EM算法, 数据拟合, 数值加速

CLC Number: 

• TP393