PH分布参数估计的确定性退火加速EM算法  

Deterministic annealing acceleration EM algorithm for PH distribution parameter estimation

作　　者：尹东亮 胡涛[2] 陈童 谢经伟 YIN Dong-liang;HU Tao;CHEN Tong;XIE Jing-wei(Dept. of Operational Research and Programming, Naval Univ. of Engineering, Wuhan 430033, China;Dept. of Management Engineering and Equipment Economics, Naval Univ. of Engineering, Wuhan 430033, China)

机构地区：[1]海军工程大学作战运筹与规划系,武汉430033 [2]海军工程大学管理工程与装备经济系,武汉430033

出　　处：《海军工程大学学报》2020年第3期49-55,共7页Journal of Naval University of Engineering

基　　金：国家自然科学青年基金资助项目(71501183)。

摘　　要：针对连续PH分布参数估计标准EM算法迭代速度慢和初值敏感易陷入局部最优化的问题,首先引入逆温因子和加速因子两参数改进算法,提出了确定性退火加速EM算法;然后,进行了详细的理论推导,给出了PH分布参数估计的具体步骤;最后,利用算例对两类典型分布和一组某装备维修时间的实际数据进行拟合,验证了改进算法的全局最优性和较快的收敛速度,拟合效果和迭代加速情况也验证了该算法的有效性和通用性。For the normative Expectation Maximization algorithm of the continuous Phase-type distribution parameter estimation,of which the iterative speed is slow and the initial value is easy to fall into the local optimization problem,the inverse temperature factor and the acceleration factor are introduced into proposing deterministic annealing acceleration Expectation Maximization algorithm.Detailed theoretical derivation is performed and specific steps for phase-type distribution estimating parameters are presented.Finally,the global optimality and a faster convergence speed of the improved algorithm are verified by using examples to fit two typical distributions and the actual data of maintenance time of a certain equipment,and then the fitting effect and iterative acceleration are demonstrated,and the effectiveness and the versatility of the algorithm are proved.

关 键 词：phase-type分布 参数估计 EM算法 确定性退火加速EM算法 

分 类 号：F253.4[经济管理—国民经济]