广义I型逐阶区间删失混合Weibull数据的参数估计

Parameters estimation of generalized progressive type-I interval-censored mixture Weibull-distributed data

源于有限混合总体的广义I型逐阶区间删失数据参数估计方法的研究不多,基于有限混合Weibull模型,讨论Expectation-Maximization(EM)算法对广义I型逐阶区间删失数据参数估计的有效性及其改进。首先给出估计参数的EM算法,通过仿真算例,说明EM算法对广义I型逐阶区间删失混合数据参数估计产生了过度迭代现象,进而,提出了停止EM算法的加权绝对偏差信息准则。改进的EM算法改善了EM算法无法确定参数估计停止迭代时刻的不足,在选择适当初值后,可快速获得满意的参数估计结果。仿真算例验证了该方法的有效性。

This paper is a response to the previously insufficient study of the method designed for the parameter estimation for the generalized progressive type-I interval-censored data based on the finite mix- ture. The study building on finite mixture Weibull-distribution model looks at the effectiveness of Expec- tation-Maximization（EM） algorithm used for the parameter estimation for generalized progressive type-I interval-censored data and produces a modified EM algorithm. The modification starts with giving the EM algorithm used for the parameter estimation, followed the simulation examples to explain the excessive it- eration produced by EM algorithm when used for the parameter estimation for the generalized progressive type-I interval-censored data, and culminates in the necessity for stopping weighted absolute deviance in- formation criterion involved in EM algorithm. The modified EM algorithm is free from the drawback inher- ent in EM algorithm incapable of determining the time right for stopping iteration when used to perform parameter estimation, thus allowing for a quick production of satisfactory estimators, following an appro- priate selection of the initial value. The simulation verifies the viability of the algorithm.