作　　者： (胡前芳）; (李保坤）;

机构地区： 西南财经大学统计学院,四川成都611130

出　　处： 《统计与信息论坛》 2017年第9期17-23,共7页

摘　　要： 针对EM算法在估计混合正态分布参数时使用不完全信息的总样本所得到的参数估计误差较大的问题,提出一种新的估计方法——TU截断改进算法。该算法根据正态分布的特点,运用部分拥有完全信息的样本将混合正态分布中的分布参数逐一估计出来。这一算法一方面克服了EM算法运用于混合分布的缺陷,另一方面改进了使用截尾数据的参数估计。仿真结果表明,TU算法比EM算法估计更精确。 As EM algorithm estimation errors are large for mixture normal distribution samples with incomplete information,a concise explanation is given for this phenomenon and a new estimation method is proposed, which is called TU algorithm. Taking advantage of normal distribution properties, with TU method we estimate the parameters of component normal distributions one by one. On one hand TU overcomes the weakness of EM in this problem, on the other hand estimation is upgraded for truncated data. Simulation results show that TU algorithm is more accurate than the EM algorithm in estimation of the mixture of normal distributions.

关 键 词： 混合正态分布 算法 算法