混合EM算法研究及聚类应用

混合高斯模型能够有效地拟合概率密度函数,常用的混合高斯概率密度模型参数估计方法是EM算法,这种算法的缺点是估计精度过分依赖于初始值,不能估计模型阶数,容易导致协方差矩阵出现奇异。基于遗传算法的Annealing-EM算法可以同时估计模型阶数和参数,有效地克服协方差矩阵出现奇异,将混合算法应用到聚类中,仿真结果表明该算法具有更好的聚类效果。

Mixed EM Algorithms and Its Clustering Application

The probability density distribution could be efficiently modeled using Gaussian mixtures.EM is one of popular algorithms for parameters estimation of Gaussian mixture probability density model.However,this method highly depends on initial parameters,and could not estimate the number of model ordens.Annealing-EM algorithms based on genetic algorithm could estimate the number and model parameters and orders,thus could offectively,and avoid the occurrence of singularity in covariancematrix.The simulation results with Matlab indicate that this methods is of excellent clustering ability.