Abstract: A method of estimating tree models via Expectation-Maximization（EM） algorithm is addressed，and a matrix version of EM algorithm which is specific to the structure characteristics of the tree models is presented.Based on original algorithm，a series of matrixes has been employed to stand for all the information including in the original model，which makes it possible to estimate the parameter for the tree models via matrix operations.Compared with the traditional algorithm，the experiment shows the proposed approach has superior structure organization with improved efficiency and flexibility，which can demonstrate the advantage under higher accuracy requirement or heavy duty situation.

Key words: tree models, EM algorithm, parameter identification, matrix operation

摘要： 研究了运用EM算法对树状模型参数辨识的方法，针对树状模型结构特性提出了一种矩阵形式的EM算法。在原有算法的基础上，由一系列矩阵分别描述出原始模型的全部信息，以矩阵运算的形式完成原算法的迭代过程，从而获得模型各参数估计值。实验结果表明，与原有算法相比，本方法具有更好表达形式和算法性能，在高精度和重负荷情况下更能显示出其优势。

关键词: 树状模型, EM算法, 参数辨识, 矩阵运算