摘要
本文将从实际评估工作中提练出来的一种评估模型推广至因变量未知且带有一般性凸约束条件的广义线性模型,证明了模型解的存在唯一性,并从解的几何背景出发,提出了基于凸集间交互投影的参数最小二乘估计的有效算法.结合模型的特点,引入EM算法给出了参数的极大似然估计.模型的提出丰富了线性模型的结构框架,算法的给出为参数估计提供了行之有效的计算方法.
In this paper, we generalize an evaluation model extracted from the practical work to the generalized linear regression model with convex constraints, whose dependent variables are unknown. On the basis of proving the existence and uniqueness of the solution, from an angle of geometric meaning,we present an iterative algorithm by alternating projection between two convex sets to solve the least square estimation. Meanwhile, in view of the characteristic of the model,we take advantage of EM algorithm to accomplish the maximum likelihood estimation of parameters. The model proposed in this paper extends the structural framing of linear model and the presentation of algorithm realizes the effective computation of parameter estimation.
出处
《应用数学》
CSCD
北大核心
2008年第4期635-639,共5页
Mathematica Applicata
基金
国家自然科学基金项目(30570611)
关键词
凸约束
广义线性回归
参数估计
交互投影
EM算法
Convex constraint
Generalized linear regression
Parameter estimation
Alternating projection
EM algorithm
