自动选取多非线性参数初始值的循环搜索法  

A CYCLIC SEARCH METHOD CHOOSING THE INITIAL VALUE OF SEVERAL NONLINEAR PARAMETERS AUTOMATICALLY

作　　者：朱珉仁[1] 李勇[2] 

机构地区：[1]东南大学数学系,南京210096 [2]香港中文大学统计系

出　　处：《高等学校计算数学学报》2006年第3期252-261,共10页Numerical Mathematics A Journal of Chinese Universities

摘　　要：1几何中心化加权最小二乘法 给定观测数据集 （xi,yi）,i=1,2…,n.In nonlinear regression analysis, one of the most important problems that the iterative algorithms for computing the MLE often meet is to choose the initial value. A bad initial value can often make the algorithm lose effect. However, until now there are still no good methods to solve the problem. In this paper, as to the MLE problem of the weighted nonlinear regression model, we propose geometric centralized weighted least square method and use the cyclic search method to automatically choose the initial value of several nonlinear parameters through the observed data. We make the algorithm less dependent on the initial value and figure out how to find the initial value concretely. At the end,several numerical examples are given in order to explain the application of this algorithm.

关 键 词：循环搜索法 非线性参数 自动选取 初始值 加权最小二乘法 中心化 数据集 

分 类 号：O229[理学—运筹学与控制论]