混合地理加权回归模型算法研究  

Algorithm for Mixed Geographically Weighted Regression Model

作　　者：覃文忠[1] 王建梅[1] 刘妙龙[1] 

机构地区：[1]同济大学测量与国土信息工程系,上海市四平路1239号200092

出　　处：《武汉大学学报（信息科学版）》2007年第2期115-119,共5页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金资助项目(49971031);国家教育部长江学者奖励计划资助项目

摘　　要：以迭代算法为基础,推导出混合地理加权回归模型的常系数(全局参数)和变系数(局域参数)的计算方法,并以上海市住宅小区楼盘销售平均价格为例进行验证。结果表明,混合地理加权回归模型的计算量略大于地理加权回归模型,但对样本数据的拟合更好,局域参数估计更稳健。An iterative algorithm is developed to estimate global coefficients and local coefficients in MGWR. First independent variables are classified two groups, Group ag in which variables are global associated with global coefficients and Group bg in which variables are local associated with local coefficients. Second assuming that ag is known, coefficients of bg is calibrated by using the basic GWR. Third ordinary linear regression （OLR） is used to estimated coefficients of ag. Material formulations of two types of coefficients and computational progress are also produced, and further tested by using average prices of house blocks in Shanghai. The experiment proves that all formulations of coefficients are available, and comparison of the two models by Akaike information criteria value shows MGWR is more appropriate and stable for the local coefficients estimates than BGWR although it requires a greater computational effort.

关 键 词：地理加权回归模型 混合地理加权回归模型 空间非平稳性 迭代算法 空间分析 

分 类 号：P208[天文地球—地图制图学与地理信息工程]