混合时空地理加权回归及参数的两步估计

摘要

In response to a phenomenon that both global stationary characteristics and spatial-temporal non-stationary characteristics exist at the same time,an approach named mixed geographically and temporally weighted regression (MGTWR) was proposed.This paper showed mathematical definition of MGTWR and gave the formula of regression parameters by using two-step estimation method.Besides,the weight calculation method and the parameter optimization method based on Akaike information criterion (AIC) were introduced.Some simulated data with different degrees of complexity were adopted to test the performance of method.Result shows that R2 are more than 0.8 when MGTWR and GTWR are used.Both MGTWR and GTWR can deal with the phenomenon that both global stationary characteristics and spatial-temporal non-stationary characteristics have.What's more,MGTWR is better than GTWR.As MGWR cannot detect temporal non-stationary characteristics,the results of MGWR are bad.In addition,the complexity of the data affects the performance of MGTWR,GTWR and MGWR.The simpler the data are,the better the results will be.%针对全局平稳特征和时空非平稳特征同时存在的现象,提出了混合时空地理加权回归方法(Mixed Geographically and Temporally Weighted Regression,MGTWR),给出了MGTWR的数学定义和回归参数的两步估计.同时,介绍了调整型带宽选择机制下的权重计算方法和基于Akaike信息准则(Akaike Information Criterion,AIC)的参数优化方法.采用复杂程度不同的模拟数据来测试方法的性能.结果表明,MGTWR和GTWR的R2大于0.8,能模拟全局平稳特征和时空非平稳特征的现象,但MGTWR显著优于GTWR.MGWR因无法探测时间平稳特征,模拟效果最差.此外,数据复杂程度影响MGTWR,GTWR和MGWR的性能,数据越简单模拟效果越好.