基于PSO-CV算法的信息扩散插值模型及其在黄河源区的应用

Information Diffusion Interpolation Model Based on the PSO-CV Algorithm and Its Application on Source Regions of the Yellow River

根据有限的雨量站观测资料进行空间插值是探索降水空间特征的有效手段。介绍了一种基于信息扩散理论的空间插值模型,并在已有经验信息扩散插值模型的基础上,考虑其关键参数估计的不足,提出一种基于粒子群算法(PSO)与交叉验证(CV)相结合的最优信息扩散插值模型。以黄河源区为例,引入反距离加权法、普通克里金法、泛克里金法及考虑高程的协克里金法作为对比参照,分别从年、季、月和日4种时间尺度评价了信息扩散插值模型的插值效果。结果表明,整体来看最优信息扩散插值模型的精度最高,反距离加权法的精度最低,经验信息扩散插值模型与其他3种空间插值方法的精度差异不大,不同方法间的精度差异随着时间尺度的减小而减小。

The spatial interpolation based on the limited observation data of precipitation stations is an effective means to explore the spatial characteristics of precipitation.This paper briefly introduces a spatial interpolation model based on the information diffusion theory,proposes an optimal information diffusion interpolation model based on particle swarm optimization(PSO)and cross-validation(CV)with consideration of the disadvantages on key parameter estimation of existing experience information diffusion interpolation model,and evaluates the interpolation effect of the information diffusion interpolation model in terms of four time scales of year,quarter,month and day with inverse distance weighting,ordinary kriging method,universal kriging method and co-kriging method considering the elevation as comparison,taking the source regions of the Yellow River as an example.The results show that overall,the optimal information diffusion interpolation model has the highest accuracy,while the inverse distance weighting has the lowest accuracy.The accuracy of experience information diffusion interpolation model does not differ much from the other three spatial interpolation methods,and the accuracy differences between different methods decrease as the time scale decreases.