中文摘要：

针对传统的时空克里格算法的精度受到时空变异函数的影响,而时空变异函数理论模型的选择常受主观因素影响和理论半变异函数局限,没有普适性的建模方法;加之参数较多估计困难,致使插值精度不高的问题,该文提出一种普适性的基于广义回归神经网络自适应时空克里格插值变异函数拟合方法,在此基础上建立了广义回归神经网络与时空克里格结合的新颖时空混合插值算法。通过与传统插值方法在民勤县地下水埋深插值中的比较研究表明,该时空混合插值算法的插值精度显著提高,并且是一个普适性的插值法。

英文摘要：

Any point in the space-time field can be interpolated by spatio-temporal Kriging method,and the interpolated accuracy is largely influenced by spatial and temporal variogram.In traditional research,choosing spatio-temporal variogram models is often influenced by subjective factors and theoretical semivariogram is limited,there is no universal method for spatio-temporal interpolation,and the parameters are complex to determine,thus resulting in poor accuracy.Aiming at this problem,the GRNN-STK hybrid spatio-temporal interpolation method was proposed in this paper by introducing GRNN neural network to adaptively fit the experimental variogram of actual data.The improved spatio-temporal kriging and several other comparative models were applied to Minqin county groundwater level data respectively for interpolation experimental comparison.Comparative results showed that the mixed interpolation accuracy of proposed method was increased significantly.It is a universal interpolation method.