稀疏数据插值问题的回归克里格方法

Regression Kriging Method for Sparse Data Interpolation

针对稀疏数据的插值问题,基于克里格方法提出了一套应用于数据量少、插值空间广泛的回归克里格方法。研究结果表明,该方法可有效解决稀疏数据无法给出合理理论变异函数的问题,且插值结果与基于最小二乘法的简单克里格相比平稳性更好。回归克里格方法应依据具体问题选择变量,以满足强相关为首要条件,过多增加相关变量往往不能取得很好的插值效果。

This paper focuses on the sparse data interpolation which is based on Kriging method, and an improved method named Regression Kriging method is derivated. This method can be applied in the issues which have sparse data and broad space. The conclusion shows that the regression Kriging method is much better than other Kriging methods and the interpolation results are much stationary than the least square method solution. Besides this, when this method is applied for multivariate scenario, the most important condition which should be satisfied is strong correlation and excessive correlation variants do not make better results definitely.