大型数据库中多维空间数据智能插值方法仿真

Simulation of Multi-Dimensional Spatial Data Intelligent Interpolation Method in Large Database

为了解决大型数据库多维空间数据中存在的零散、稀疏问题,需要进行数据插值处理。针对采用当前插值方法对于样点空间分布及结构约束考虑较少,取样点数据的数量不能全面反映大型数据库中多维空间数据的整体分布情况,容易引起数据遗失、插值精度低的问题,提出一种基于径向基神经网络算法(RBF)的大型数据库中多维空间数据智能插值方法。首先对多维空间数据集进行时空分区,分别在时间和空间依据异质协方差模型计算多维空间缺失数据的估计值,进而采用关联系数确定大型数据库时空权重、融合时间和空间估计值获得多维空间缺失数据的估计结果。采用RBF算法对大型数据库中多维空间全局样本数据进行分类,再对各维数据进行训练。将蚁群算法和模拟退火算法相融合,并以此对径向基层训练过程进行优化,构成大型数据库中多维空间数据智能插值模型。实验结果表明,所提方法相比其它插值方法具有更优越的插值精度和适用性。

This article proposes an intelligent interpolation method of multi -dimensional spatial data in large database based on Radical Basis Function （RBF） neural network algorithm. Firstly,time -space division was carried out for multi - dimensional spatial data set and estimation value of missing data of multi - dimensional space was cal- culated in time and space respectively according to heterogeneous covariance model. And then space- time weight, fusion time and space estimation value were confirmed using correlation coefficient to obtain estimation result of the missing data. The research also employed the RBF algorithm to classify global sample data of multi - dimensional space in large database and trained each dimension data. Moreover, integrated with the ant colony algorithm, the re- search optimized grassroots - level training process. Experimental results show that the method has preferable interpo- lation precision and applicability compared with other interpolation methods.