摘要
提出了基于神经网络残余Kriging的场地液化势数据预测模型。利用传统的地质统计学方法中的交叉验证技术来寻找网络参数Spread。在最优网络参数下,GRNN网络较好地映射了场地液化势的非线性趋势,再借助于Kriging对残余分量进行数据空间结构分析和估值。计算结果表明,GRNN网络预测的非线性趋势较好地刻划了场地液化势数据的全局特征,其非线性映射能力要高于趋势面技术。去势后的场地液化势残余分量满足本征假设条件,可以很方便地用Kriging方法来估值。本法可以嵌入岩土工程决策系统对未勘察区域进行场地液化预测和评价。
This paper presents the prediction model for liquefaction potential of a two-dimension site based on the neural network residual Kriging. Neural parameter Spread can be searched by means of the cross-validation technique of Kriging. The large-scale nonlinear trend of spatial data can be modeled by means of neural network with the optimal parameter and the residual component can be analyzed by the ordinary Kriging. The study result indicates that the generalized regression neural network can map thenonlineal drift （trend）. The result obtained in the paper is better than that of the ordinary Kriging estimator. The model can be applied to the decision and evaluation system based on GIS to predict and evaluate the engineering properties of unexplored sites.
出处
《成都理工大学学报（自然科学版）》
CAS
CSCD
北大核心
2005年第4期368-372,共5页
Journal of Chengdu University of Technology: Science & Technology Edition
