中文摘要：

针对导致黄河下游三角洲地区土壤盐渍化的浅层地下水因素,以该地区典型区域为研究对象,将人工神经网络引入地下水矿化度的模拟和预测中,建立了基于土壤盐分、地下水埋深和pH的地下水矿化度预测的BP神经网络模型,并与多元回归模型在拟合精度和预测性能方面进行了比较。结果表明：研究区域地下水矿化度与土壤盐渍化程度呈显著的相关性,多元回归模型能较好地拟合地下水矿化度;通过网络训练确定了地下水矿化度的BP神经网络的拓扑结构为5∶8∶1,BP神经网络的拟合精度明显优于多元回归模型;统计检验表明BP神经网络的预测性能亦优于多元回归方法,其预测精度提高了50.1%。该研究可为黄河三角洲地区盐渍化的水盐调控和预测预报提供理论基础与决策依据。

英文摘要：

In view of the shallow groundwater factor resulting in soil salinization of Lower Yellow River Delta, artificial neural net- work was introduced for modeling and prediction of groundwater mineralization which was performed in typical region of this area. Back propagation neural network （BPNN） model was established for modeling groundwater mineralization with surface soil salinity, sub-surface soil salinity, subsoil salinity, groundwater depth and pH value included as input vectors. Then the fitting precision and prediction accuracy between BPNN and multiple regression models were compared. Results indicated that groundwater mineralization showed significant correlation with soil salinization across the study area, and multiple regression model fitted the groundwater mineralization well. The BPNN topological structure of groundwater mineralization was determined as 5 ： 8 ： 1 through network training, and the fitting precision of BPNN was superior to multiple regression model. Statistical test showed prediction accuracy of BPNN was also higher than multiple regression model with an increase by 50. 1%. The study provides a theoretical and decision- making basis for soil water and salt regulation and forecasting of soil salinization in Yellow River Delta.