中文摘要：

为开展长江河口地区土壤盐分动态的中长期模拟与预测，采用人工神经网络中应用较为成熟和广泛的BP网络建立长江河口地区土壤盐分与降雨量、蒸发量、长江水电导率、内河水电导率、地下水位、地下水电导率6因子间的非线性神经网络响应模型。网络模型结构为6-11-1，隐含层单元数用“试错法”确定。选择合适的参数训练和学习网络模型后，对河口地区2003年各月平均根层土壤电导率进行预测，并与线性回归模型预测结果进行比较。结果表明：BP网络模型较线性回归模型具有更高的预测精度，平均相对预测误差为7.3％，预测值与实测值相关性良好，可以满足实际应用需求。

英文摘要：

In order to conduct a medium-long term simulation and prediction of the soil salt dynamics in the Yangtze River estuary, a nonlinear artificial neural network response-model among 6 factors of soil salt, rainfall, evaporation, river water EC, freshwater EC, water table and groundwater EC was established with the BP network which has been applied maturely and widely in all kinds of artificial neural networks. The structure of the network model was fixed on 6-11-1. The cells number of the hidden layer was confirmed using the Trial-and-Error method. After the network model was trained and tested by choosing appropriate parameters, it was applied to predict the average root-layer soil EC in the Yangtze River estuary in every month of 2003. The predicted result from the network model was compared with that from the linear regression model. Results showed that the BP network model had higher precision in predicting the salt dynamics than the linear regression model. The average relative error was 7.3% and the correlativity between predicted value and actual value was all right, both of which showed that the simulation and prediction of the BP artificial neural network could meet the need of the practical application.