黄土区不同土地利用方式下径流量的神经网络模拟

Runoff Simulated by Neural Network Under Different Landuses on the Loess Area

基于黄土坡面降雨—径流关系的复杂性且非线性特性,引用3层前馈型BP网络模型,对不同土地利用方式(草灌地、刈割地、翻耕地)径流量进行模拟,以植被盖度、降雨强度、坡度、土壤前期含水率和土壤容重5个因子作为输入层变量,次降雨下径流量作为输出层变量,并利用野外人工模拟降雨试验所得到不同降雨强度下各类土地利用径流小区的径流量实测资料,对网络进行模拟训练和预测,取得了较好的结果,平均误差不超过10%。研究结果表明,与传统回归统计方法进行了误差比较,该模型的预测精度更高。

Based on the complex nonlinear characteristics of rainfall runoff on slope, a three-layer feedforward back-propagation network model for runoff in different landuses （grass and shrubby slope, cutting slope and plowing slope） is established. Structure of the model has five input variables, including vegetation coverage, rainfall intensity, gradient, antecedent soil moisture content, and soil bulk density and one output variable of runoff amount from single rainfall event. The network model is validated by using the data ob- served from field simulated rainfall experiment. The BP network model is compared with conventional method as well. Results show that the BP network model may improve the precision of forecast.