中文摘要：

净灌溉需水量是估算农业灌溉用水量的参考依据。该文以西北干旱内陆区石羊河流域上中下游的古浪县,凉州区,民勤县为研究对象,在分析农业净灌溉需水量宏观驱动力因子（1959-2005）的基础上,以关键因子作为输入项,区域农业净灌溉量为输出项,分别建立农业净灌溉需水量的多元线性回归模型、人工神经网络BP模型以及人工神经网络集成模型。并对不同模型的模拟效果进行比较;通过对时间序列的离散化蒙特卡洛（MC）设计,采用不确定性评价指数（d-factor）对3种模型模拟的不确定性进行分析。结果表明：与多元线性回归模型和神经网络BP模型相比,神经网络集成模型具有较高的模拟精度,并能合理地指示影响因素与净灌溉需水量的不确定性变化。

英文摘要：

Net irrigation requirement is the major component of agricultural water use.In this study,the regional net irrigation requirement of agricultural area in Shiyang river basin in semi-arid region of Northwest China was analyzed.Based on the zones of local agricultural water use,the key factors are chosen by driving factor analysis（1959-2005）,and then regional net irrigation requirement was simulated using multiple-linear regression（MLR）,artificial neural network（ANN） and Ensemble ANN models.Then discrimination time-series Monte-Carlo（MC） simulation is used for inputs uncertainty analysis of these models.Results suggest that compared with MLR and ANN,the Ensemble ANN model show better ability in the simulation of regional net irrigation requirement with smallest error and lowest uncertainty index.