中文摘要：

开展高山寒区径流预报对合理开发利用我国西北地区水资源有重要意义，由于恶劣自然环境造成的观测困难、干扰因素较多等问题，建立简单有效的径流预报模型是研究高山寒区水文规律的途径之一。近年来，人工神经网络技术作为一种简单有效的新方法被广泛应用于水文预报，但在冰川融雪为主的流域径流预报中的应用较少，本文以乌鲁木齐河源1号冰川区为研究对象，构建了高山寒区冰川作用区径流预报的前馈型人工神经网络模型（BP-ANN）。通过1号冰川水文站各水文要素之间的相关分析初步确定网络的输入，以Nash效率系数最大等为目标函数，优选网络结构，并在此基础上对所优选网络结构的合理性及模型预见期进行了分析。

英文摘要：

Artificial neural network （ANN） model is intensively exploited for hydrological prediction. This is the case of predictions for snow/glacier melting which is of critical importance for sustainable utilization of water resources in the arid plain oases of Northwest China. In this paper, a parsimonious optimization ANN model based on back propagation algorithm is developed to simulate and predict runoff in high and cold mountainous regions. The source drainage area of the Urumqi River in Northwest China is selected for the case study. The network inputs, i. e., the preceding daily positive accumulative temperature and previous runoff are determined by the correlative analysis, and the network structure is optimized with the maximum Nash coefficient as the objection function. The detailed study has also been conducted to test the effect of alternative inputs and forecasting periods on model performance which suggests that 3 - 3 - 1 network, i.e., 3 inputs （2 days preceding positive accumulative temperature and one day preceding runoff） and 3 hidden nodes, is the optimal network structure. The reasonability of optimized network structure and the effect of different forecast periods for model performance have also been studied, which would be helpful for runoff prediction in high and cold mountainous regious.