The surface runoff of inland rivers in arid area of Northwest China is a valuable available water resource.However,due to the complexity of runoff composition,runoff simulation is complex,and it is difficult to actually utilize the water resources.This paper applies the back-propagation (BP) neural network combined with the degree-day factor of snowmelt runoff model (SRM) to analyze and simulate the runoff process of Kaidu River Basin in Xinjiang.The results show that:①The simulation accuracy is high.The daily average temperature is the main factor affecting the runoff and process in the target river basin.②Using autocorrelation coefficient method to process daily average flow series can significantly improve the simulation accuracy of annual average daily runoff,especially for daily runoff in the snow melting period (March to May) of target river basin that cannot be effectively simulated by ordinary input.This study provides a new way for the rapid simulation of runoff supplied by glacier,snow melting and precipitation in arid area of Northwest China,and an effective reference for rational utilization of water resources. 西北干旱区内陆河流的地表径流是宝贵的可利用水资源,但是由于径流组成的复杂性导致径流模拟复杂,难以满足实际应用的需要.应用Back-Propagation (BP)神经网络结合Snowmelt-Runoff Model (SRM) 融雪径流模型中度日因子法对新疆开都河流域进行径流过程分析,模拟取得较好结果,同时发现:①日平均气温是影响目标流域径流量和过程的主要影响因子;②使用自相关系数法处理日均流量序列对全年日均流量模拟精度均有明显提升,尤其对普通输入无法有效模拟的目标流域融雪时间段(3-5月)的日均流量改进显著.为西北干旱区冰川、融雪和降水复合补给的径流的快速模拟提供了新的途径,也为水资源的合理利用提供了有效参考.