基于SVR的无实测资料小流域山洪灾害临界雨量预估模型及应用--以河南新县为例

Threshold Rainfall Prediction Model for Flash Flood Disaster in Ungauged Small Watersheds Based on SVR:A Study Case in Xinxian County of Henan Province

山洪灾害是我国防灾减灾工作中亟待解决的突出问题,临界雨量是山洪灾害预警预报的重要指标,传统的临界雨量计算方法各有所长,对资料的要求各不相同,但对于无实测资料小流域防灾对象临界雨量计算的研究却寥寥无几(这里无实测资料指的是既没有雨量、洪峰资料也无野外断面调查资料)。鉴于此,基于支持向量机回归拟合算法(SVR),应用比拟法建立模型,对无资料小流域山洪灾害临界雨量进行推求。结果表明:(1)SVR模型具有很好的泛化性,预测精度高、误差小,结果较优;(2)利用SVR模型预估的临界雨量值与真实值偏差小,可以作为山洪灾害预警预报指标,为当地决策部门提供参考。

Flash flood disaster is a prominent problem to be solved in disaster prevention and mitigation in China.The rainfall threshold is one of the important indicators of flash flood disaster forecasting.The traditional calculation methods have their own advantages and different requirements for data.For some ungauged area,however,data are hard to collect.There are few studies on the calculation of rainfall in disaster prevention objects in small watersheds without measured data.In view of the condition mentioned above,based on the support vector machine regression fitting method(SVR),this paper established a model to deduce the rainfall threshold of flash flood disaster in small watersheds without data by using the comparison method.The results show that:(1)the SVR model has good generalization,high prediction accuracy,small error and better forecasting;(2)the difference between modeled rainfall threshold and real value is small and this model could be used as an early warning and forecasting of flash flood disaster.Additionally,it can also provide reference for departments of local decision-making.