基于递推最小二乘改进算法的洪水预报模型研究

由递推最小二乘算法估算出的自回归系数在一定条件下具有最佳的统计特性,但在实际应用中,这种方法往往难以动态地把握水文现象的动态特性.为提高自回归洪水预报模型的精度,分别用衰减记忆、有限记忆及2种算法相结合的方法对基本的递推最小二乘算法进行改进,并利用这几种改进算法对白马寺水文站的实测径流序列进行了模拟演算.结果表明,这3种改进的递推最小二乘算法,都可以使自回归洪水预报模型取得较好的预报效果,但实际应用时应根据不同预报的侧重点选择相应的算法.

Flood forecasting model based on improved recursive least square method

The auto-regression(AR) flood forecasting model parameters estimated by the recursive least square(RLS) method have optimum statistical characteristics under certain condition;however,it is difficult for the method to reflect the dynamic character of hydrological phenomenon in practical application.For improvement of the precision of the AR model,the basic RLS procedure was improved,and three improved forms,i.e.the faded-memory RLS procedure,the fixed-memory RLS procedure,and a combined form of the above two procedures,were deduced.The three improved RLS procedures were applied to simulate the measured runoff series of the Baimasi hydrological station.The results show that the accuracy of flood forecasting by the AR model based on three improved RLS procedures is satisfactory,but rational procedure should be selected according to the practical situation.