Runoff prediction plays an important role in the optimal allocation of water resources and flood control and drought relief.To solve the problem of large prediction errors caused by non-smoothness and extreme values of runoff se-ries and improve the prediction accuracy,this paper proposes a combined prediction model(LCD-SSA-BiLSTM)based on local characteristic-scale decomposition(LCD),sparrow search algorithm(SSA)and bi-directional long short-term mem-ory(BiLSTM)to study the monthly runoff series of four stations in the upper reaches of Fenhe River(Fenhe Reservoir Station,Shangjingyou Station,Lancun Station and Zhaishang Station).Nash efficiency coefficient(NNSE),mean abso-lute error(MMAE),root mean square error(RRMSE),and qualification rate(QQR)are used to quantitatively evaluate the prediction results.Compared with the single BiLSTM model,EMD-BiLSTM model,LCD-BiLSTM model and EMD-SSA-BiLSTM model,the results show that the LCD-SSA-BiLSTM model has higher prediction accuracy with MMAE of 10.346×104-124.629×104 m3,RRMSE of 19.416×104-191.284×104m3,NNSE of 0.975-0.988,and the QQR of all four hydrological stations were 90%and above,and the prediction accuracy was grade A.Thus,the LCD-SSA-BiLSTM mod-el is an effective method to predict non-stationary monthly runoff series.
upper reaches of the Fenhe RiverBiLSTM modelLCDmonthly runoff prediction
NETL
NSTL
维普
万方数据
