Monthly runoff prediction using modified ceemd-based weighted integrated model

Abstract

Due to the nonlinear characteristics of runoff data and the poor performance of the single prediction model, a weighted integrated modified complementary ensemble empirical mode decomposition (MCEEMD)-based model was proposed to predict the monthly runoff of three hydrological stations in the lower reaches of the Yellow River. In this model, particle swarm optimization (PSO) was used to optimize the parameters of support vector regression (SVR), back propagation neural network (BP), long short-term memory neural network (LSTM) that constitute it. The weight coefficients and frequency terms decomposed by MCEEMD were used to obtain the final prediction results. Results indicated that this model performs better than other models, with the Nash–Sutcliffe efficiency (NSE) reaching above 0.92, qualification rate (QR) reaching above 75% and all error indicators being minimal. In addition, considering the influence of extreme weather and climate anomalies, the integrated model combined the atmospheric circulation anomalies factors and the results can still be improved. It can be verified that this weighted integrated model can be used for the stable and accurate predication of medium-and long-term runoff.