Abstract: Extreme-point Symmetric Mode Decomposition (ESMD) is used to predict runoff series based on a runoff forecasting model to solve two problems after runoff series decomposition－large fluctuation ranges of high frequency components and poor forecast accuracy. We use the stationary processing technique of the ESMD to decompose the runoff series, select the best prediction method by analyzing the characteristics of different frequency components, combine Particle Swarm Optimization (PSO) and Least Square Support Vector Machines (LSSVM) for the prediction of high-frequency components, and use the back-propogation (BP) neural network for the prediction of mid- and low-frequency components. A combined ESMD-PSO-LSSVM-BP forecasting model is constructed to forecast annual and monthly runoffs at three hydrological stations in the upper and middle reaches of the Xijiang River. The results show this model, using different forecasting methods for different frequency components, improves the runoff forecasting accuracy significantly.

Key words: Xijiang River basin, runoff forecast, non-stationary, combined forecasting models, extreme-point symmetric mode decomposition