A hybrid variational mode decomposition and sparrow search algorithm-based least square support vector machine model for monthly runoff forecasting

Abstract

Monthly runoff forecasting has always been a key problem in water resources management. As a data-driven method, the least square support vector machine (LSSVM) method has been investigated by numerous studies in runoff forecasting. However, selecting appropriate parameters for LSSVM is the key to obtaining satisfactory model performance. In this study, we propose a hybrid model for monthly runoff forecasting, VMD-SSA-LSSVM for short, which combines variational mode decomposition (VMD) with LSSVM and the parameters of LSSVM are optimized by a sparrow search algorithm (SSA). Firstly, VMD is utilized to decompose the original time series data into several subsequences. Secondly, LSSVM is employed to simulate each subsequence, for which the parameters are optimized by SSA. Finally, the simulated results for each subsequence are accumulated as the final results. The validity of the proposed model was verified by forecasting monthly runoff for two reservoirs located in China. Four frequently-used statistical indexes, namely the Nash efficiency coefficient, root mean squared error, correlation coefficient and mean absolute percentage error were used to evaluate model performance. The results demonstrate the superiority of VMD-SSA-LSSVM over the compared models in terms of all statistical indexes, indicating that it is beneficial for enhancing monthly runoff forecast accuracy.