Runoff Forecasting of Machine Learning Model Based on Selective Ensemble

Reliable runoff forecasting plays an important role in water resource management. In this study, we propose a homogeneous selective ensemble forecasting framework based on modified differential evolution algorithm (MDE) to elucidate the complex nonlinear characteristics of hydrological time series. First, the same type of component learners was selected to form the average ensemble model, which was then trained using the training set to obtain preliminary prediction results. Subsequently, the MDE method was applied to improve the performance of the differential evolution algorithm with respect to low solution accuracy and premature convergence. MDE assigns weights according to the performance of each component learner in the ensemble model to obtain the selective ensemble model structure on the validation set. Finally, the selective ensemble framework was verified on the test set. Experiments were conducted on the runoff data of four important hydrological stations in the Yangtze River Basin. The results showed that the forecast framework can obtain better prediction accuracy and generalization performance than the average ensemble models composed of four classical learners, and can improve prediction accuracy for hydrological forecasting.