Abstract: Influenced by such factors as meteorology, climate change and human activities, the runoff series present unsteady and nonlinear characteristics, which brings a new challenges to the accurate prediction of runoff. In order to improve the prediction accuracy of a single runoff prediction model, a new combined runoff prediction model（EEMD-VMD-SSA-KELM） based on ensemble empirical mode decomposition（EEMD） algorithm, variational modal decomposition（VMD） algorithm and kernel extreme learning machine（KELM） which is optimized by sparrow search algorithm（SSA） is proposed. Firstly, EEMD algorithm is used to decompose the runoff sequence into trend component, detail component and random component, and then VMD algorithm is used to further decompose the random component with the highest frequency into several components with different frequencies that are more stable than the random component to reduce the instability of the runoff sequence. Secondly, the KELM model is established for each component to predict, and SSA is used to optimize the kernel parameters and penalty coefficients of the KELM model. Finally, the prediction results of runoff series are obtained by accumulating the prediction results of all components. The proposed model is applied to the daily runoff prediction in the flood season of Cuntan Hydrological Station in Yichang, Hubei. The proposed model is applied to the flood season daily runoff prediction of Cuntan Hydrometric Station in Yichang, Hubei Province, and is compared with BP neural network model, Least Squares Support Vector Machine Model（LSSVM）, KELM model, etc. The results show that the prediction accuracy of the model combined with data decomposition algorithm is obviously better than that of the single BP model, LSSVM model and KELM model, and the prediction accuracy of the model combined with EEMD algorithm and VMD algorithm was better than that of the model combined with EEMD algorithm only. The prediction accuracy of KELM model is better than that of LSSVM model; The optimization precision of SSA was better than that of particle swarm optimization（PSO） algorithm. The prediction accuracy of EEMD-VMD-SSA-KELM model is the highest, which can accurately simulate the change trend of daily runoff in flood season with complex multi frequency information, and can provide a reference for hydrological forecasting and related forecasting research.