Accurate runoff prediction is of great significance for agricultural irrigation, reservoir scheduling, flood control and disaster mitigation in the basin. Aiming at the strong nonlinearity and non-stationarity of the runoff series, a hybrid model for monthly runoff prediction, VMD-(CNN-LSTM, ELMAN), is proposed. Firstly, VMD is used to decompose the runoff sequence into multiple modal components, and the sample entropy (SE) of each modal component is calculated, according to which the components are divided into high-frequency and medium-low frequency components. Then the CNN-LSTM model is used for the prediction of high-frequency components, the ELMAN model for the medium-low frequency components. Finally the predictions results are summed up. The model is then applied to the monthly runoff prediction of Baimasi Station and Heishiguan Station in the middle and lower reaches of the Yellow River Basin, and the prediction results are evaluated compared with those of CNN-LSTM, ELMAN, VMD-CNN-LSTM models. Research results show that the NSE values of the prediction results of this model are all greater than 0.99, which is superior to other models, indicating that the VMD-(CNN-LSTM, ELMAN) model has high prediction accuracy and can be applied to actual monthly runoff prediction of the basin.