In order to improve the prediction accuracy and computational efficiency of the hydrological model and to solve the problem of difficulty in extracting multi-source hydrological features by traditional methods, a monthly runoff prediction model based on temporal convolutional network (TCN) and long and short term memory (LSTM) with bimodal decomposition is proposed. The complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) generates high-frequency eigenmode functions (IMFs), which have an impact on the prediction results. To reduce this effect, a variable mode decomposition (VMD) is performed on the high-frequency IMF components after the primary mode decomposition. A CEEMDAN-VMD-TCN-LSTM prediction model was established, and the monthly runoff prediction was completed with an LSTM network after extracting features from the decomposed runoff series by TCN. The stability and accuracy of the model were verified by taking the monthly runoff from 1957-2021 at Wulong Station in the Wujiang River basin as an example. It was found that the coupled CEEMDAN-VMD-TCN-LSTM model outperformed other comparative models in terms of computational efficiency, accuracy and network structure, with a prediction accuracy of 95%. The effectiveness of the model in hydrological prediction in the Wujiang River basin is verified.