The combination of empirical model decomposition and long short-term memory network(EMD-LSTM)model was used to forecast the hydrological time series data. The median filter for data preprocessing was used first and then EMD was applied to decompose the time sequence. The LSTM model was used to prediction for each characteristic series from EMD, and each prediction sequence was superposed to obtain the final prediction result. Based on the data of the instantaneous flow, water velocity and water level per hour of a certain river in the South-to-North Water Diversion Project, the EMD-LSTM was used for modeling. The experimental results show that this method can realize accurate prediction of water level, water velocity and instantaneous flow for 12 h and 6 h continuously, has higher accuracy compared with the LSTM model and provides decision basis for the forecast of hydrological time series and the real-time dispatching of water resources.