Abstract: To improve the prediction accuracy of mine water inflow, a mine inflow prediction model (CEEMD_GRU) based on the combination of complementary ensemble empirical mode decomposition (CEEMD) and gated recurrent unit (GRU) neural network. First, the one-dimensional water inflow data was decomposed into several intrinsic mode function (IMF) components and a residual margin by CEEMD algorithm. The fluctuation characteristics of the water inflow data at different time scales were reflected by the intrinsic mode components while the trend characteristics of long-term changes of the data was reflected by the residual margin. Then, the GRU neural network model was established for each component, and the study of one-dimensional data was transformed into the study of the decomposed multidimensional sub-components so as to train and learn the time-series change rules of each component and make predictions. Finally, the predicted water inflow was obtained by combining the predicted results. By comparing CEEMD_GRU with back propagation (BP), SVM and GRU neural network, the results show that the RMSE based on CEEMD_GRU prediction model is reduced by 60.8% on average. This study provides a new idea for further improving the prediction accuracy of mine water inflow.