In response to the problems of low accuracy and poor reliability of water consumption prediction due to the strong randomness and non-stationary state exhibited by the water consumption signal, this paper proposed a hybrid water consumption prediction model based on improved EEMD-WOA-SRU. Firstly, the LSTM prediction method was used to suppress the endpoint effect of the EEMD to obtain the improved intrinsic mode functions (IMF). Then the whale optimization algorithm (WOA) was used to optimize the simple recurrent unit (SRU) and predicted each component. Finally, the final prediction results were obtained by accumulation. The experimental results show that the decomposition error of the EEMD is reduced by 0.94% on average; Compared with the SRU, the average absolute error of EEMD-WOASRU model prediction is reduced by 45.42%, the root mean square error is reduced by 50.43%, and the reliability is improved by 52.38%. It can provide a basis for water resources decision making.