Abstract: It’s known that the Three Gorges Reservoir and Gezhouba Hydropower Station have close hydraulic connection. However, there is an inconsistency between the Three Gorges Reservoir’s outflow and Gezhouba Hydropower Station’s inflow by operational monitoring, which brings uncertainty to the inflow and power output prediction of Gezhouba Hydropower Station. In order to solve these problems, a prediction model was proposed for Gezhouba Hydropower Station’s inflow based on eXtreme Gradient Boosting (XGBoost) and Autoregressive Integrated Moving Average model (ARIMA) during the non-discarded water period. What’s more, another prediction model was proposed for daily power output of Gezhouba Hydropower Station based on Bayesian Ridge Regression. And then, these two models were combined to predict the daily power output with unknown reservoir inflow. Through the experimental analysis during the non-discarded water period in 2019, the results showed that the proposed inflow prediction model performed better than traditional three-day average conversion coefficient method, greatly reducing the prediction error. Further more, the daily power output prediction model has high precision accuracy and strong noise robusticity, and it can be applied to making power-generation plan of Gezhouba Hydropower Station during non-discarded water period.