Rolling predictive optimal scheduling of reservoirs for flood control and power generation under prediction uncertainty

Abstract The Xiajiang Reservoir, located in the middle Ganjiang River of the Yangtze River Basin, requires scientific inflow prediction and optimised scheduling to balance the flood control and power generation. Traditional methods are constrained by an overreliance on historical data, insufficient flood samples, and poor high-flow performance, necessitating improved models. Focusing on the Xiajiang Reservoir, this study developed a multi-model series–parallel coupling inflow prediction model to test different coupling modes and a rolling prediction optimisation scheduling model integrated with the inflow model. Using typical design floods as input, we analysed the impacts of the flood forecast period and dynamic flood control water level on benefits and explored performance across flow scenarios. The key results showed that the parallel coupling framework achieved excellent short-term prediction, whereas the series coupling model had obvious limitations, and the series–parallel model underperformed. The rolling scheduling modelwith dynamically updated hydrological dataeffectively balances the two objectives. Within 72 h, the maximum outflow first decreased and then stabilised; moderately raising the dynamic flood control water level boosted power generation with minimal flood risk. Additionally, traditional forecasting struggled with high flows because of scarce samples, whereas abundant low-flow samples ensured higher accuracy and better scheduling results. This study strengthens the scientific basis for inflow prediction and scheduling in the Xiajiang Reservoir and provides a valuable reference for similar reservoirs in the Yangtze River Basin.