Application of hierarchical optimization of gates and units, and flow-classification water level control rules, and model predictive control in reservoir real-time optimal scheduling

Study region: This study used the Xiajiang Reservoir (China) as a case study. Study focus: This study focuses on the following issues in reservoir scheduling: low efficiency of gate-unit optimisation, difficulty in controlling end-period water levels, and difficulty in achieving real-time scheduling with uncertain incoming water. This study constructs a hierarchical real-time optimisation scheduling (ROS) model, proposes an end-period water level control strategy integrating flow-classification water-level control rules (FWCR), and introduces model predictive control (MPC) to address the uncertainty of inflow. The main contributions of this study are as follows: (1) Hierarchical optimisation strategy (HOS) decomposes high-dimensional optimisation into upper-level unit-virtual aggregated gate joint optimisation and lower-level multi-gate discharge optimisation; (2) FWCR ensures end-period water level safety; (3) Constructed an MPC framework with ROS module to achieve real-time scheduling under water uncertainty. New hydrological insights for the region: (1) The hierarchical optimisation scheduling model (HOSM) outperformed the traditional optimisation scheduling model (OSM), with the 20-year flood peak shaving rate increasing from 7.09 % to 9.62 %; (2) With FWCR, HOSM further raised peak shaving to 9.88 %, reduced gate operations (29−22), increased power generation by 22.6 %, and lowered end-period water level by 1.02 m; (3) Optimal MPC parameters (36–48 h prediction horizon, 2–3 h decision step size) achieved 10.81 % peak shaving; feedback correction module max water level by 0.08–0.15 m. The HOS-MPC system based on FWCR provides a reference for real-time scheduling under uncertain inflow conditions.