Dynamic compensation for flood and sediment regulation in the Yellow River using a digital twin

Study region This study examines the 128-km Xiaolangdi–Huayuankou reach of the middle and lower Yellow River, China. Reservoir releases affect Huayuankou after a travel-time lag and interact with sediment transport and channel-bed adjustment. Study focus We develop a digital twin-enhanced bi-level multi-agent framework in which the basin authority sets a time-varying sediment-flushing subsidy and the reservoir operator selects release and sediment-control responses. The hydrodynamic-sediment digital twin serves as a state-transition and feasibility operator within a nested real-coded genetic algorithm–particle swarm optimization procedure. The framework is evaluated using a 240-h reconstruction of the 2021 Autumn Flood under centralized coordination, decentralized operation without compensation, and decentralized operation with dynamic compensation. New hydrological insights for the region Downstream risk depends on reservoir-response timing relative to flood-wave propagation, not on inflow magnitude alone. Without compensation, Huayuankou water level peaks at 94.14 m and remains above the 93.56-m threshold for 49 h, while cumulative sediment evacuation is about 15,000 t. Dynamic compensation lowers the peak to 93.45 m and increases sediment evacuation to about 22,500 t, 50% above the no-compensation case and close to the centralized benchmark of 23,500 t. Local perturbation tests indicate that reoptimization moderates the tested inflow and sediment-load effects, whereas higher electricity prices increase required compensation. These findings apply only to the reconstructed event and tested ranges.