Optimization of river environmental management based on reinforcement learning algorithm: a case study of the Yellow River in China

Generating scientific management strategy contributes to the sustainable development of river ecological environment. In this study, a multi-objective coupled water and sediment regulation model aiming at minimizing sedimentation and inundation loss as well as maximizing ecological value in the lower Yellow River has been developed. A reinforcement Q-learning algorithm was used to obtain optimized strategies from the multi-objective of sediment reduction, flood control and ecological restoration under different hydrological years. The results showed that the simulated channel sedimentation is very close to the measured value, which proves the applicability of the developed model. Under dry, normal and wet hydrological year, the effects of various regulation strategies on silt reduction, flood control and ecological restoration were obviously different. The regulation scheme of discharge at 3700 m³/s was verified to be suitable for dry and wet year, and that of discharge at 2600 m³/s was more suitable for normal year. Increasing the spacing of the beach area was better in normal year and wet year. Our findings suggested optimized strategies to address environmental challenges of the lower Yellow River in different hydrological years. This paper provides a reliable reference for improving the management of the lower Yellow River.