Online Control of the Raw Water System of a High-Sediment River Based on Deep Reinforcement Learning

Water supply systems that use rivers with high sedimentation levels may experience issues such as reservoir siltation. The suspended sediment concentration (SSC) of experiences interannual variation and nonlinearity due to its close relationship meteorological factors, which increase the mismatch between river water source urban demand. raw system scheduling problem is expressed a pump station control involves real-time SSC changes. To lower intake pumping station’s energy consumption, deep reinforcement learning (DRL) model based on prediction was developed. framework consists DRL model, hydraulic for simulating system, neural network predicting SSC. tested using data from Yellow River withdrawal in China an average capacity 400,000 m3/d. strategy created this study can reduce consumption per unit by 8.33% annual 37.01%, when compared manual strategy. Meanwhile, algorithm had good response robustness uncertain imperfect predictive data.