Reinforcement Learning Method for Multi-objective Optimization of Papermaking Wastewater Treatment Process.

Due to different sources and the water using habits, the wastewater of papermaking fluctuates sharply over time. Quality control of the effluent in the papermaking wastewater treatment process (PWTP) is rather challenging and costly. Concerns are growing about the effects, especially of the greenhouse gas (GHG) emission, of PWTP on the environments. In order to realize the economic, energy-efficient, and eco-friendly objectives, with subject to the effluent quality, this paper proposed a multi-agent deep reinforcement learning framework to simultaneously optimize process cost, energy consumption, and GHG emission in the PWTP. Benchmark simulation model No.1 (BSM1) was used to simulate biological treatment process of papermaking wastewater. The default wastewater influence data of BSM manual was used to train, and the real data collected from papermaking industry was applied to verified the model system. The results demonstrate that the proposed system effectively identifies optimal control strategies for multiple targets, outperforming traditional methods.