A multi-agent AI reinforcement-based digital multi-solution for optimal operation of a full-scale wastewater treatment plant under various influent conditions

Optimal operating systems have been widely employed to improve the economic and environmental performance of wastewater treatment plants (WWTPs) for efficiently treating discharged water pollutants. However, the multi-optimization approach applied to WWTPs has been inadequate, resulting in a low optimization performance owing to varying influent conditions and inherent complex interactions between manipulated variables. Therefore, this study developed a digital multi-solution for optimal operation of a WWTP based on multi-agent reinforcement learning. Dynamic influent conditions with low, normal, and high influent chemical oxygen demands and total nitrogen composition ratios were generated using a k-means clustering algorithm to accurately reflect operating conditions. A game abstraction method based on a two-stage attention network (G2ANet) algorithm was employed to simultaneously search for three operational setpoints: dissolved oxygen, external sludge recycling, and external carbon dose. The adaptability of the G2ANet-based digital multi-solution for determining setpoints was evaluated by employing newly measured one-year influent data. The results demonstrated the ability of an intelligent G2ANet-based digital multi-solution in identifying optimal setpoints to improve the performance of WWTPs and outperform manual operating systems under varying influent conditions. It reduced aeration energy by 25 % and improved effluent quality by 7 %, while maintaining a similar pumping energy. The G2ANet-based digital multi-solution can innovatively contribute to the smart operation of WWTP operation systems owing to its environmental stability and adaptability.