Deep Learning-Based Multiobjective Optimization for Balancing Effluent Quality, Operational Cost, and Greenhouse Gas Emissions in Wastewater Treatment Plant Control

The low-carbon and sustainable operation challenges the wastewater treatment plant (WWTP), notably as influent temperatures vary seasonally. Optimizing operational parameters is a feasible approach, but current methods generally face a large computational cost and imprecise optimization. In response, this study developed a novel seasonal multiobjective optimization method based on deep learning to trade-off effluent quality index (EQI), operational cost index (OCI), and greenhouse gas (GHG) emissions. The crucial control and operation variables were identified for both winter and summer using Sobol's sensitivity analysis, serving as optimization candidates and inputs for the data-driven models. Then, deep neural network models were constructed on a seasonal basis to approximate EQI, OCI, GHG emissions, and crucial effluent quality limitations. Furthermore, multiobjective optimization for winter and summer was performed based on the preference-inspired coevolutionary algorithm. The results show that the optimized scenarios reduce the level of the OCI by 23.92% and 40.94% and the level of GHG emissions by up to 7.72% and 13.91% in winter and summer, compared to the base cases. The novel optimization approach simplifies and improves performance trade-off for low-carbon WWTPs. It also facilitates online fine-tuning of WWTP operating parameters for different seasons, particularly in regions with significant temperature changes.