Risk-Aware, Low-Carbon Operating Windows for Pharmaceutical Wastewater Treatment via Uncertainty-Aware Bayesian Optimization

Abstract

Pharmaceutical manufacturing effluents containing antibiotics and other micropollutants pose persistent challenges to environmental protection and process safety. Their treatment is increasingly constrained by high energy demand, chemical consumption and the associated carbon emissions. Here we compiled literature-sourced operational data from conventional activated sludge (CAS), membrane bioreactor (MBR) and ozonation systems treating pharmaceutical wastewater, and constructed Gaussian process regression (GPR) surrogate models for removal performance, antibiotic risk quotients, resource consumption and indirect greenhouse-gas emissions. To address the sparsity and heterogeneity of available data sets, we developed a data-augmentation-assisted Bayesian multiobjective optimization framework combining hybrid sampling with uncertainty-aware search to map trade-offs among effluent quality, risk reduction and carbon-intensive inputs. Compared with baseline conditions, optimized operating windows increased overall removal efficiency by approximately 10–15% while reducing energy consumption, chemical dosage and CO2-equivalent emissions by roughly 12, 15 and 8–9%, respectively. Ozonation showed the strongest risk-reduction capability, shifting compound-specific risk quotients below regulatory thresholds for several high-priority antibiotics, whereas CAS showed limited improvement. This workflow operationalises low-carbon strategies for pharmaceutical wastewater treatment and offers a transferable approach for other hazardous industrial effluents.