A Hybrid Data-Driven Intensified Mechanistic Modeling Framework for Accurate and Reliable Wastewater Treatment...

Computational modeling guides the wastewater treatment (WWT) design. However, nonlinearities, stochastic influent fluctuations, and complex microbial interactions challenge model predictive accuracy and robustness. Mechanistic models lack fidelity under dynamic conditions due to static parametrization. Data-driven models have emerged to capture such dynamics, but their unconstrained flexibility can produce unstable predictions, which are unacceptable in high-stakes WWT. This study proposes a hybrid data-driven intensified mechanistic modeling (HyDIM) framework, which functionalizes kinetic parameters using a data-driven component to enable adaptive responses, while the mechanistic core ensures predictive robustness. HyDIM was validated in the high-stakes sulfide-laden WWT system, where precise control is critical to mitigating hydrogen sulfide release and optimizing elemental sulfur recovery. Compared to the mechanistic model, HyDIM improved prediction accuracy under dynamic conditions, with R² values for sulfide, sulfur, and sulfate increasing from 0.30, 0.09, and 0.02 to 0.66, 0.77, and 0.74, respectively. Predictive robustness was also enhanced, with extreme deviations reduced by 78.2% relative to the data-driven model. Using HyDIM, a bi-objective optimization strategy achieved stable long-term performance, with 96% sulfide removal and 91% sulfur recovery, a 41% improvement in sulfur recovery over the mechanistic strategy. Overall, this work presents a novel modeling paradigm for reliable decision support and control in WWT.