Abstract: Objective: The article proposes optimizing state-space models applied to activated sludge processes in wastewater treatment plants, addressing the limitations of manual or deterministic calibration. Method: It integrates Bayesian optimization for hyperparameter tuning with Bayesian inference through Hamiltonian Monte Carlo sampling using the NUTS algorithm. The analysis used 2,500 simulated records involving substrate and nitrogen inputs, logarithmic transformations, and interaction terms. Results: The model explained 59.4% of the variability in substrate removal efficiency and 32.9% of the variability in nitrogen efficiency, with RMSE values of 14.69 and 4.41, respectively. Estimated parameters, including yield Y=0.6 and decay rate b=0.05, were consistent with reported literature values. Conclusion: The proposed framework improves calibration stability, reproducibility, and uncertainty quantification in activated sludge modeling. Nevertheless, the study recognizes the need to include toxic, microbiological, and hierarchical factors to strengthen nitrogen prediction and support more reliable operational and regulatory decisions in future complex dynamic environmental treatment systems.

Keywords: Activated sludge, Bayesian optimization, Bayesian inference, State-space models, Calibration, Wastewater treatment, Uncertainty

Pages: 736-747