Development of artificial neural network model for anaerobic digestion-elutriated phase treatment

Abstract

Nonlinear autoregressive exogenous (NARX) neural network models were used to forecast the time-series profiles of anaerobic digestion-elutriated phase treatment (ADEPT). Experimental data from the operation of the pilot plant and lab-scale reactor were used for calibration, validation, and practice tests. Anaerobic digestion-elutriated phase treatment removed approximately 87% of volatile solids with a relatively brief hydraulic retention time of 7 days. The self-built machine learning algorithm provided confident predictions of the volatile-solids removal efficiency, biogas production, and methane content, with mean square error values of 0.32, 0.02, and 0.16, respectively. Time-series simulations of nonlinear autoregressive exogenous models demonstrated that ADEPT can improve organic removal and biogas production by maintaining the pH at 6.0–6.5 and 7.0–7.5 in the acidogenesis and methanogenic reactors, respectively. Applying nonlinear autoregressive exogenous neural network models to ADEPT allows high-rate anaerobic digestion without over-acidification. © 2025 Elsevier Ltd