Model predictive control of anaerobic digestion processes using a long short-term memory network predictor

This article presents an evaluation of a model predictive control (MPC) strategy that employs long short-term memory (LSTM) networks as internal predictive models for anaerobic digestion (AD) processes. The primary objective was to develop and validate a data-driven control approach utilizing readily available online measurements. The strategy was tested in two simulated AD environments: the simplified Anaerobic Model No. 2 (AM2) and the detailed Anaerobic Digestion Model No. 1 (ADM1). LSTM networks were effectively trained to predict methane flow rates from simulated data, including stochastic disturbances. The integrated LSTM-MPC framework demonstrated robust methane flow rate setpoint tracking and ensured process stability in both environments, even amidst nonlinear operating conditions and influent disturbances. Importantly, the computational requirements remained feasible for real-time applications in these typically slow processes. The findings suggest that the LSTM-MPC strategy is a promising and computationally efficient alternative for controlling AD processes, providing a practical solution compared with traditional mechanistic model-based approaches that often rely on more complex and less accessible measurements.