Abstract:
In the context of energy transition, anaerobic digestion offers a sustainable way to produce
biogas from organic waste. Yet, its biological complexity hinders real-time control, especially
due to unmeasurable variables like biomass. This work proposes an industrial supervisor for
monitoring and controlling an anaerobic digestion bioreactor, using the AM2 model. Three
observers are developed and compared: asymptotic, interval-based, and one using machine
learning (Random Forest). Results show each approach offers unique advantages in accuracy,
robustness, and uncertainty handling.
