(562c) Reinforcement Learning Based Control for Biorefining Process Under Uncertainty

Excess waste is a major issue affecting global sustainable development and has led to an influx of pollution and greenhouse gas emissions. Biorefineries are an alternative pathway for waste that are able to convert low-cost biomass into value-adding biproducts. Here in, we present a reinforcement learning based framework to control the uncertainties associated with a centralized biorefinery plant. Anaerobic digestion, a well-known and utilized biorefinery process, is used to show the performance and robustness in production and uncertainty control. Our framework outperforms traditional control strategies in short-term scenarios, achieving fast target tracking, increased precision, and accuracy. When considering long term plant control, with inventory limitations, our framework shows adaptive and robust behavior satisfying downstream demands. We have shown that our reinforcement learning framework is a promising and scalable solution for uncertainty issues in real-world biorefining processes, contributing to the advancement of sustainability and waste management.