Reinforcement Learning Approaches for the Optimization of the Partial Oxidation Reaction of Methane
Marius Neumann, Stefan Palkovits
Abstract
Optimizing reactions in the chemical industry is one of the major challenges in the pursuit of economic and ecological sustainability. With ongoing research in this field, the amount of available data has greatly increased, which makes it suitable for machine learning approaches. In this paper, the application of reinforcement learning for finding optimal reaction conditions of the partial oxidation of methane (POX) is tested. Q-learning (QL) agents and deep deterministic policy gradient (DDPG) agents are trained to maximize H2 production by partial oxidation of methane in a simulated plug flow reactor. Although the QL agent showed promising results in a simplified environment, it was not able to achieve improvements in the simulation environment. A clear superiority of the DDPG agent was observed, as it was able to maximize H2 production by adjusting temperature, pressure, flow velocity, and substrate composition. This proves that reinforcement learning is applicable for reaction optimization and a promising concept to improve efficiency in chemical processes.