Reinforcement Learning Approaches for the Optimization of thePartial Oxidation Reaction of Methane

Optimizing reactions in the chemical industry is one of the majorchallenges in the pursuit of economic and ecological sustainability. With ongoingresearch in thisfield, the amount of available data has greatly increased, whichmakes it suitable for machine learning approaches. In this paper, the application ofreinforcement learning forfinding optimal reaction conditions of the partialoxidation of methane (POX) is tested. Q-learning (QL) agents and deepdeterministic policy gradient (DDPG) agents are trained to maximize H2production by partial oxidation of methane in a simulated plugflow reactor.Although the QL agent showed promising results in a simplified environment, it was not able to achieve improvements in thesimulation environment. A clear superiority of the DDPG agent was observed, as it was able to maximize H2production by adjustingtemperature, pressure,flow velocity, and substrate composition. This proves that reinforcement learning is applicable for reactionoptimization and a promising concept to improve efficiency in chemical processes.

Automated summary

This study tests the application of reinforcement learning in finding optimal reaction conditions, and found that the deep deterministic policy gradient (DDPG) agent showed clear superiority in a simulated environment, maximizing H2 production. It demonstrates the potential of reinforcement learning in optimizing efficiency in chemical processes.