A Reinforcement Learning Approach to Optimize CO2 Removal from Flue Gas in Carbon Capture Applications

Abstract

Abstract Reinforcement learning (RL) has gained interest in chemical process and production industries for control and optimization of the process. The RL algorithm does not need any kind of field knowledge to start the optimization process. The algorithm will be trained and adjusted during the learning and training steps. This paper presents an evaluation of the RL approach in the context of CO2 removal from flue gas for carbon capture applications. A carbon capture plant is simulated using a commercial process simulator to be served as the RL algorithm’s environment. Observed values and control parameters required for plant optimization are defined and fed into the algorithm. These parameters were used for reward, policy, and state definitions. The agent in the RL model uses a deep deterministic policy gradients (DDPG) algorithm for process optimization. Two neural networks were used as the critic and actor for policy generation. The optimization target is the minimization of the required operating energy while removing the optimal amount of CO2 from the flue gas. The RL algorithm was used to train the critic and actor networks toward the optimum policy for operating the carbon capture plant. The result of the project is an assessment of the capture plant response to the changes in CO2 composition and inlet gas flow rate. The goal of the plant operation is to minimize the required operating energy while removing the optimal amount of CO2 from the flue gas. The results of the trained RL algorithm have been evaluated under different operating conditions. The required energy was compared with the base case; i.e., the case with recommended operating conditions by the simulator vendor. The results showed that trained RL algorithm can operate the capture plant with lower energy costs in more than 70% of the cases. The paper presents an evaluation of the RL method for CO2 capture process optimizations in digital transforming and implementing artificial intelligence (AI) in oil and gas industries.