Process Simulation and Optimization of Hydrogen Liquification Using Reinforcement Learning

Abstract

Abstract Hydrogen liquification is one of the main steps for using hydrogen in obtaining carbon neutralization. This process consists of cooling hydrogen to approximately 20 K and converting ortho-hydrogen to para-hydrogen. This paper presents a study in which the authors simulate the Linde-Hampson hydrogen liquification process using a commercial process simulator. Then, the process optimization is evaluated using a reinforcement learning method. In the first step, the Linde-Hampson process is simulated in a steady-state mode using a commercial process simulator. The process uses liquified natural gas (LNG) as a pre-cooling medium. Process optimization is conducted in the next step to minimize the energy consumption of the hydrogen liquification process. A reinforcement learning (RL) method, using a deep deterministic policy gradient algorithm (DDPG), is implemented for process optimization. The reward for the RL optimizer is defined to minimize the operating cost for hydrogen liquification according to the cost and rate of consumption of electricity and LNG. An Artificial Neural Network (ANN) is used to construct the Actor-Critic sections of the RL algorithm. Capability of the RL algorithm is evaluated versus a classic optimization method. Case studies at different electricity and LNG prices were conducted to evaluate the trained RL model versus the base process optimized by the process simulator. The comparison between hydrogen liquification operating cost of the trained model with base model showed improved performance for the RL model.