Cost reduction of CO2 capture processes using reinforcement learning based iterative design: A pilot-scale absorption-stripping system

An economical and practical way of designing the optimal condition for CO2 capture processes is proposed. This learning strategy, called reinforcement learning based iterative design, is developed to learn the optimal condition from hybrid information. One is from simulation data, the other, from real plant data. The simulation data is easily accessible, but the optimal condition is restricted to a simulated model being selected. To make up the mismatch, new info from the real process is explored. Only fewer operating data supplemented from the real process is used to update the learning scheme, so time, costs, and efforts can be saved. The fused info from the two kinds of data is also proposed. To demonstrate the effectiveness of the proposed method, design of a pilot-scale CO2 absorption-stripping experiment is conducted for recovery of CO2 from flue gases.