Flexible operation of Post-combustion CO2 capture process enabled by NARX-MPC using neural network

Post-combustion CO2 capture (PCC) process is expected to serve as the role of a bridge solution during the transition from fossil-based to renewable energy, but extensive energy consumption is a significant hurdle to its widespread commercial adoption. Operating a PCC process economically often requires large operating condition changes in response to power plant load changes and time-varying energy prices. The 'flexible' operation re-quires a controller that aids in such a dynamic operating environment of the PCC process. To balance model complexity with prediction accuracy, we adopt NARX-MPC which employs a neural network-based NARX model as a prediction model for MPC. Under various dynamic operation scenarios, the proposed NARX-MPC is eval-uated and compared with a linear MPC (LMPC) controller. Our observations indeed reveal that the NARX-MPC delivers significantly superior closed-loop control performance over the LMPC, especially in scenarios involving large changes in the power plant load or inlet flue gas flow rate and simultaneous changes in the flue gas flow rate and the CO2 capture rate.