Adaptive Subspace System Identification for CO₂ Capture Processes

The CO₂ capture process with amine solvent based absorption and stripping is the most significant industrial method for the removal of carbon dioxide from coal and natural gas-fired power plants. Several dynamic models have been developed for the CO₂ capture process. However, most of these models require reaction kinetic and thermodynamic properties of the species considered in the process. Therefore, to obtain the simulation results, rigorous property calculations are programmed in specialized software that runs in off-line hardware. Such a modeling implementation approach is useful for model based control design and optimization strategies. Nevertheless, the application of such type of models during process operations is limited because simulation convergence, time and computational resources are limited in industrial setups. A practical approach to obtain a dynamic model from a process plant is to use data-driven empirical models, where the model is made to match the process measurements. Among empirical models, the subspace system identification algorithms have been well accepted in industry not only because of their simplicity and robustness, but also because they provide state space form models that are very convenient for prediction, process monitoring and model based control. However, CO₂ capture process operating changes are subject to fluctuations in electric power consumption as supplementary fuel is burned during peak hours of electric energy usage. Therefore, empirical models should adapt accordingly to different process loads, including startup and shutdown procedures that are particularly common in CO₂ capture operations. This work applies adaptive subspace system identification to a CO₂ recovery pilot plant. The empirical model is updated by collecting data at every sampling time from more than seventy sensors and making typical operational changes to perturb the process. Such changes are expected to excite the different modes of operation to determine the number of states required to describe the dynamic response of the process, which can change depending of the process operating mode. The development of such an adaptive empirical model represents a valuable tool for advanced process control applications, where minimal energy consumption and optimal CO₂ recovery should be achieved at different operating circumstances.