(596c) An Integrated Approach for Real-Time Economic Optimization, State Estimation, and Control for a Post-Combustion Carbon Capture Absorber Section Under Uncertainty

Model-based optimization and control provide plant operators with systematic methods to ensure the safety, productivity, and cost effectiveness of a process despite unexpected disturbances, diurnal cycles, and seasonal process changes. A common approach to this is the so called two-layer structure in which a steady-state economic optimization problem is solved at a slow timescale to generate operating set points for a dynamic optimal control problem that is executed at a faster timescale. While the top optimization layer generates cost-effective operating points, the bottom control layer ensures that these points are actually reached by the closed-loop system. Although this integrated approach adds complexity into the hierarchical manufacturing framework, it has been shown to return significant improvements in the process economics. In particular, such economic improvements are critical in the energy sector where prices must be kept low as the product is a public utility that must be widely accessible.

With the increased effect of global warming owing to the release of anthropogenic greenhouse gas emissions from industry, the energy sector is under scrutiny for its largely emissive operations. In particular, combustion emissions remain prevalent as much of the world is still reliant on carbon-based fuels for energy production [1]. To this end, more attention is being placed on emission mitigation strategies. A category of technologies poised to address the emissions of these carbon-based fuels is known as carbon capture and storage (CCS). Within this wider umbrella of technologies, solvent-based post-combustion CO₂ absorption is arguably the most mature method of CO₂ removal. In this technology, the crucial unit for removal is the absorber column, for which a mechanistic dynamic model has previously been developed [2]. This model consists of a large-scale partial differential algebraic system (PDAE) of equations, which, although more difficult to implement that simpler models, provides an accurate treatment of the phenomena occurring inside the column. This level of detail allows for a hierarchical implementation of a real-time optimization (RTO) and nonlinear model predictive control (NMPC) structure that uses the same model. This operational two-layer structure, as described above, has been shown to result in cost-effective operation of complex systems, such as the absorber column. While the economically optimal operation of the solvent-based CO₂ capture system has been addressed [3-5], the two-layer structure has yet to be implemented and tested comprehensively. Furthermore, the key issues of process uncertainty and state estimation in this system have not been addressed comprehensively for this system.

In this work, we present an implementation of the aforementioned two-layer optimization and control structure for the solvent-based CO₂ absorber downstream from a coal-based power plant subject to uncertainty. In the economic optimization layer, a cost function has been built featuring energy costs, carbon tax costs, and solvent degradation cost of the process. In the control layer, the performance of an NMPC and a linear MPC is contrasted. State estimation is also addressed in this study. An important aspect of the model used in the two layers is its axially discretization, which models the substantial variation in states along the length of the column and is crucial to the accuracy of the model. However, due to the axially discretization of the PDAE system, the model requires information on all of the absorber states at all axial discretization points (110 process state variables). Hence, a moving horizon estimator (MHE) is implemented whereby only a small portion of the total states needed for the model are assumed to be available for online measurement. Moreover, uncertainties are introduced at the different operational levels via some key parameters (i.e. prices, inlet compositions, and thermodynamic coefficients) and the inherent robustness of the nonlinear model to these uncertainties is assessed. The effect of some of these uncertainties has been previously studied [6,7] and shown to have a substantial effect on the absorber.

The performance of the linear and nonlinear MPCs are first analyzed in a single-layer implementation (i.e. no economic optimization) subject to a disturbance rejection (load-following) scenario with measurement and process noise. This will illustrate the performance difference between the control-layer models. Moreover, the entire proposed two-layer structure is subjected to fluctuations in carbon price and parametric uncertainty in the process model in a repeated set point update scenario; this aims to simulate the substantial variation to which the absorber could be subjected to in day-to-day operations. This will elucidate the inherent robustness in the NMPC owing to its highly detailed model relative to the linear model, as well as show the performance improvement afforded by the RTO layer. The trade-offs between the two control-layer models will also be discussed.

This study presents the first comparison of linear and nonlinear control-layer model for the CO₂ absorber subject to price and parametric uncertainty. Moreover, it also presents the first combined optimization/control/estimation structure for the post-combustion CO₂ capture plant. These novelties present an increasingly holistic picture of how to approach the control and optimization of the absorber unit, which remains in its industrial infancy, thereby advancing its development towards eventual industrial-scale deployment.

References

[1] International Energy Agency (IEA), “Coal Information 2018”, Coal Information, 2018. Available: https://doi.org/10.1787/coal-2018-en.

[2] N. Harun, T. Nittaya, P. Douglas, E. Croiset and L. Ricardez-Sandoval, "Dynamic simulation of MEA absorption process for CO2 capture from power plants", International Journal of Greenhouse Gas Control, vol. 10, pp. 295-309, 2012. Available: https://doi.org/10.1016/j.ijggc.2012.06.017.

[3] Chan, L. and Chen, J., “Improving the energy cost of an absorber-stripper CO2 capture process through economic model predictive control”, International Journal of Greenhouse Gas Control, 76, pp.158-166, 2018. Available: https://doi.org/10.1016/j.ifacol.2018.09.284.

[4] Decardi-Nelson, B., Liu, S. and Liu, J., “Improving Flexibility and Energy Efficiency of Post-Combustion CO2 Capture Plants Using Economic Model Predictive Control”, Processes, 6(9), p.135, 2018. Available: https://doi.org/10.3390/pr6090135.

[5] Panahi, M. and Skogestad, S., “Economically efficient operation of CO2 capturing process. Part II. Design of control layer”, Chemical Engineering and Processing: Process Intensification, 52, pp.112-124, 2012. Available: https://doi.org/10.1016/j.cep.2011.11.004.

[6] G. Patron and L. Ricardez-Sandoval, "A robust nonlinear model predictive controller for a post-combustion CO2 capture absorber unit", Fuel, vol. 265, p. 116932, 2020. Available: https://doi.org/10.1016/j.fuel.2019.116932.

[7] Cerrillo-Briones, I. and Ricardez-Sandoval, L., 2019. Robust optimization of a post-combustion CO2 capture absorber column under process uncertainty. Chemical Engineering Research and Design, 144, pp.386-396.