(274c) Application of Dynamic Reduced-Order Modeling and Advanced Process Control on UKy-CAER CO2 Capture Pilot Plant Using CCSI Tools

Solvent-based processes have remained the predominant solution to post-combustion CO₂ capture for power plants. Such processes often suffer sub-optimal transient performance due to process delays involving large solvent holdups, unmeasured disturbances from solvent degradation, limited availability of measured variables, and fluctuations in ambient conditions. In this work, the 2MWth slipstream-based CO₂ capture system designed and operated by University of Kentucky’s Center for Applied Energy Research (CAER) at the E.W. Brown generating station is utilized as a test-bench for developing and implementing Advanced Process Control (APC) approaches to improve transient performance and reduce energy penalty. This pilot plant involves two highly-interacting recirculation loops – a solvent loop and a desiccant loop – which result in significantly large time delays and oscillatory behavior during transient operations and serve as the major motivation for the proposed process control improvements. This work, as part of the U.S. Department of Energy’s Carbon Capture Simulation for Industry Impact (CCSI²) project, utilizes a suite of tools and models from the open-source Carbon Capture Simulation Initiative (CCSI) Toolset for the development of control strategies to efficiently reject process disturbances and optimize the set points within this industrial CO₂ capture process.

The CCSI’s APC Framework implements Nonlinear Model Predictive Control (NMPC) and utilizes computationally efficient dynamic reduced models (D-RMs) as “fast” yet accurate predictive models. Such models are generated from the transient data measured at CAER’s pilot plant by using the open-source CCSI tool D-RM Builder, which is a data-driven nonlinear system identification tool (Ma et al., 2016). Preliminary studies were conducted following certain design-of-experiments approaches and multiple sets of transient data were collected. These data were thereafter used for offline sensitivity and controllability analysis and more importantly towards developing dynamic reduced order predictive models. The developed D-RMs, with limited number of “most-influencial” input and output variables, demonstrated good prediction of plant behavior during a set of validation studies. These D-RMs were used for offline simulation-based APC studies utilizing APC Framework (Omell et al., 2016; Mahapatra et al., 2018) and suggested the potential to reduce the settling time by 60% while incurring 5% reduction in utility costs. This paper will present results from preliminary simulation-based studies utilizing the CCSI tools and discuss the development of a data-driven communication interface between APC Framework and pilot plant’s distributed control system (DCS) as part of our on-going research.

Reference

1. Ma, J., Mahapatra, P., Zitney, S. E., Biegler, L. T. & Miller, D. C. (2016). D-RM Builder: A software tool for generating fast and accurate nonlinear dynamic reduced models from high-fidelity models. Computer and Chemical Engineering, 94, 60-74.
2. Omell, B. P., Ma, J., Mahapatra, P., Yu, M., Lee, A., Bhattacharyya, D., Zitney, S. E., Biegler, L. T. & Miller, D. C. (2016). Advanced Modeling and Control of a Solid Sorbent-Based CO2 Capture Process. IFAC-PapersOnLine, 49(7), 633-638.
3. Mahapatra, P., Ma, J. & Zitney, S. E. (2018). Nonlinear Model Predictive Control using Decoupled A-B Net Formulation for Carbon Capture Systems – Comparisons with Algorithmic Differentiation Approach. American Control Conference, Jun 27-29, Milwaukee, WI.