(188y) Nonlinear System Identification and Dynamic Real-Time Optimization of Postcombustion CO2 Capture Processes for Cycling Applications

The climate change and its correlation with the greenhouse gases has increased the interest in the implementation of carbon capture technologies for existing fossil fueled power plants (FFPP) and the utilization of renewable energy sources^[1,2]. However, as the renewables are intermittent and not dispatchable^[3], the FFPP designed to operate at base-load conditions are required to cycle their load. Power plant cycling will introduce inefficiencies to the plant that can result in increased emissions and coal consumption^[2]. The objective in this work is to develop a framework for generating optimal dynamic trajectories for postcombustion methylethylamine (MEA)-based carbon capture systems in cycling FFPP due to renewables. To truly represent the system and mitigate the effect of disturbances, a multivariable and dynamic nonlinear system model is desired for dynamic real-time optimization (DRTO) applications. As a detailed first-principles model of the integrated plant is intractable for online optimization, a reduced-order model will be developed employing system identification techniques.

A simplified model enables the implementation of nonlinear programming (NLP) optimization techniques for high-dimensional systems in a computationally tractable manner^[4]. The proposed DRTO aims to minimize a single objective cost function that considers different carbon capture scenarios such as cap and trade of carbon credit for distinct carbon capture rates. The produced DRTO trajectories will be ultimately provided to MPC controllers^[5]to ensure the power plant can follow the optimal profiles for cycling. Based on the optimization and control objectives, the control structure is chosen for the model identification task. For the power plant application, the flue gas, steam and lean solvent flowrates are the main inputs of the system, whereas power consumption, lean solvent CO₂ loading and carbon capture rate are the main outputs.

For the system identification, different nonlinear autoregressive moving average with exogenous inputs (NARMAX) models are analyzed. The Akaike information criterion (AIC) is employed in the model selection, as it rewards goodness of fit and penalizes increased number of estimated parameters^[4]. Following this criterion, model complexity is increased until the AIC value changes significantly. If there is no significant difference in the AIC indicator for different models, then the simpler model is chosen. The objective in this selection is to obtain a model as simple as possible that can still provide an accurate representation of the system for optimization purposes. The designed input sequence is a pseudo random binary signal (PRBS) for all inputs. Additionally, for the flue gas flowrate, different ramping rates will be tested to represent the cycling of the FFPP. The developed model is then employed for DRTO purposes and optimal trajectories are generated for cycling. In this presentation, results for different cycling scenarios associated with different carbon capture landscapes will be discussed, aiming the optimal operation of future supercritical FFPP.

References

[1] ZHANG Q., TURTON R., & BHATTACHARYYA D. (2016). Development of Model and Model-Predictive Control of an MEA-Based Postcombustion CO₂ Capture Process. Industrial and Engineering Chemistry Research. 55, 1292-1308.

[2] BENTEK ENERGY LLC. (2010). How Less Became More: Wind, Power and Unintended Consequences in the Colorado Energy Market. Prepared for Independent Petroleum Association of Mountain States.

[3] TRIFKOVIC, M., MARVIN, W. A., DAOUTIDIS, P., & SHEIKHZADEH, M. (2014). Dynamic real-time optimization and control of a hybrid energy system. AIChE Journal. 60, 2546-2556.

[4] BHATTACHARYYA D., & RENGASWAMY R. (2009). Dynamic modeling and system identification of a tubular solid oxide fuel cell (TSOFC). Proceedings of the American Control Conference. 2672-2677.

[5] XIN, H., LIMA F. V., (2016) Design and implementation of model predictive control strategies for IGCC power plant cycling with carbon capture. AIChE Annual Meeting, San Francisco, CA.