(428a) Nonlinear Predictive Control of an Industrial Selective Catalytic Reduction Unit with Time-Varying Time Delay

Under load-following operation of coal-fired power plants, it is challenging to control nitrogen oxide (NO_x) concentration at the outlet of selective catalytic reduction (SCR) units to comply with environmental regulations. For controlling NO_x, excess ammonia (NH₃) that often far exceeds the stoichiometric requirement is injected in SCRs leading to NH₃ slip that gets released to the atmosphere and increases plant operating costs. In SCR units, NH₃ adsorbs on the catalyst reacting with NOx in the gas phase; it is practically impossible to measure NH₃ loading on the catalyst. Thus, even when NH₃ loading is depleting or increasing, its effect may not be reflected in the real-time outlet NO_x concentration. This leads to a complex, time-varying, time-delay response that is difficult to estimate due to complex adsorption-desorption dynamics of NH₃ on the catalyst coupled with internal and external mass transfer limitations. Thus, the industry-standard feedback-augmented feedforward (FBAFF) controllers and conventional model predicative controllers exhibit poor control performance for SCR control.

In this work, first a comprehensive first-principles dynamic model of the SCR unit is developed. In the existing literature, mass transfer limitations, both internal or external, have been neglected [1]–[3]. In addition, the competing NH₃ oxidation reaction, that can consume available NH₃, particularly when the system experiences thermal disturbances, is often neglected [4]. Additionally, most of these models, including those cited previously, are validated only with lab- or bench-scale data, thus there is a distinct lack of industrial-scale validation for SCR unit models. Data-driven models that have been developed using industrial data lack the predictive capability of first-principles models [5]–[7]. In this work, a first-principles model was developed with consideration for the adsorption-desorption dynamics of NH₃ on the catalyst, the SCR reduction reaction, and the competing NH₃ oxidation reaction, as well as the internal and external mass transfer limitations inherent to the SCR system. Furthermore, the model was validated with industrial data for a wide range of load-following operating conditions to ensure the results were applicable for an industrial-sized SCR unit.

In the power plants, control of the SCR is largely handled via FBAFF control using feedforward models of varying complexity coupled with a feedback trim to control NO_x emissions via NH₃ injection [8]. In the literature, some applications of model predictive control (MPC) have been reported with the goals of limiting both NO_x emissions and NH₃ slip, but considering very simple plant models that lack sufficient details to capture the complex SCR dynamics [9]–[11]. A few papers have considered limiting ammonia slip using a state constraint in their MPC formulations, but these formulations do not necessarily minimize NH₃ slip [12], [13].

In this work, three nonlinear MPC (NMPC) formulation are developed. In the first formulation, tight control of the outlet NO_x concentration with the standard quadratic objective function is exhibited. A second NMPC is proposed using a soft constraint on the outlet NO_x while minimizing the amount of NH₃ slip. Finally, a third NMPC implementation is proposed with explicit accounting for the time delay due to the adsorption-desorption dynamics. Because the time delay is unknown, a state-observer based approach is used whereby an initial estimate of the time delay based on the kinetic parameters of the SCR reactions is updated recursively.

Performance of the NMPC controllers is studied with respect to industrial data for a load-following scenario with comparison drawn against a well-tuned FBAFF controller. Results are discussed from the perspective of controller performance via the standard metrics along with economic considerations in the form of reagent consumption. The study shows that when time-delay is estimated recursively, it leads to considerably superior control performance compared to other approaches especially for rapid changes in the incoming NOx flow and concentration.

[1] I. Nova, L. Lietti, E. Tronconi, and P. Forzatti, “Dynamics of SCR reaction over a TiO 2-supported vanadia-tungsta commercial catalyst,” Catal. Today, vol. 60, pp. 73–82, 2000.

[2] L. Lietti, I. Nova, S. Camurri, E. Tronconi, and P. Fonatti, “Dynamics of the SCR-DeNO, Reaction by the Transient-Response Method,” AIChE J., vol. 43, no. 10, pp. 2559–2570, 1997.

[3] T. Qin, T. Yang, Y. Lv, F. Hong, and Q. Yao, “Dynamic modeling for SCR system of coal fired power plant,” in Chinese Control Conference, CCC, Jul. 2016, vol. 2016-Augus, pp. 2006–2010, doi: 10.1109/ChiCC.2016.7553660.

[4] E. Muñoz, P. Marín, S. Ordóñez, and F. V. Díez, “The role of reaction kinetics and mass transfer in the selective catalytic reduction of NO with NH3 in monolithic reactors,” J. Chem. Technol. Biotechnol., vol. 90, no. 7, pp. 1299–1307, Jul. 2015, doi: 10.1002/jctb.4437.

[5] Y. Lv, J. Liu, T. Yang, and D. Zeng, “A novel least squares support vector machine ensemble model for NOx emission prediction of a coal-fired boiler,” Energy, vol. 55, pp. 319–329, Jun. 2013, doi: 10.1016/j.energy.2013.02.062.

[6] S. M. Safdarnejad, J. F. Tuttle, and K. M. Powell, “Dynamic modeling and optimization of a coal-fired utility boiler to forecast and minimize NOx and CO emissions simultaneously,” Comput. Chem. Eng., vol. 124, pp. 62–79, May 2019, doi: 10.1016/j.compchemeng.2019.02.001.

[7] Y. Lv, X. Lv, F. Fang, T. Yang, and C. E. Romero, “Adaptive selective catalytic reduction model development using typical operating data in coal-fired power plants,” Energy, vol. 192, Feb. 2020, doi: 10.1016/j.energy.2019.116589.

[8] T. Leopold, “Tuning Ammonia Flow to Optimize SCR Performance,” Power, Jan. 12, 2010.

[9] J. Zhang, B. Jia, B. Zou, X. Tian, and C. Liu, “A multiple model control method of coal-fired power plant SCR-DeNOx system,” in Communications in Computer and Information Science, 2018, vol. 925, pp. 306–313, doi: 10.1007/978-981-13-2381-2_29.

[10] K. Zhang, J. Zhao, Y. Zhu, and Z. Xu, “Model predictive control case study: Selective catalytic reduction (SCR) system in coal-fired power plant,” in Chinese Control Conference, CCC, Aug. 2016, vol. 2016-August, pp. 4300–4305, doi: 10.1109/ChiCC.2016.7554021.

[11] B. Shen, N. Zhao, T. Liu, F. Wu, and C. Zuo, “Modeling and simulation of selective catalytic reduction for flue gas denitration in power plants,” in Advanced Materials Research, 2012, vol. 383–390, pp. 6580–6586, doi: 10.4028/www.scientific.net/AMR.383-390.6580.

[12] S. Shah, S. Abrol, S. Balram, and J. Barve, “Optimal Ammonia Injection for Emissions Control in Power Plants,” IFAC-PapersOnLine, vol. 48, no. 30, pp. 379–384, 2015, doi: 10.1016/j.ifacol.2015.12.408.

[13] H. Peng, H. Shioya, and R. Zou, “A predictive control strategy for nonlinear nox decomposition process in thermal power plants,” IEEE Trans. Syst. Man, Cybern. Part A Syst. Humans, vol. 36, no. 5, pp. 904–921, 2006, doi: 10.1109/TSMCA.2005.855920.