(689c) Estimation-Based Model Predictive Control with State-Dependent Objective Prioritization; An Application to a Natural Gas Combined Cycle Power Plant

For control of systems where state, input, and/or controlled variables are unmeasured or are measured, but are corrupted with high noise, an estimator-based model predictive control (EB-MPC) can result in superior performance when the estimated variables are of critical importance to the objective or constraint of the MPC. This work will present the development of an EB-MPC algorithm for a system where some of the critical disturbance inputs and constraints for the MPC are unmeasurable.

This work is focused on the control of a system where objectives are prioritized based on the operating state of the plant. While multi-objective MPC (MO-MPC) considers multiple control objectives, the control objectives considered in MO-MPC are competing and therefore a compromise must be made at every instant in terms of the performance of each of the objectives. This work focuses on the control problem where there are multiple control objectives for the MPC, but they are mutually exclusive and priority of the MPC control objective changes based on the operating state of the plant. Furthermore, new manipulated or disturbance variables are added or some of the existing manipulated variables are excluded conditioned on the objective selection. The resulting MPC problem can lead to discontinuity in the optimization problem resulting in difficulty in solving the MPC optimization problem and unacceptable and poor control performance. While there are several papers that can be found in the literature on MO-MPC [1-5], to the best of our knowledge, there is currently no work available in the open literature on MPC with state-dependent objective prioritization.

A generic algorithm is developed for the MPC based on the user-provided criterion/criteria for selection of the objective function. The algorithm can accommodate any number of objectives and any number of conditions for selection/exclusion of objective functions. It yields smooth transition from one objective function to another readily accommodating structural changes in the MPC, as desired, conditioned on the selected objective function thus offering favorable properties for the Jacobian and Hessian for the underlying optimization problem. The structure is flexible and can be used for simultaneous selection of multiple objectives, if desired, as opposed to a single objective based on the state of the system. The EB-MPC algorithm with state-dependent objective prioritization is developed with due consideration of synchronization errors between estimator and MPC, model mismatch, measurement noise, and prediction uncertainty for the estimated variables.

The algorithm is applied to a superheater system of a natural gas combined cycle (NGCC) power plants where a water spray at the inlet of the superheater is the key manipulated variable where the key control objective under nominal condition is to maintain the outlet temperature of steam from the system. However, due to disturbances, the water spray can excessively increase leading to saturation at the inlet of the superheater. This, in turn, leads to a two-phase flow causing water hammer that can be very damaging for the plant and should be avoided. Thus, under this situation, the priority of the control objective changes to controlling the saturation at the inlet of the superheater. As the magnitude of disturbance increases further, the outlet temperature approaches the upper bound of safe equipment operation. Thus control priority changes to satisfying safety constraints. For this system, the key variables that estimator exchanges information is the heat transfer rate that is a complex function of many operating variables, thermo-physical properties, transport characteristics, and design variables. The estimator algorithm is a modified extended Kalman filter that is developed for estimation for differential algebraic systems with uncertain algebraic states. The estimator includes a detailed 2-d superheater model with rigorous models for heat exchange between the flue gas side and tube m through tube wall heat exchange and between the tube and flowing steam. Performance of the EB-MPC is evaluated for varying magnitudes of disturbances. Smooth transition from one control objective to the other with desired characteristics of the manipulated variables are observed.

[1] E. C. Kerrigan and J. M. Maciejowski, “Designing model predictive controllers with prioritised constraints and objectives,” in Proceedings. IEEE International Symposium on Computer Aided Control System Design, 2002, pp. 33–38, doi: 10.1109/CACSD.2002.1036925.

[2] J. Barreiro-Gomez, C. Ocampo-Martinez, and N. Quijano, “Evolutionary game-based dynamical tuning for multi-objective model predictive control,” in Lecture Notes in Control and Information Sciences, 2015, vol. 464, pp. 115–138, doi: 10.1007/978-3-319-26687-9_6.

[3] C. E. Cortés, D. Sáez, F. Milla, A. Núñez, and M. Riquelme, “Hybrid predictive control for real-time optimization of public transport systems’ operations based on evolutionary multi-objective optimization,” Transp. Res. Part C Emerg. Technol., vol. 18, no. 5, pp. 757–769, 2010, doi: 10.1016/j.trc.2009.05.016.

[4] N. Ryu and S. Min, “Multiobjective optimization with an adaptive weight determination scheme using the concept of hyperplane,” Int. J. Numer. Methods Eng., vol. 118, no. 6, pp. 303–319, 2019, doi: 10.1002/nme.6013.

[5] B. M. Ticha, V. Sircoulomb, and N. Langlois, “Priority switching approach for multi-objective MPC for brandy distillation systems,” 7th Int. Conf. Control. Decis. Inf. Technol. CoDIT 2020, pp. 1093–1098, 2020, doi: 10.1109/CoDIT49905.2020.9263917.