Model reduction and MIMO model predictive control of gas turbine systems

Abstract

This paper presents the development and experimental implementation of an online, fully nonlinear model predictive controller (NMPC) for a gas turbine. The reduced-order, internal model used in the controller was developed from an original high-order, physics-based model using rigorous time scale separation arguments that may be extended to any gas turbine system. A control problem for a prototype gas turbine system is then formulated as a MIMO optimisation problem that may be addressed through NMPC. The controller objective is to regulate compressor exit pressure while tracking compressor bleed demand. The constraint set includes the range of the control actuators, the valid range of the internal model, and safety limits such as the compressor surge margin and maximum turbine inlet temperature. This formulation also includes derivation of the conditions sufficient to prove nominal asymptotic stability of the closed-loop system.Controller performance is then evaluated through representative tracking and regulation experiments. This highlights the advantages of the proposed controller's constraint handling and disturbance rejection. More broadly, these experiments demonstrate that a nonlinear model predictive controller can be successfully deployed on a gas turbine, in real time, without the need for any linearisation.