Non-linear control is common in industry. Indeed, there is a growing use of constrained model predictive control methods, which are inherently nonlinear. This means that it is not feasible to directly apply traditional indirect identification procedures which assume that the control law is linear. In this paper, we address this problem by using a virtual closed loop where the virtual controller is chosen to be linear irrespective of the linearity of the true controller. We employ Subspace identification procedures.
43rd IEEE Conference on Decision and Control 2004. Proceedings of the 43rd IEEE Conference on Decision and Control 2004 (Bahamas 14-17 December, 2004) p. 364-369