Robust model predictive control of nonlinear systems using input-output models

Presents a framework for nonlinear input-output-model-based predictive control from a Bayesian decision theoretic view point. The parameters are modelled as random variables and their probability distributions are computed and used explicitly in the control computation. The framework naturally yields on-line model refinement and “cautious control” against parametric uncertainties. This is important as nonlinear input-output models often contain a large number of parameters and input excitation needed for acceptable parameter estimation is difficult to achieve off-line. The feasibility of the framework is demonstrated by deriving a prototype algorithm for the second order Volterra model. The algorithm is interpreted in the classical model predictive control (MPC) framework and connections to other robust control strategies are discussed