Nonlinear model predictive control with enlarged terminal sets using support vector machine

Abstract

In this paper, model predictive control (MPC) of nonlinear systems subject to input and state constraints is considered, for which nominal closed-loop stability is guaranteed. We propose the use of a large terminal invariant set and an estimate of the terminal cost to reduce the online computational burden of MPC. These terminal sets and costs are learned off-line via support vector machine method. Its main advantage with respect to other well-known techniques is the reduction of online computational effort by relaxing the terminal constraints. An example illustrates the efficiency of the approach.