On a quasi-convex method for designing robust predictive controllers

Abstract

A systematic method of designing robust predictive controllers for systems with parametric ellipsoidal uncertainty is proposed. Ellipsoidal uncertainty descriptions arise in many engineering applications and are relevant to predictive control operations where model parameters are often found by fitting experimental data. A significant feature is that the robust predictive controller retains the servo performance of a nominal predictive controller designed using conventional methods. The synthesis procedure involves solving a quasi-convex optimization problem that has analytic expressions for the gradients. The optimization problem is based on rigorous theoretical foundations for robust stability, and convergence to the global solution is guaranteed. An illustrative design example is given.