Optimization problems in model predictive control (Q4371071)

Some basic concepts of model predictive control (MPC) are reviewed and the optimization problems that arise here are discussed. Standard optimization methods provide efficient solutions if the model is linear and the goal function is quadratic. The ``tail'' problem can be avoided by the assumption that, after a finite horizon, a stabilizing linear control law is used. However, for nonlinear models, MPC needs solutions of nonconvex optimization problems, and the authors can not show a reasonable way how to overcome theoretical and computational troubles. It is mentioned that control laws generated by MPC may be discontinuous and robustness properties are in question here. The review is rather complete with respect to the western publications but all achievements of Russian scientists are absolutely ignored. The authors' attention is concentrated on quadratic goal functions though, for many applications, e.g. inventory-production problems, linear goal functions are more adequate, and some new variants of MPC appear here. For the entire collection see [Zbl 0872.00046].