Multipoint Iterative Learning Model Predictive Control

Abstract

Iterative learning model predictive control (ILMPC), endowed with the merits of iterative learning control and model predictive control, has excellent abilities of disturbance rejection and constraints handling. It has been widely applied in regulation of industrial batch processes for its remarkable reference tracking performance. In practice, the presence of strong system nonlinearity fundamentally challenges the model mismatch cyclewise invariance assumption that underlies numerous synthesis of many ILMPC methods. It may further induce additional conservatism and thus ultimately leads to performance degradation. To circumvent this issue, in this paper a novel ILMPC scheme relying upon past error compensation from multiple time instances is presented, in which the associated weights are adaptively optimized, thereby resulting in considerable enhancement of robustness. These superior performances are confirmed by numerical experiments on injection molding process, a typical batch process.