Sensitivity-Assisted multistage nonlinear model predictive control: Robustness, stability and computational efficiency

Mandar Thombre, Zhou Joyce Yu, Johannes Jäschke, Lorenz T. Biegler

Abstract: Highlights•A robust multistage NMPC strategy with a sensitivity-based approximation - samNMPC - is proposed, where the scenario tree is decomposed into critical and noncritical scenarios based on likelihood of constraint violations.•A reduced problem is solved over the critical scenarios, and the noncritical scenarios are substituted with their sensitivity-based approximations.•Critical scenarios grow linearly instead of exponentially with the length of robust horizon, allowing for computations with longer robust horizons.•Stability properties of the proposed samNMPC method are discussed.•The samNMPC method is compared to the ideal multistage NMPC with the help of two case studies, and is shown to be significantly faster while having similar performance and robustness.