A learning- and scenario-based MPC design for nonlinear systems in LPV framework with safety and stability guarantees (Q6600975)

In the work, a learning-based model predictive control (MPC) design approach is proposed for systems described within the linear parameter-varying framework. Bayesian inference neutral network approach is used to learn from input-output data in a linear parameter state space model with epistemic uncertainty quantification. Scenario-based MPC is proposed to consider safety when generating scenarios. To guarantee the closed-loop stability, a parameter dependent terminal cost and a controller were designed, which can improve the control performance. In the end of the paper, results of a numerical experiment are shown.