New MPC method controls unknown nonlinear systems with stability guarantees
Robust controller learns from noisy data, guarantees stability for nonlinear dynamics.
A new paper proposes a data-driven robust min-max MPC scheme for unknown nonlinear systems with process disturbances. The unknown dynamics are represented using vector fields built from basis functions, yielding an equivalent linear form with unknown matrices characterized via set-membership learning from noisy data. Two online scenarios are considered: noise-free state measurements and states corrupted by process disturbance. For each case, a Lyapunov-based semidefinite program computes a stabilizing state-feedback controller. The scheme guarantees recursive feasibility and either exponential or robust stability, depending on the presence of process disturbance. Simulation on benchmark examples shows effectiveness compared to existing data-driven and model-based controllers.
- Method bridges gap between data-driven MPC and nonlinear control by handling process disturbances.
- Uses set-membership learning from noisy data to characterize unknown system matrices.
- Guarantees recursive feasibility and exponential/robust stability via Lyapunov-based SDP synthesis.
Why It Matters
Enables safe control for real-world nonlinear systems where models are unknown and data is noisy.