Distributed and learning-based model predictive control
S.H. Mallick (TU Delft - Mechanical Engineering)
B. De Schutter – Promotor (TU Delft - Mechanical Engineering)
A. Dabiri – Copromotor (TU Delft - Mechanical Engineering)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
This dissertation advances Model Predictive Control (MPC) by addressing two major challenges: computational complexity and model uncertainty. The research focuses on distributed control and learning-based approaches to facilitate MPC for hybrid systems, large-scale networks, and systems with limited model knowledge.
To reduce computational burden, new distributed MPC methods for piecewise affine systems are developed, providing efficient convex optimisation-based solutions with guarantees on consistency and feasibility. Learning-based policies are also integrated with MPC, shifting computationally intensive tasks offline and enabling efficient control of hybrid systems and autonomous vehicles.
To address uncertainty, reinforcement learning (RL) is combined with MPC to learn uncertain controller components from data. Novel distributed MPC-RL frameworks are proposed for networked systems. Furthermore, centralised MPC-RL controllers are proposed for applications such as greenhouse climate control and energy systems. The results demonstrate that distributed and learning-based MPC can significantly improve scalability, efficiency, and performance in complex real-world control problems.