Contributions to distributed MPCcoalitional and learning approaches

Laburpena

A growing number of works and applications are consolidating the research area of distributed control with partial and varying communication topologies. In this context, many of the works included in this thesis focus on the so-called coalitional MPC. This approach is characterized by the dynamic formation of groups of cooperative MPC agents (referred to as coalitions) and seeks to provide a performance close to the centralized one with lighter computations and communication demands. The thesis includes a literature review of existing distributed control methods that boost scalability and flexibility by exploiting the degree of interaction between local controllers. Likewise, we present a hierarchical coalitional MPC for traffic freeways and new methods to address the agents' clustering problem, which, given its combinatoria! nature, becomes a key issue for the real-time implementation of this type of controller. Additionally, new theoretical results to provide this clustering strategy with robust and stability guarantees to track changing targets are included. Further works of this thesis focus on the application of learning techniques in distributed and decentralized MPC schemes, thus paving the way for a future extension to the coalitional framework. In this regard, we have focused on the use of neural networks to aid distributed negotiations, and on the development of a multi agent learning MPC based on a collaborative data collection.