Adaptive Control for Hybrid Uncertain Systems


Staff Mentor:

dr.ir. S. Baldi (Simone)



Keywords:

Hybrid systems; Learning and adaptive control

Description:

A hybrid system is a dynamic system that involves both continuous and discrete event dynamics. As a result of its double nature, the state of the hybrid system can not only flow, but also jump (as a consequence of discrete events and logics).

Hybrid systems are ubiquitous in our modern technological world, covering many applications in transportation (urban traffic control, autonomous vehicles, etc.), energy (smart grids and buildings, renewable energy management, etc.), and many other crucial fields with high societal impact. The relevance of this kind of systems calls for control techniques that can tame the complexity of the problem and, most importantly, provide with efficient solutions also in the presence of uncertainty. When the dynamics of the hybrid system or the way in which they interact are uncertain, model-based control techniques are no more applicable, and appropriate adaptive solutions must be found. Cognitive adaptive control is a recently developed method for adaptive optimal control of uncertain systems aiming at minimizing an objective function related to system performance. Cognitive adaptive control combines: an approximate solution to the Hamilton-Jacobi-Bellman (HJB) equation associated with the optimal control problem; and a cognitive adaptive optimization algorithm that online updates the control law to approach the optimal solution defined by the HJB equation.

The extension of cognitive adaptive control to hybrid systems is to a great extent an unexplored research field. The aim of this research proposal is to investigate and develop efficient adaptive methods for uncertain hybrid systems that are able to compute control actions driving the hybrid system dynamics toward optimal operational regimes. In order to deal with the challenging scale and complex dynamics of real-world systems several aspects that could be considered are: hybrid modelling (hybrid problem formulation and appropriate extension of the HJB equation); tractability and scalability to large-scale systems; extension to general classes of hybrid systems.

Prerequisites: research oriented attitude


© Copyright Delft Center for Systems and Control, Delft University of Technology, 2017.