Robust Model-Free Control, Optimization, and Learning in Cyber-Physical Societal Systems

April 2, 2018

Jorge I. Poveda

University of California, Santa Barbara

The deployment of advanced real-time control and optimization strategies in socially-integrated engineering systems could significantly improve our quality of life while creating jobs and economic opportunity. However, in cyber-physical systems such as smart grids, transportation networks, healthcare, and robotic systems, there still exist several challenges that prevent the implementation of intelligent control strategies. These challenges include the existence of limited communication networks, dynamic environments, multiple decision makers interacting with the system, and complex hybrid dynamics emerging from the feedback interconnection of physical processes and computational devices. In this talk, I will present a set of tools for the analysis and design of model-free feedback mechanisms that can cope with these challenges, and that are suitable for the real-time control and optimization of cyber-physical societal systems. The first part of the talk will focus on the problem of designing a class of robust model-free adaptive pricing mechanisms for systems such as the smart grids, transportation networks, and the Internet, where users behave in a selfish way, and where the objective of the social planner is to maximize the total welfare of the system. Next, I will show that this problem belongs to a broader family of model-free extremization problems, and I will present a general framework for the design of a family of algorithms that can successfully optimize the performance of cyber-physical systems having unknown mathematical models. Finally, I will illustrate how these results can be extended to achieve distributed control of large-scale autonomous systems by implementing novel robust coordination and synchronization feedback mechanisms. The talk will finish by discussing some future directions and preliminary results in the areas of data-driven hybrid control and security in stochastic learning dynamics.

Published on April 2nd, 2018Last updated on March 29th, 2018