The climate crisis is likely to be the largest factor in inequality growth over the next century. Mitigating these effects require novel techniques in both decarbonizing energy production, and minimizing current energy consumption. In this talk, I discuss how large-scale infrastructure system control can address both problems simultaneously. One challenge with increasing the penetration of renewable energy in the power grid is that this results in higher levels of uncertainty and variability of supply and demand in the grid. I argue that electricity demand, such as from large residential buildings, can be used to help balance these fluctuations in supply in real-time by varying the building demand (known as demand response). This necessitates the control of large numbers of individual apartment units, as well as heating/cooling energy systems, in a coordinated fashion in order to produce a desired aggregate electricity demand during real-time operation. Via novel distributed control of the energy consumption of buildings, one can provide a service to the electrical grid by enabling real-time demand response. This allows online balancing of the supply and demand of the electrical grid, all while minimizing the electrical bill of the building occupants.
The theoretical core behind distributed control algorithms lies in fundamental network science, and the interplay of machine learning techniques with robust control systems. The presence of networks in distributed control systems leads to interesting theoretical questions regarding scalability and modularity of such methods: how large of a network can you control, and how does your system behave when new systems are added to the network? I also discuss how fundamental network science can inform us about how we can modify pre-existing infrastructure networks for improved control, or how to design them a priori. When using machine learning tools in the loop with a control system, it is difficult to design the controller using traditional design criteria. I discuss how one can achieve such desirable properties of control systems, such as model convexity, with state-of-the-art machine learning tools. I will address areas of infrastructure control that highlight the importance of each of these theoretical questions, and demonstrate the viability of our distributed control methods with experimental results from the NEST Smart Building Demonstrator facility.
Bio: Mathias Hubaba de Badyn is a postdoctoral scholar in the Automatic Control Laboratory at the Swiss Federal Institute of Technology in Zürich. He obtained his PhD in distributed control in 2019 at the William E. Boeing Department of Aeronautics and Astronautics at the University of Washington with Mehran Mesbahi, as well as a Master's of Mathematics in Optimization in 2019 and a Master's of Aeronautics and Astronautics (Control Systems Track) in 2017. His undergraduate degree was in the Combined Honours in Physics and Mathematics program at the University of British Columbia. Mathias’ research interests mesh fundamental network science with control, optimization, and learning for smart city infrastructure systems.
