Research > Faculty Projects

CPS: Breakthrough: Collaborative Research: The Interweaving of Humans and Physical Systems: A Perspective from Power Systems

Principal Investigator
Baosen Zhang

Co-PI(s)

Sponsor
National Science Foundation

Award Period
10/01/2015 - 09/30/2018

Abstract
As information technology has transformed physical systems such as the power grid, the interface between these systems and their human users has become both richer and much more complex. For example, from the perspective of an electricity consumer, a whole host of devices and technologies are transforming how they interact with the grid: demand response programs; electric vehicles; "smart" thermostats and appliances; etc. These novel technologies are also forcing us to rethink how the grid interacts with its users, because critical objectives such as stability and robustness require effective integration among the many diverse users in the grid. Our project studies the complex interweaving of humans and physical systems, with power systems as our primary focus. Traditionally, a "separation principle" has been adopted at the interface between humans and physical systems. On one hand, engineers studying physical systems would take user preferences (e.g., utility functions) as exogenous parameters, and optimize the resulting system. On the other hand, economic mechanism designers would focus on finding ways to elicit preference information, taking as a given the underlying physical system operation. Crucially, this separation principle requires that preferences are well-defined, stable, and quickly discoverable. This separation is no longer adequate: users' preferences are often not well-defined; unstable over time; and take time to discover. A single, static "utility function" cannot capture the complex, evolving nature of user's preferences. For example, users do not interact with their smart homes by specifying utility functions; instead, they express preferences implicitly through choices like "do laundry while I am asleep." These preferences and resultant choices can vary greatly over time, leading to significant system-wide consequences. For example, at each time instant, the power grid must regulate physical quantities such as voltage and frequency while matching against the dynamically discovered preferences of users. Our project has two main thrusts. First, we take a fundamentally dynamic view of user preferences. We observe that two new "learning loops" are present in the interface between users and physical systems. Individual users do not know their preferences; instead they learn the value they place on different choices through repeated interaction and exploration. As a result, the system cannot understand user preferences in "one-shot"; rather, repeated interaction and learning with the user is required. Our first thrust studies the entanglement of these "learning loops": with power systems as a concrete focus, we develop new models for user behavior, as well as new approaches for the system to effectively learn user preferences. Our second thrust focuses on the larger system. How do we control and operate a physical system, in the presence of the interacting "learning loops", while mediating between many competing users simultaneously? We apply ideas from the mean field games and optimal power flow to capture, analyze, and transform the interaction between the system and the ongoing preference discovery process. Our methods will yield guidance for market design in power systems where user preferences are constantly evolving.

Research
 
EE logo