Increasingly, customer-sited distributed energy resources (including distributed generation, energy storage systems, controllable appliances, and electric vehicles) are being used as assets to support the operation of electric power systems through grid service programs, in which customers are compensated for their devices' participation or performance in responding to power grid needs. In this dissertation, we study how to best control customer energy resources (with a focus on battery energy storage systems) in order to reduce customer costs while participating in utility tariffs and grid services programs. We focus on a commonly implemented, but not widely studied, class of grid service programs known as baseline-based demand response, where customers are compensated based on how much their electric load during grid service events differs from a "baseline load" based on their prior electric load. Optimizing control under baseline-based demand response presents additional challenges (compared to demand response based on time-varying pricing without baselines) due to uncertainty in current effective electricity prices and significant time delay between control decisions and learning the cost impact of these decisions.

This work aims to develop methods and insights to improve the utilization of customer resources in grid services programs, in order to ultimately reduce system cost, improve reliability, and support integration of renewable generation. Control methods studied and developed in this work can be used to manage customer resources in real time or to support studies of grid service program designs.

This dissertation addresses our topic from several angles: (1) analytically, with mathematical analysis of the underlying optimization problems, (2) numerically, by developing and evaluating performance of control algorithms based on dynamic programming, stochastic programming with model predictive control, imitation learning, and reinforcement learning, and (3) practically, by demonstrating and testing control algorithms on a hardware-in-the-loop test bed.

Doug Ellman is a PhD candidate in the Department of Electrical & Computer Engineering at the University of Hawaiʻi at Mānoa who has worked as a research assistant for the Hawaiʻi Advanced Wireless Technologies Institute (HAWTI). Doug earned a Bachelor of Arts in Physics from Princeton University and a Master of Science in Technology and Policy from the Massachusetts Institute of Technology. Doug has worked in a variety of roles in the clean energy industry (for Cubic, SolarCity, and Tesla) and as a physics teacher (at Maryknoll and ʻIolani).