Add to Google Calendar
Date:  Thu, August 29, 2019
Time:  2:00pm - 4:00pm
Location:  Holmes Hall 389
Speaker:  Douglas Ellman, EE Ph.D. Candidate

Abstract:

Demand response (DR) refers to a variety of mechanisms that aim to modify customers’ electricity consumption in order to improve the operation of the electric grid. Baseline-based DR is a widely-used type of DR that pays customers based on how much they change energy consumption compared to some baseline. While the goal of baseline-based DR is to modify energy consumption during DR events (e.g. peak demand days), baseline-based DR may also create unintentional and undesirable incentives for customers to modify baseline energy consumption in order to increase DR payments. However, for DR programs with baselines based on multiple days and with uncertain DR event schedules, determining optimal customer behavior is challenging. Without accurate models of customer behavior, utilities risk designing DR programs that encourage undesirable customer actions, and may incorrectly estimate cost and performance of DR programs.  

This presentation will describe research on optimization models to minimize costs for customers in baseline-based DR programs, and applications of the models to DR programs that are operating in Hawaii and New York. A model based on dynamic programming optimizes decisions for a DR customer that can modify electric load by certain amounts each day at a given costs. A second model based on model predictive control and stochastic programming obtains near-optimal decisions for a DR customer with a battery energy storage system and solar under a certain electricity tariff. Future work will explore the use of reinforcement learning to find near-optimal customer actions under multiple uncertainties (event schedule, customer demand, renewable energy production).