EAGER: Real-Time: Learning, Selection, and Control in Residential Demand Response for Grid Reliability

点击翻译按钮获取中文摘要

As renewable energy sources increase and conventional generators retire, demand response (DR) has been utilized to address the reliability issue on balancing real-time demand and supply in power grids. However, the potential of residential DR, which is the largest share of electricity demands, has not been fully exploited in practice. Existing pilots reveal many issues, such as i) small monetary rewards which play a limited role in user participation, ii) user dissatisfaction when utility companies exploit DR resources extensively, and iii) the lack of reliability due to the unpredictability of user behavior. In collaboration with ThinkEco Inc, this proposal will develop novel and applicable approaches for residential DR with provable guarantees. The method will learn DR behavior, select the correct residential users, and automatically control residential appliances -- all in the service of enhancing system reliability. The research will be tested and validated on real-world residential DR programs using ThinkEco platforms. The research results will advance real-time learning for human-in-the-loop societal systems with applications ranging from transportation to power grids to AI-enabled systems of the future. The team is strongly committed to providing opportunities in STEM to K-12, women, and under-represented minorities. Moreover, the close collaboration between academia and industry promises a fast and effective transition of academic results to industry practice. Specifically, by understanding users' energy consumption behavior from both historical and real-time measurements, and adjusting user selection and control strategies in real-time, this proposed research will invent DR learning and control mechanisms to satisfy various power grid operation requirements. A major theme in this proposal is to close the loop between learning (exploration) and control (exploitation) in human-in-the-loop societal systems: how to learn (explore) user behavior while taking good control actions (exploitation) at the same time. There is a fundamental tradeoff between exploration and exploitation, and the proposed research aims to uncover the tradeoff and design real-time decision-making rules to achieve near-optimal performance for residential DR. Different from the conventional approaches to learning in computer science or statistics, this proposal aims to tackle the challenge of intertwined interactions between human users and the engineered systems.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.