Data-Driven Approaches to Demand Response: Studies on Thermostatically Controlled Loads and Electric Vehicles

This dissertation presents a set of studies that use data collected on thermostatically controlled loads (TCLs) and electric vehicles (EVs) to (i) understand and improve their participation in demand response (DR) strategies, and (ii) quantify their capabilities as DR providers. First, we focus on TCLs and formulate a centralized control scenario in which a large, heterogeneous population of TCLs are controlled to provide ancillary services. We investigate the system-level benefits that such a control scenario can impart, and focus on evaluating modeling strategies that capture TCL population dynamics when disturbances to individual TCLs are considered. We then discuss the upstream communication requirements from the TCLs to the main controller and the characteristics of the underlying cyberinfrastructure, and argue that end users will prefer a strategy in which they do not have to share real-time information with the centralized controller. Using measurements obtained at a neighborhood-level load aggregation point, we develop state estimation techniques to replace the need for upstream information exchange between individual end use loads and the controller. Our results have important policy implications for appliance standards and the participation of loads in DR services. We show that the proposed aggregate TCL modeling strategy better captures the dynamics of an aggregate TCL population with no additional computational burden in comparison to state-of-the-art strategies. We also show that if individual thermal parameters of TCLs are available to the main controller, the controller can improve on its knowledge of the state of the TCL population using measurements obtained from a neighborhood-level load aggregation point. Following our study of TCLs, we focus on centralized control of an aggregation of commercial EV charging stations. We investigate the benefits of managed EV charging to different stakeholders engaged in the reliable operation of the power grid. To do this, we leverage data collected from smart devices to overcome limiting assumptions commonly made in the literature regarding: (i) driving patterns, driver behavior and driver types; (ii) the scalability of a limited number of simulated vehicles to represent different load aggregation points in the power system with different customer characteristics; and (iii) the charging profile of EVs. As part of this study, we investigate the relationship between the EV infrastructure availability, EV load flexibility and benefits to stakeholders, a relationship which has implications for future improvements to DR programs. We show that managed EV charging can decrease the contribution of EV charging loads to the system peak load by approximately 40%, and reductions up to 24% in the monthly bills are possible for EV aggregations.