Quantifying the Impact of Participant Override Behavior on a Summer Demand Response Program

Demand response (DR) programs are an important tool for maintaining grid reliability and for decarbonization of the power sector. Summer residential air condition (AC) direct load control (DLC) is a common DR program. In DLC programs, consumers cede control of their thermostat to the DR provider for a small number of events to reduce AC load. DLC programs often include an override feature, which allows participants to deviate from the DR provider's thermostat settings. No existing research has used empirical data to quantify the impact of override behavior on the reliability contribution of a DLC program. Ignoring overrides could significantly overestimate the system value of DLC programs, particularly in a warming world. We design a model to predict the cooling load of thermostat's participating in ecobee's Donate Your Data initiative under various temperature conditions, then quantify the program's reliability contribution using the effective load carrying capability (ELCC) metric. We use this model to evaluate the performance of 403 ecobee thermostats participating in Southern California Edison's 2019 Smart Energy Program at a range of AC unit efficiencies. We find that overrides reduce the program's ELCC, or reliability contribution, by roughly 10 percentage points (or from 34% to 25%). Across DLC events, roughly 15% of participants use overrides to exit the event. This override behavior was similar for all events, regardless of duration. Overrides led to a 50% reduction in cooling load savings, with larger reductions for longer DLC events. In particular, the impact of override behavior was lowest in the first hour of events and decreased in value until the last hour, when it caused an increase in cooling load relative to if there had been no participation. This research indicates DR providers should consider the impact of override behavior and the length of events when determining the reliability contribution of their program.