Real-world Experience with Residential Demand Response Baselines

Baseline, or counterfactual, load profiles are almost always needed to evaluate the performance of load flexibility (demand response, DR) programs. Baseline methods are well-established for commercial and industrial (C&I) DR, but it is not clear that methods developed for C&I sites will work equally well for groups of residential connected devices, such as thermostats and electric vehicles (EVs). This paper uses historical data from hundreds of DR events involving many thousands of smart thermostats and EVs to evaluate several baseline methods used in industry today. The results suggest that standard statistical methods frequently lead to inaccurate baselines. Baselines for HVAC cooling DR events are frequently biased when the event day follows a number of colder days. Inaccuracies also emerge when baseline methods are applied to smaller groups of EVs, in which behavioral variance results in large deviations from one day to the next. Results from an engineered baseline for EVs suggest that more accurate baseline methods are possible.