Thermal discomfort prediction with sparse residential thermostat dataset

We develop a probabilistic method for predicting the thermal comfort of residential occupants during demand response (DR) events. Specifically, we estimate the probability that occupants will change the thermostat setpoint, by calculating their discomfort tolerance based on the degree and duration of discomfort. We also show that we can predict this discomfort tolerance reliably. The primary advantage of our approach is that it requires minimal data, in contrast with other thermal comfort prediction models, i.e., only historical thermostat setpoints from connected thermostats (CTs) and weather data. Since CTs are a prerequisite for DR event participation, this approach requires no additional capital on the part of the utility and is nonintrusive for the customer. At the same time, accurate predictions of occupant comfort allow utilities to tailor DR events for each customer to minimize the likelihood of customer opt-out while maximizing load flexibility.