Impact of Disturbances on Modeling of Thermostatically Controlled Loads for Demand Response

Aggregations of thermostatically controlled loads (TCLs) have been shown to hold promise as demand response resources. However, the evaluation of these promises has relied on simulations of individual TCLs that make important assumptions about the thermal dynamics and properties of the loads, the end-user's interactions with individual TCLs and the disturbances to their operation. In this paper, we first propose a data-driven modeling strategy to simulate individual TCLs-specifically, household refrigeration units (HRUs)-that allows us to relax some of these assumptions and evaluate the validity of the approaches proposed to date. Specifically, we fit probability distributions to a year-long dataset of power measurements for HRUs and use these models to create more realistic simulations. We then derive the aggregate system equations using a bottom-up approach that results in a more flexible [linear time invariant (LTI)] system. Finally, we quantify the plant-model mismatch and evaluate the proposed strategy with the more realistic simulation. Our results show that the effects of invalid assumptions about the disturbances and time-invariant properties of individual HRUs may be mitigated by a faster sampling of the state variables and that, when this is not possible, the proposed LTI system reduces the plant-model mismatch.