Benchmarking Of Aggregate Residential Load Models Used For Demand Response

Residential loads can provide ancillary services such as frequency regulation to an electric power system. Aggregate load models aim to capture the dynamics of the demand-responsive loads in an accurate and computationally-tractable way, which allows the models to be incorporated into controllers and observers used by demand response providers. A variety of aggregate models have been developed; however, their accuracy has not been benchmarked against one another in comparable scenarios. This paper compares the accuracy of two Markov-based and one transfer function-based aggregate air conditioner (AC) models against a realistic simulation model with timevarying outdoor air temperature and temperature-dependent AC parameters. We also extend the existing models to cope with the time-varying outdoor air temperature. We find that 1) the more detailed Markov model is more accurate, 2) updating the Markov transitions as a function of the outdoor temperature trend decreases prediction error in both Markov models, 3) the transfer function model performs worst, likely because the simulation scenario differs significantly from the assumptions used to develop the model.