Noise and Parameter Heterogeneity in Aggregate Models of Thermostatically Controlled Loads

Aggregate models are used in the analysis and control of large populations of thermostatically controlled loads (TCLs), such as air-conditioners and water heaters. The fidelity of such models is studied by analyzing the influences of noise and parameter heterogeneity on TCL aggregate dynamics. While TCLs can provide valuable services to the power systems, control may cause their temperatures to synchronize, which may then lead to undesirable power oscillations. Recent works has shown that the aggregate dynamics of TCLs can be modeled by tracking the evolution of probability densities over discrete temperature ranges or bins. To accurately capture oscillations in aggregate power, such bin-based models require a large number of bins. The process of obtaining the Markov state transition matrix that governs the dynamics can be computationally intensive when using Monte Carlo based system identification techniques. Existing analytical techniques are further limited as noise and heterogeneity in several thermal parameters are difficult to incorporate. These challenges are addressed by developing a fast analytical technique that incorporates noise and heterogeneity into bin-based aggregate models. Results show the identified and the analytical models match very closely. Studies consider the influence of model error, noise and parameter heterogeneity on the damping of oscillations.