Abstract Thermostatically controlled loads (TCLs) have great potential to participate in demand response (DR) programs due to their thermal storage and transfer characteristics. In this paper, we propose a grouping control strategy based on customer comfort and power matching degree for aggregated TCLs. We also model an individual TCL based on equivalent thermal parameter (ETP). The aggregated TCLs are modeled with the time-varying Markov chain and 2D state-bin transition with the consideration of their heterogeneities. A load selecting process is introduced based on a multi-scale priority sorting composed of normalized temperature distance, power rating similarity and controlled times. These indexes increase the accuracy and speed of the tracking response to the optimum control signal. The simulation results verify that the proposed control strategy has a comprehensive performance on the tracking response, customer comfort, and the fairness of TCLs participated in DR programs.