Abstract: Thermostatically controlled load（TCL） is a valuable demand response resource of the power grid, which usually participates in load regulation in the form of clusters. Selecting the appropriate units in the cluster is one of the critical links for TCL to participate in the load regulation of the power grid. Considering the efficiency of information interaction and the difference between users′ participation willingness, this paper proposes a random selection model for TCL units. A general normalized temperature state index is established for various TCL types. Taking the normalized temperature state as the main reference of response priority evaluation and introducing participation willingness coefficient, the controlled probability function of the TCL units is constructed. A random selection algorithm of TCL units is proposed, which can guarantee the user experience and perform the regulation target accurately. The random selection process of TCL units based on cluster status statistics and control signal broadcast is designed. Through the example simulation, the effectiveness of the proposed model is verified.