The residential air conditionings (RACs) are widely considered as one of the most important demand response (DR) resources due to the thermal storage characteristics. However, due to the uncertainties of the outdoor environment and the customers' behaviors, the RACs' operation states and power consumption are difficult to predicate. Facing this issue, this paper proposes a dynamic control strategy for the RACs to participate in DR program considering these uncertainties. Firstly, a single dynamic RAC model considering the uncertain environment and customer behaviors is developed. On this basis, a dynamic aggregate model of RACs is established with different number of RACs. Then, the dynamic aggregate model is identified by actual operation data. A dynamic rolling control strategy-based temperature set-points for large-scale RACs to participate in DR program is formulated. Moreover, the DR provided by RACs is divided into three levels according to the-power reduction, where the corresponding control strategies at each level are proposed. Finally, the proposed models and methods are verified by employing the actual data of the urban residential communities in Changzhou City, China. The simulation results show that the proposed control strategy is accurate and effective.