Massive air-conditioner resources on the consumer side have a huge potential to be involved in system frequency control, and massive scattered individual air-conditioners can be aggregated into air-conditioner groups to participate in system automatic generation control (AGC). As the operating status of individual air-conditioners is random, it is necessary to accurately evaluate the frequency control reserve capacity of air-conditioner groups. In this paper, a data-driven nonintrusive load disaggregation (NILD) method is proposed to incorporate massive air-conditioner groups into AGC frequency control reserve. First, the proposed method constructs a sequence-to-sequence model to nonintrusively forecast the frequency control reserve capacity of individual air-conditioners. Considering the fact that the participation of air-conditioners in AGC may affect the comfort of users, the feasibility of individual air-conditioners is evaluated to participate in AGC frequency control and a selection mechanism is constructed to determine whether they can be included in the frequency control reserve of air-conditioner groups. In the proposed method, individual air-conditioners within a sub-station that participate in AGC are aggregated to derive the air-conditioner group frequency control reserve of the sub-station, and the air-conditioner group frequency control reserve of all sub-stations are aggregated to derive the entire air-conditioner group frequency control reserve. The experiments simulate different degrees of system frequency drops in which air-conditioner groups work cooperatively with thermal units for AGC frequency control. The simulation results demonstrate the validity of the proposed method.