A probabilistic equivalent method for doubly fed induction generator (DFIG) based wind farms is proposed in this paper. First, the wind farm equivalent model is assumed to be composed of three types of equivalent DFIGs with different dynamic characteristics. The structure of equivalent model remains constant, whereas the parameters change with the migration of different scenarios in the wind farm. Then, historical meteorological data are utilized to investigate the probability distribution of key equivalent parameters, such as capacity, wind speed and electrical impedance to the point of common coupling. Each type of equivalent DFIG is further clustered into several groups according to their active power output. Combinations are created to generate representative scenarios. The probabilistic equivalent model of wind farm is finally achieved after removing invalid combinations. Most matched representative scenarios can be predicted according to the real-time measurement. The equivalent model is applied to the probabilistic power flow calculation and the stability analysis of test systems.