Electric vehicles（EVs）are important new adjustable load resources,and predicting their flexibility is an essential prerequisite for implementing optimized dispatch. A method to infer the fleet charging feasibility domain based on normal charging session data of EVs is proposed,forming a historical dataset on the flexibility of EV fleets. Subsequently,considering the high stochasticity of charging load,a probability prediction method for the flexibility of EV fleets based on Gaussian process regression is proposed. The obtained probability prediction results can be used to establish chance constraints for the optimization problem of EV fleet and convert them into deterministic constraints under specific confidence levels. Lastly,simulation verification is conducted using actual charging data. The results show that the proposed method can accurately predict the charging flexibility of EV fleets from both energy and power aspects. By adjusting the confidence level,it is possible to balance the economy and the implementability when optimizing the EV fleets.