Aiming at the charging/discharging scheduling problem for charging station aggregation of electric vehicles, an optimal energy scheduling strategy is proposed for a electric vehicle aggregator (EVA) that takes into account the demands of vehicle owners with the goal of minimizing the long-term power purchase cost of EVA. Firstly, adequate consideration of vehicle owners demands and the time-varying nature of external grid tariffs, an operational framework for EVA energy scheduling management is established. Secondly, the electric vehicles (EVs) are classified into three charging modes according to the difference of users' charging demands, that is, two-way-dispatch EVs, one-way-dispatch EVs and fast-dispatch EVs, and load models are established respectively. Then, based on reinforcement learning theory the real-time energy scheduling strategy is designed for EVA. Finally, the reasonableness and effectiveness of the proposed algorithm are verified by simulation examples of real data and comparing with other greedy algorithms. The results show that the first two scheduling modes based on the proposed strategy can save 54.1% and 47.5% of the cost of EVA in one month, compared with the scheduling mode under the greedy algorithms.