This paper proposes a hierarchical framework for the real-time dispatch of Active Distribution Networks (ADN) with high penetration of Electric Vehicles (EVs) and renewable energy, based on near term forecast information and long term value function approximation. To solve the curse of dimensionality after large-scale EVs access, in the upper layer, an equivalent EV cluster model is developed; while in the lower layer, the power allocation algorithm considering the diverse characteristics of EVs is proposed to make specific strategies for each EV while tracking the upper layer instruction. To handle the stochastic EV charging behaviors, maximum output of renewable energy and real-time price, the near term forecast information of those stochastic factors are considered in the proposed method. Further, value function approximations are adopted to evaluate the value function of the system after the near term decision, by which way the real-time dispatches of EV clusters, renewable energy, and electricity purchase can be given stage by stage. Simulation cases verify the effectiveness and the robustness of the proposed model in the real-time coordinated dispatch of ADN with high penetration of EVs. This work is supported by Science and Technology Project of China Southern Power Grid Co., Ltd. (No. GDKJXM20172831).