With the proliferation of the distributed energy resources (DERs), the scheduling and control of the distribution network have become more complicated. To cope with the uncertainty nature of distributed generation, a multi-timescale optimal dispatch method in active distribution network (ADN) based on the model predictive control (MPC) is proposed in this paper. First, based on MPC, a hierarchical scheduling framework for ADN is established, including long-timescale stage, and short-timescale stage. Then, via coordinated control of various resources in the ADN, i.e., distributed generators, energy storage, capacitor banks and OLTC transformer, the impact of intermittent renewable energy and load forecast errors can be reduced. Finally, considering the coupling characteristics of active and reactive power in the ADN, a joint active and reactive power optimization model is proposed to further reduce the network loss. Numerical simulation on a modified IEEE-33 distribution network system verifies the correctness and superiority of the proposed scheduling approach.