Large scale distributed generation (DG) connected into the distribution network (DN) increases the difficulty of DN to cope with the uncertainty and centralized control. Therefore, this paper proposes a multi-timescale distributed active-reactive coordinated optimization method based on DG cluster (DGC) of DN. At the day-ahead level, a centralized active-reactive power coordinated optimization model is established to arrange the power and start-stop plan of long-time adjustment equipment. Then, considering the influence of DG and load forecast errors on voltage, a DGC cluster partition model is established based on day-ahead scheduling result. At the intraday level, targeting the minimum of optimal cost and voltage deviation, a distributed active-reactive power coordinated optimization model is established based on DGC and SADMM, to revise the day-ahead power plan of each continuous adjustment equipment. Finally, taking the IEEE33-node system as an example, the effectiveness of the proposed method is verified.