Aiming at the characteristics of large-scale data in the distribution network with distributed generation, as well as the limitations of the traditional grid partitioning method based on complex network theory, such as irrational partitioning and insufficient reactive power regulation capability, a two-stage division method based on complex network theory is proposed in this paper. In the first stage, based on mapping partition, load nodes are initially partitioned according to the principle of "minimum electrical distance" so that the reactive power resources can better coordinate and control the load nodes in the region. At the same time, the initial association size in the second stage is reduced. In the second stage, the Louvain community discovery algorithm is used for partition aggregation, and the initial partition is adjusted by improving the partition modularity function to ensure the reactive power adjustment capability of the partition. The IEEE 69-node system is used as an example to verify the effectiveness of the proposed method in this paper. The number of partitions and the amount of inter-area reactive power transfer after partitioning are reduced compared with the traditional method. The results show that the proposed partitioning method has stronger reactive power regulation capability and is more suitable for autonomous control of regional voltage and reactive power in distribution networks.