A cluster partitioning method is proposed to address the issues of overvoltage and power flow reversal caused by the integration of high penetration distributed photovoltaic power sources into the distribution network, as well as the problems of long transmission distance, too many transformation devices, and immature technology brought about by centralized photovoltaic integration. This method considers both the system’s structural and functional characteristics. Firstly, the structural indicators adopt improved modularity indicators. In contrast, the functional indicators adopt cluster active and reactive power balance and node membership indicators, constructing an optimization model for the comprehensive indicator system. Secondly, to fully demonstrate the expression of the indicator system model, some improvements have been made to the traditional genetic algorithm, and the adjacency matrix has been established as the encoding method of the algorithm. Finally, an improved IEEE33 node and a 10kV actual distribution network in a certain area were analyzed and verified as examples. The simulation results show that the proposed cluster partitioning method improves power balance by more than 30% while ensuring minimal changes in modularity, indicating that the method has certain effectiveness in cluster partitioning.