This paper concentrates on the leader-following formation control problem of multi-agent systems in the case of enormous quantity agents inside. A new method based on min-weighted persistent formations is proposed to model the distributed interaction topology. The multi-agent topology can be distributed by the same-size k-means algorithm. The control for persistent formation is presented and the stability analysis of the control input is analyzed. The simulations experiments of the distributed generation and control of persistent formation illustrate the effectiveness of the algorithm. The distributed method can avoid the frequent changes of the topologies of the optimally persistent formation and relatively independent movements of every node can also save time.