Aiming at the effect of community characteristics of complex networks on collaborative control of multi-agent systems,a fast flocking control algorithm based on community division is proposed for the multi-agent system with the characteristic of the Erdos-renyi(ER) random network and Barabasi-albert(BA) scale-free network, which considers the relative concentration characteristics of agents in every community and incorporate attractive force of virtual leader between different communities, so the "block" phenomenon in evolution process because of communication constraint can be avoided and algebraic connectivity of the system is improved. Simulation results show that the convergence characteristics of the flocking behavior of the multi-agent system with complex networks are affected by the average degree of ER and BA networks and the power exponent of the degree of BA network distribution, and optimizing the number of partition communities is beneficial to improve the convergence speed.