For dynamical changed network topology and unstable links, this paper proposes a mobility prediction-based weighted clustering algorithm for vehicular ad-hoc network with high mobility nodes. In the CH election stage, it computes vehicle’s priority through the similarity, intimacy and centrality, the vehicles with the highest priority in their neighborhoods are selected as the CHs, this scheme suitable for large scale network. NS2 simulation results show that the proposed clustering scheme improves the stability of cluster structure, prolongs the CH duration time, improves the delivery ratio.