To explore the trajectory distribution and movement pattern of the process of lane-changing, the simulation experiment of lane-changing on the freeway based on a driving simulator was performed. The data of the lane-changing trajectory were collected. Then, Spectral clustering, hierarchy(HDBSCAN), distribution-based (gaussian mixture model, GMM) and density-based (DBSCAN) approaches were used to cluster and analyze the lane-changing trajectory data. The trajectory distribution and motion pattern of the lane-changing process were obtained. Meanwhile, the four methods were compared and evaluated, and the optimal lane-changing mode was obtained according to the actual driving situation. The results show that the clustering effect of DBSCAN method is the best among the methods mentioned above and the lane-changing motion pattern is the optimum under the condition that the trajectory-similarity measure is based on the Hausdorff distance. The research results can provide a reference for the trajectory prediction and planning of lane-changing for autonomous vehicles.

CICTP