A Flight Trajectory Prediction Method based on Trajectory Clustering

Trajectory prediction is the core technology of next generation air traffic system, flight data mining can offer great favor to implement of trajectory prediction. In this paper, a trajectory prediction method based on space-time clustering of historical flight data is proposed. Firstly, trajectories are sampled based on the key points and expressed as sets of sample points. Then, cluster sample points by using local density based spatial clustering algorithm with noise (LDBSCAN). LDBSCAN combines local outlier factor (LOF) with density based spatial clustering algorithm (DBSCAN), it can erase the influence of outlier sample points. After removing outlier points' information, the rest of sample points can represent normal trajectories' regular pattern. Fuse sample points as a experiential trajectory by using inverse distance weighting (IDW). Finally, considering the existing trajectory information, predict the trajectory in a future time according the experiential trajectory and its error character. At the end of this paper, an experiment shows that this method can erase the abnormal trajectories' influence and predict trajectory accurately..