ANOMALY DETECTION BASED ON MULTI-DIMENSIONAL FLIGHT TRAJECTORY PROFILE

With the development of aviation safety management, research has gradually shifted from learning from past accidents to actively identifying safety hazards during routine operations. How to effectively identify abnormal trajectories from the massive data is still an open question. To identify abnormal flight trajectories, this research proposes a novel method of trajectory representation based on three-channel images. We attempt to model the latitude, longitude, flight level and ground speed of the aircraft as pixel information of the image using semi-annual Automatic Dependent Surveillance-Broadcast (ADS-B) flight trajectory data from Hong Kong International Airport. A Deep Convolutional Autoencoder (DCAE) is utilised to extract lowdimensional feature representations of image-based trajectories, and the Gaussian mixture model (GMM) clustering method is performed for similarity and anomaly detection. The results indicate the DCAE model has good performance in trajectory feature extraction and abnormal trajectory recognition, which provides ideas for multi-parameter trajectory prediction and multi-dimensional meteorological image fusion.