The invention relates to a four-dimensional track online abnormity detection method based on unsupervised learning. On the basis of the method for measuring the distance between the tracks, the historical four-dimensional track data of the selected take-off and landing airports are subjected to segmented clustering by adopting a density clustering algorithm, and the representative tracks in trackclusters are extracted, so that the inter-pair track model of each take-off and landing airport is accurately established. For the flights flying in real time, according to an inter-flight-path distance measurement method and a flight path model, the flight grouping degree and the flight abnormality probability are calculated, whether the current state of the flights is abnormal or not is judged according to an abnormality threshold value, and the flight path model is updated in real time.