Ship Trajectory Clustering Analysis And Anomaly Detection Based On Machine Learning

ea transportation is an important part of the logistics system.In recent years,the continuous growth of sea transportation trade urgently needs more ships to be put into sea transportation.The increase in the number of ships has brought new challenges to the maritime safety supervision department.Traditional maritime traffic manual monitoring methods are inefficient and lack of specificity.Therefore,based on the original ship trajectory data provided by the Automatic Identification System(AIS),this paper obtains high-quality ship trajectory data through data cleaning,processing and other means.DBSCAN algorithm is used to realize the clustering of ship trajectory and then obtain the ship’s normal motion trajectory.A ship trajectory prediction model based on transformer is proposed,and the normal motion trajectory is taken as the training set training model.Then,the trained ship trajectory prediction model is used to predict the target ship trajectory.By comparing the predicted position of the ship with the actual position of the ship,appropriate threshold value is set to finally realize the abnormal detection of the ship trajectory.The main work of this paper is as follows:(1)AIS data preprocessing.The SQL database was established,the AIS trajectory data in the target sea area was selected by Datagrip,and the AIS data was preprocessed.The Minimum description length(MDL)criterion is used to extract the features of ship trajectory and simplify the compression of ship trajectory,which lays a foundation for clustering and anomaly detection.(2)Extract the ship’s normal trajectory based on DBSCAN clustering algorithm.A feature expansion method of ship trajectory including ship position,speed and heading is proposed.Dynamic Time Warping(DTW)trajectory similarity measurement algorithm is used to achieve ship trajectory clustering based on DBSCAN clustering algorithm.The accuracy of similarity measurement between ship trajectories and the accuracy of clustering effect are improved.(3)Abnormal detection of ship trajectory based on deep learning.Considering the particularity of ship trajectory,a ship trajectory prediction model based on transformer is established.The self-attention mechanism of transformer is used to capture the long-term dependence relationship between ship trajectory information,and the threshold based anomaly detection method is used to realize the comparison and analysis between predicted trajectory data and actual trajectory data.Thus complete the detection of ship track anomalies.AIS ship trajectory data in the sea area near Yantai Port were collected for experimental demonstration.The results showed that this research method could effectively identify ship abnormal trajectory,which provided certain reference significance for maritime traffic supervision and safety guarantee.