AIS Track Clustering Analysis And Abnormal Track Detection

Monitoring and maritime management are of great significance to vessels safety.The mandatory installation of the AIS system and the establishment of coastal VTS have brought great convenience to the maritime supervision department.However,manual monitoring is the main way to monitor marine traffic for the maritime department.It is difficult for some busy ports to rely on manual monitoring to guarantee port safety.In order to monitor the trajectory of the vessels in real time,automatically detect abnormal vessels and prevent dangerous activities at sea,this paper proposed an abnormal trajectory detection based on machine learning.The thesis took AIS data as a sample and combines cluster analysis and machine learning to complete the detection of abnormal trajectories.The main work of this paper can be summarized as the following four aspects.(1)processing of AIS data.Based on the introduction of AIS information,it carried out database storage,extraction,cleaning,data conversion and other processing to obtain the rectangular range of AIS data from Dalian Port to Yantai Port according to the geographic range.Then the complex trajectories were sorted and segmented to obtain many complete trajectories,in order to lay the foundation for subsequent research.(2)A clustering algorithm based on the similarity of trajectory structure was used to realize the cluster analysis of ship trajectories in the sea area near Dalian Port and Yantai Port.Firstly,the complete trajectories were segmented according to the threshold of the azimuth angle to obtain sub-trajectory segments.Based on the analysis of the track direction and distance of the trajectories,the similarities between the sub-track segments were calculated.The clustering algorithm based on the similarity of the trajectory structure was used to realize the clustering of vessel trajectories in the rectangular range from Dalian Port to Yantai Port.In this paper,the obtained clustering results were analyzed.Through experiments,trajectories in the set waters are clustered.According to the clustering results,the trajectories in the waters near Dalian port and Yantai port were divided into six categories.(3)Classifying the trajectories of the second class to distinguish between abnormal trajectories and normal trajectories based on the determination method of abnormal trajectories,and then extracting the trajectory features.Firstly,this paper determined the abnormal trajectory points according to the rules of the abnormal trajectory points,and then sorted the ratio of the abnormal points of the trajectories to from large to small.According to the order of sorting,the top N%of the trajectories were abnormal trajectories and the labels were 0.The rest were normal tracks,the labels were 1.After trajectory classification,feature extraction of the trajectory was performed,the selected area was meshed,and the trajectory data was converted into image feature.This paper discussed the threshold N%,taking N equal to 3,5,10 and 20,respectively,to obtain different numbers of abnormal trajectories and obtain different detection results through abnormal trajectory detection.(4)Finally,an abnormal trajectory detection based on supervised learning was proposed.During the experiment,the training data and test data were substituted into the broad learning system and the convolutional neural network for training to obtain the prediction result of the test data,and then the predicted labels were compared with the real labels.The experimental results show that the recognition accuracy of the broad learning system was higher than that of the convolutional neural network,and the time of the broad learning system model was significantly less than that of the convolutional neural network model.