Research On Ship Behavior Analysis And Anomaly Detection Based On Multi-source Spatiotemporal Data

With the development of shipping industry and marine monitoring technology,the multi-source spatiotemporal data of marine traffic has increased rapidly.The regulatory requirements of maritime authorities cannot be met only by traditional manual guard.At present,most situation awareness systems simply display the collected multi-source spatiotemporal data without exploring the potential knowledge behind the data.Problems such as low association degree,shallow mining depth and insufficient intelligence level exist in the utilization of multi-source spatiotemporal data.Therefore,based on multi-source spatiotemporal data,it is of great research significance to analyze ship behavior by using data association and mining techniques,so as to realize the key supervision of specific types of ships and abnormal ship behavior by maritime authorities.This paper studies how to make use of the advantages of multi-source spatiotemporal data to improve the situational awareness of ships in relevant sea areas."Situational awareness" includes two meanings:one of which focuses on ship static characteristics,and mainly studies ship classification based on trajectory association and behavior analysis;Another meaning focuses on ship dynamic characteristics,and mainly studies the detection of abnormal ship behavior from two aspects:law learning and domain knowledge.The main contents and innovations are as follows:Firstly,the ship classification method based on trajectory association and behavior analysis is studied.Specifically,aiming at the problem that the classification model is difficult to build due to the insufficient perception dimension of radar spatiotemporal data,the trajectory generation and association algorithms are designed.The algorithms realize the effective association of multisource spatiotemporal data by means of attribute constraints,double projection curve fitting and time registration,so as to expand the perception dimension of radar spatiotemporal data.In addition,in the training stage of the ship classification model,in order to solve the problem of class imbalance of ship trajectory data,a heterogeneous ensemble learning method based on EasyEnsemble and SMOTE is proposed.This method can achieve class balance without information loss and reduce the impact of noise on the final classification results.The experimental results show that the proposed method can improve the accuracy of ship classification and realize the recognition of ships of minority class.Secondly,the detection method of abnormal ship behavior based on law learning is studied.Aiming at the problems of low efficiency and difficult to quantify the effect of anomaly detection in traditional methods,this paper presents an abnormal ship behavior detection method(eD-LW method)based on emd-DBSCAN algorithm and LSTM network with weighted loss function.Specifically,the emd-DBSCAN algorithm is used to cluster ship trajectories.The algorithm takes the trajectory as a whole and measures the difference between trajectory behaviors based on the earth mover’s distance,so as to obtain more reasonable and accurate anomaly identification results.Then the LSTM network with weighted loss function is used to train the abnormal behavior detection model.This method can make full use of the timing characteristics of trajectories,and improve the detection effect of the model through the adjustment of the weight in the loss function.The experimental results show that the eD-LW method can realize the quantitative evaluation of detection effect,and can effectively improve the detection efficiency on the premise of ensuring the detection accuracy.Thirdly,the detection method of abnormal ship behavior based on domain knowledge is studied.Traditional methods mainly rely on AIS data for anomaly detection,but AIS device has the phenomena of intentional closing and malicious data spoofing,which leads to the problem of poor reliability of anomaly detection.Combined with maritime domain knowledge,this paper analyzes and detects the abnormal behaviors such as AIS closing and data spoofing.Specifically,based on the distance between trajectory points and D-TRAP,trajectory association is carried out from the perspective of trajectory points.This method can improve the association effect between multi-source spatiotemporal data during AIS illegal operation,and then the corresponding detection algorithms are designed by establishing the associated map.At the same time,in order to accurately locate the abnormal behavior of ships after the closure of AIS,the abnormal speed behavior location algorithm combining rules and clustering and the abnormal direction behavior location algorithm combining partition and the earth mover’s distance are proposed.The location algorithms can make use of the characteristics of abnormal behavior to reduce the impact of behavior component fluctuation on anomaly location.Experimental results show that the proposed algorithms can make full use of multi-source spatiotemporal data to effectively detect AIS illegal operations and other abnormal behaviors,and can accurately locate the abnormal behavior after the closure of AIS.