Ship Trajectory Outlier Detection Based On AIS Data And Recurrent Neural Network

In order to solve the engineering applicability problem of AIS(Automatic Identification System)trajectory data in the field of maritime supervision and application,this paper takes" Ship Trajectory Outlier Detection Based on AIS Data and Recurrent Neural Network" as the topic,and explores a new method of anomaly detection driven by AIS trajectory data.The research work could guarantee the safety of vessel navigation through improving the maritime supervision ability and provide the support for data science in the field of maritime engineering applications.This research work includes 4 key aspects:quality management of AIS trajectories,similarity measurement of ship trajectories,trajectory density clustering and ship trajectory outlier detection.The research employs a framework consisting of data preprocessing,pattern recognition and pattern based anomaly detection.That is to say,in our research the AIS data is cleaned up,besides,the pattern of ship trajectories is extracted by density clustering algorithm,then ship trajectory outlier dectection is carried by trained nerual network.The objective of this thesis is to develop a set of application method for ship trajectory outliers detection considering the character of AIS data in the water traffic,and mine the valuable information hidden in the historic ship trajectory data,which may improve the maritime supervision.For the fact that there are a lot of errors in the original dataset of AIS trajectory and the lack of systematic consideration of AIS trajectory quality,this paper summarizes a set of quality dimensional system consisting of physical integrity,spatial logic integrity and time accuracy by observing,processing and analyzing a large number of real AIS trajectories data.Besides,A universal trajectory data cleaning method is proposed for the errors in spatial trajectory data and a processing method using the generation time is proposed to correct the recorded time.Similarity measurement and parameter determination is the key step in the clustering process.In the term of similarity measurement,AIS ship trajectory is a typical asymmetric time series data.For the lack of considering water traffic situation factors in ship trajectory similarity measurement,this paper proposes an improved method based on DTW distance(Dynamic Time Warping).The method considers the local trend characteristics of the ship trajectory and the displacement information of the ship's motion.In the term of parameter determination of DBSCAN algorithm(Density-Based Spatial Clustering of Applications with noise),The experimental data in the ship trajectory data mining has the characteristics of large volume,complex distribution and high dimension.Therefore,in the ship trajectory clustering analysis,the parameters of DBSCAN algorithm are particularly difficult to determine.Aiming at the problem of parameter selection of DBSCAN algorithm in water traffic environment,this paper proposes a parameter determination method based on statistical distribution.the proposed method has the characteristics of simple implementation and less human intervention,which means the proposed method is more suitable for application in ship trajectory clustering.In the research of ship trajectory outlier detection,considering the influence of low quality trajectory data on accuracy and the influence of excessive parameter setting on implementation efficiency,this paper proposed a method that combine the DBSCAN algorithm and recurrent neural network,and the detection process is converted into a simple judgment of geographical distance,avoiding the complicated parameter settings.The proposed application method has the characteristics of high detection accuracy and insensitivity to low quality trajectories in the normal model.Besides,based on the AIS real ship trajectory from the water area in Ningbo-Zhoushan harbour(Jan.2015),the verification experiment is carried out.The experimental results verify the effectiveness of the proposed method in improving the quality of AIS trajectory,normal pattern extraction and identifying abnormal behavior of ships.This research has practical significance for improving the maritime supervision ability and ensuring the safety of ships sailing at sea,and provides an important theoretical basis for the application of data science in the maritime field.