Design And Implementation Of Real-time Vessel Anomaly Detecting System Based On AIS Data In Maritime Key Area

With the development of global seaborne trade,the number of vessels and the loading of dangerous goods are increasing all over the world,and the potential security maritime issues show a clear rising trend.Automatic Identification System(AIS)has been developed to provide rich data source for vessel anomaly detecting research.How to use modern data mining and huge data processing technology to detect the abnormal behavior of vessels has become a key research content of maritime situation awareness system.However,most of current anomaly monitoring methods judge a real track with normal models generated from historical trajectory or self-defined rules refined from statistics of vessel data.These methods have difficulty to meet the anomaly detecting needs under condition of massive real-time data because of mutually restricted efficiency and accuracy.This paper takes the project named "AIS Data Pretreatment Technology and Vessel Abnormality Detecting System”as the background.In order to solve the problems mentioned above,a method based on DBSCAN algorithm is presented after the discussion of mining maritime key area.On this foundation,through the analysis and extraction of anomaly characteristics from historical AIS data,a method based on logistic regression is introduced to build vessels anomaly behavior model.Finally,a distributed real-time vessel anomaly detecting system is designed and implemented with the knowledge of maritime key area and the vessel anomaly model.The validating work is conducted by using real AIS data.The results demonstrate that the proposed mining method can effectively mine key areas and export information about the areas.It also proves that the vessel anomaly detecting model can achieve high efficiency and prediction accuracy.The system is running on a cluster consists of three machines,it can handle tens of thousands of records per second,and the performance can be easily expanded by increasing work nodes.The comprehensive results indicate that the system has high performance and can provide a better automation detecting service for the supervision department.