Research On Anomaly Detection Methodof Ship Navigation Trajectories Based On Data Mining

With the rapid development of inland waterway transport,the safety of inland waterway ship has become an urgent problem to be solved.There are ship collision,grounding and other water traffic accidents in the inland waterway every year.The ship trajectory anomaly detection,can detect the ship who deviate from the normal track of the ship,so as to improve the navigation safety and reduce accidents.In this thesis,oriented to the feature of inland waterway,according to the historical AIS data to establish the normal trajectory model of ship navigation,using data mining technology to detect the abnormal trajectory of the ship.The concrete research content and innovation are as follows:1)The trajectory similarity measurement method ofmulti characteristic distance was studied,and the normal trajectory of the ship was established.Calculate their distance from three aspects of ship sub trajectory position,course and speed,and a weighted sum of the formation of multifeature distance of the ship trajectoryin order to accurately measure the similarity between the sub trajectoryof the ship.On the basis of this,the multi feature fast clustering algorithm is designed to get the normal trajectory data of the ship.2)Based on the multi feature clustering method of ship trajectory anomaly detection was proposed.Based on the normal trajectory data of the ship,the model of the normal trajectory of the ship was built.In the ship trajectory outlier detection,from two aspects considered inland waterway characteristics:(1)considering the width of the inland waterway,the location of anomaly detection model was constructed and the effect of channel width on the anomaly detection is reduced by calculating sampling pointsmedian.(2)according to the characteristics of inland waterway,considering the course and speed to construct course and speed of the anomaly detection model,so that it can better conform to the characteristics ofupper channel(heading angle is larger)speed slow,Launching channel(heading angle is smaller)speed fastly.Finally,the feasibility and effectiveness of this anomaly detection method was verified by using the AIS data of the QingShanJiawaterway.3)In order to study the K value of the single grid's GMM(Gauss mixture model)and the denoising method,the IMCFSFDP algorithm wasproposed.Based on the CFSFDP [37]algorithm,the improved algorithm of IMCFSFDP was proposed:(1)clustering center selection of CFSFDP algorithm depends on the decision diagram,designed the additional function ?to automatically determine the cluster center,in order to get effective K value of the GMM model;(2)according to the characteristics of ship trajectory data in the grid,Using the local density formula has been improved,reducing the probability of the local density conflicts(local density identical);Using the formula of minimumdistance with high density has been improved,to avoid multiple clustering center in a cluster;designedthe additional function?,CFSFDP algorithm is solved because of the difference between clusters is too large(in the opposite direction)can not be effective denoising problem.4)Based on the improved GMM,ship trajectory anomaly detection was proposed.On the basis of the research content(3),the Gauss mixture model of normal locus point of the ship was constructed on each grid.In anomaly detection,based on the point of the ship trajectory anomaly detection algorithm was designed,by calculating the trajectory of the abnormal rate todetect abnormal of the ship trajectory.Finally,using the AIS data of the QingShanJiawaterway,the experiment verified the effectiveness and feasibility of the anomaly detection method based on the improved GMM.