Ship Dynamic Outlier Detection Method Based On Trajectory Data Mining

In order to better monitor,identify and find outlier behaviors of ships such as crossing the channel,speeding,decelerating,etc.,and issue alerts to ships that may have potential accidents to minimize the risk of accidents,we choose the topic "Research on Dynamic Outlier Detection Method" to explore and analyze regional ship navigation data,which is expected to improve ship navigation safety and provide theoretical support for improving maritime supervision capabilities.The purpose of this study is to solve the problem of real-time detection of stray ships to reduce the risk of accidents in the research waters.Therefore,an adaptive density clustering algorithm is proposed to extract the characteristic information of ships in the research waters,such as track,speed,and heading.By studying the Graphics Processing Unit optimization algorithm,we improve the speed performance of ship trajectory similarity measurement and clustering respectively,and greatly reduce the time overhead in the process of extracting the ship’s normal trajectory features.Based on the extracted ship’s normal trajectory,it uses machine learning algorithms to establish a normal and outlier discrimination model for ship track position,speed,and heading,then realizes real-time dynamic outlier detection of ships in the research waters.The main contents of this article are as follows:(1)Multi-feature extraction method of ship normality.The feature similarity measurement between ship trajectories is accelerated by GPU multi-thread parallel computing.The algorithm is combined with the K-mediods clustering algorithm to establish a fast track feature extraction model based on prior knowledge,which accelerates the extraction of location feature data of normal ship in the research waters.After the universality of the adaptive density clustering algorithm is enhanced,the accelerated similarity measurement algorithm combined with the adaptive density clustering algorithm is used to extract the normal ship speed and heading feature data in the research waters.(2)Dynamic Outlier Detection Method for Ship Navigation Behavior.Based on the machine learning method,we selecte five machine learning algorithms to construct the ship dynamic outlier detection classifier,and the seven-fold cross-validation was used to compare the accuracy of each model,and the algorithm with the highest accuracy was selected to form the final dynamic detection model.(3)Validation of dynamic outlier detection method based on sample generationTaking the navigable ships in the confluence of the southern trough of the Yangtze River as the research object,the outlier trajectory samples are generated based on the Monte Carlo idea,and the training data set is formed together with the historical trajectory samples.The data set is judged in real time to verify the effectiveness and rationality of the outlier detection method proposed in this paper.The main conclusions of the research are as follows:(1)The similarity measurement algorithm combined with GPU accelerates the process of K-mediods clustering algorithm for the extraction of normal track features in the study waters.The extraction results are reasonable,and the silhouette coefficient is 0.868.The speed performance of track extraction is improved by 50%,which speeds up the track data input preparation process for the ship dynamic outlier detection method.(2)In the process of using the proposed adaptive density clustering algorithm to verify the extraction of ship characteristics in the research waters,the adaptive density clustering combined with the accelerated similarity measurement algorithm can shorten the time overhead of heading extraction by 30%,and can also extract the normal speed and the heading feature in the research waters.(3)We establish an outlier discrimination model based on machine learning model.The results of the case study show that the optimal model is selected by comparing the methods,and the real-time outlier trajectories can be discriminated with high accuracy,which verifies the effectiveness and rationality of the feature extraction and outlier identification algorithms in this paper.