Research on Ship Abnormal Behavior Detection Method Based on Graph Neural Network

Abstract

Because of the existing maritime intelligent supervision, ship abnormal behavior detection ignores the interdependence between different time series of ships and different abnormal behavior characteristics, which leads to poor real-time performance and limited detection accuracy of abnormal detection. This paper explores the mechanism of ship abnormal behavior monitoring and detection based on graph attention network. This mechanism uses a sliding window to generate fixed data input, uses graph attention network to capture the dependence between ship behavior characteristics, and determines the threshold of anomaly according to the sea environment set in this paper. In order to complete the online real-time detection of ship abnormal behavior with multivariate and multidimensional time series, improve the accuracy of ship abnormal behavior detection and reduce the blind area of VTS (Vessel Traffic Service) system monitoring.