Machine learning-based prediction of the behavior of marine traffic participants and discovering non-standard marine traffic situations

Scientific research problem – what are the theoretical and practical assumptions to create machine learning methods to ensure safe navigation in marine traffic. These methods are based on analysis of the historical navigation data and predict the behavior of marine traffic participants. After defining safe marine navigation as a research object and substantiating the benefits of the usage of the machine learning methods, the different methods based on machine learning were applied to perform historical data aggregation, modeling, and evaluating situations of vessel maneuvering. The historical marine navigation data aggregation algorithm was proposed and examined. This historical marine data aggregation is intended to facilitate and accelerate analyze of marine data. Two different methods were applied and examined intended to predict the monitored vessels’ future nearest turning point. As well, this method is applied to predict the nearest turning points of the route of the monitored vessel. The proposed and examined method is designed to detect a deviation from the planned or predicted route of the monitored vessel. This dissertation proposed a new complex method for the prediction and evaluation of vessel turning points based on context information. This evaluation is intended to caution about an unusual situation based on the prediction and before any indication. The proposed methods’ possibilities and limitations for practical usage with real-time navigational data are defined and described. Based on the empirical research of the proposed models, created a complex methodology to ensure safe vessel navigation in marine traffic