ANOMALY DETECTION IN SENSOR DATA IN APPLICATION TO THE ANALYSIS OF MARITIME VESSEL MOTION
The article describes algorithms for anomalies detection in the sensory data in the application to the analysis of maritime traffic. The modern vessel is equipped with many sensors, continuously recording the performance of its various subsystems. The collection and storage of such information provide the possibilities of using intelligent data analysis systems. These tasks concern both the issues of ensuring maritime traffic safety (analysis and prevention of dangerous maneuvers) and the problems of increasing economic efficiency (increased fuel consumption) for ship-owners. The onboard sensors can generate numerical data with a frequency from ten seconds for telemetry up to one minute for navigational parameters. Given such a large amount of information, it becomes obvious the need for the development of automated decision support systems. Applied areas of such systems can serve tasks of preventing dangerous maneuvers, predicting maintenance, preventing collisions, optimizing fuel consumption. In the article, the application unsupervised learning for the analysis of navigational data (ship coordinates, speed, course, depth, etc.) and an example of predicting of fuel consumption based on regression models are considered. A description of various mathematical approaches and its demonstration on real data is given. In conclusion, the possible development and improvement of the given methods are considered.
Keywords: anomaly detection, sensor data, extreme maneuvering, prediction of fuel consumption.