Data-driven framework for extracting global maritime shipping networks by machine learning

Maritime shipping network is essential for ship routing, scheduling, and flexibility analysis of the shipping system. This paper proposes a framework for extracting global maritime shipping traffic networks using automatic identification system (AIS) data based on machine learning methods. The framework consists of berthing area identification, trajectory selection and separation, waypoint area identification, edge generation, and network construction. Simultaneously, a route planning method using the A* algorithm based on a probability-directed graph model is proposed to verify the effectiveness of the maritime shipping network. The real-world global AIS data of bulk carriers in 2018 was used to extract maritime shipping networks to prove the framework. The framework successfully extracts maritime shipping networks containing 2769 berthing areas and 2688 waypoint areas over the world's oceans, and the results demonstrate that the estimated networks can be used to analyze the speed of navigation on edges and the size of flows between nodes. Additionally, along with the estimated shipping networks, distance-based route planning is still more stable even if generated routes considering node connection probabilities usually match the observed trajectories. It is concluded that the proposed framework and methods may help (1) provide a thorough framework to obtain and analyze maritime shipping traffic networks and (2) enrich route planning methods by considering historical navigation patterns.