Prediction of ship following behavior in ice-covered waters in the Northern Sea Route based on hybrid theory and data-driven approach.

The Northern Sea Route (NSR) is perennially ice-covered. As shipping demand increases, the inherent theoretical following model is difficult to describe the increasingly complex ship-following behavior under icebreaker escort operation. This paper aims to establish an innovative model, employing a hybrid approach that integrates theoretical principles and data-driven methodologies. Key factors of sea ice, such as the thickness and density, are comprehensively considered. The model's foundation is grounded in a backward-looking velocity difference framework, accounting for the motion states of preceding and trailing ships. This framework guides the design of the data-driven model, enhancing its interpretability. A deep learning-based long short-term memory network is employed to formulate a ship-following model, tailored to the challenging ice-infested regions. Moreover, an attention mechanism is introduced to augment the model's capability to focus on and extract critical information from the input sequences of ship-following behaviors. A case study including two-ship and multi-ship following scenarios demonstrates good model performance with a mean absolute percentage error (MAPE) of less than 2 %. This research brings an important advancement for predicting ship-following behaviors in the NSR, which could be used to enhance navigation safety and efficiency in the Arctic region. • A ship following model is proposed to improve navigation safety and efficiency in the NSR. • Theory-driven and data-driven approaches are combined in the proposed model. • Ship speed and course during formation are identified in ice-covered waters. • Case studies in two ships and multi-ships scenarios are conducted to verify the model strength. [ABSTRACT FROM AUTHOR]