Abstract: Objectives Aiming at the limited generalization ability and lack of interpretability of decision results in existing intelligent ship navigation decision systems, research on the application of large model technology is conducted focusing on the tasks of open-sea ship autonomous collision avoidance, centering on both data and algorithms. Methods Firstly, ship encounter scenarios are extracted from AIS navigation data to establish a comprehensive scenario library. Secondly, the scene-instruction mapping module converts scene information into descriptions, providing the foundation for the large model. By analyzing key components of ship navigation decision-making tasks, we adopt a hierarchical reasoning framework comprising cognition, analysis, and decision-making, and propose a large model tailored for ship navigation decision tasks. This model achieves progressive capabilities in navigation scenarios, generation of driving decisions, and planning of navigation trajectories. Finally, both qualitative and quantitative experiments were conducted to validate the proposed method. Results The model demonstrates high-level navigation scenario comprehension through its performance in scenario-based QA tasks. For the task of decision-making and trajectory planning: direction action classification achieves an F1-score of 0.92 , speed action classification attains an F1-score of 0.82, and the trajectory planning maintains sub-10-meter errors. These quantitative results demonstrate the model's effectiveness in ship navigation decision-making and path planning.Conclusions The results indicate that the proposed method provides a novel technical approach for the development of ship autonomous navigation.