The problem of ship trajectory prediction has a long-standing history within the broader domain of time series forecasting. Historically, trajectory prediction tasks involved dividing a trajectory into training and testing sets. However, due to the emphasis on the completeness of ship trajectories and their relatively weak periodicity, this approach undoubtedly resulted in information loss as some important patterns and features were not adequately learned. In order to capture the entirety of ship trajectories, we propose a pre-training model specifically designed for ship trajectory prediction. Employing a single-decoder architecture and fine-tuning on specific sub-tasks, this model enables direct input of data segments, allowing for immediate initiation of the prediction process. Our approach not only achieves impressive prediction accuracy but also opens up new possibilities for downstream tasks. The pre-trained model extends its applicability beyond traditional sequential trajectory prediction. By adjusting the downstream linear layers, the model can be tailored to produce a wider array of desired outcomes. This adaptability empowers the model to address a variety of tasks, thereby enhancing its utility in ship trajectory forecasting.