There are a lot of researches for the prediction of ship movements based on Automatic Identification System (AIS) information for applications such as collision and grounding prevention and sea congestion forecasting. However, most of them are for normal time, and it is also important to detect the sea congestion at the time of disaster in order to support the control of ships to their destinations and to prevent collision. During disaster, it is assumed that ship movements are different from normal. However, there are few available data for the prediction during disaster, which makes it more difficult to improve the accuracy of the prediction. Therefore, we have developed a new model to represent the movement of a ship for the prediction at the time of disaster. In this paper, we evaluate the results of 30 minutes ahead and describe that our model showed more accurate result compared with the rule-based model and machine learning model assumed as the benchmark. In addition, our model reproduced the actual congestion situation in some areas and times by predicting the position of each ship. It is suggested that this model can be applied to the decision-making to solve ship congestion in disaster situations.