Real-time prediction of ship trajectory is important to prevent collision and eliminate potential navigational conflicts, and has significant significance for improving navigational safety. However, the declining accuracy of the long-term prediction and the inadequacy of the model to adapt to the complex situation have seriously affected the reliability of ship trajectory prediction. To overcome this problem, this study proposes an advanced prediction model that combines a Test-Time Training (TTT) layer with a long short-term memory network (LSTM). The proposed TTT-LSTM model utilizes LSTM layers to capture the dependency of time series data in ship movement. Moreover, it achieves real-time parameter adjustments through the integrated TTT layer, enabling the model to maintain both real-time responsiveness and predictive efficiency as new data is received. This paper collects the data of the Automatic Identification System (AIS) in the sea area around Xiamen Port and uses it as an empirical case for model validation. Experimental results show that compared with the baseline model, the proposed TTT-LSTM model has a significant improvement in both prediction accuracy and response time. Therefore, it can provide more reliable navigation decision support, reduce navigation risk and optimize route planning.