In terms of the accuracy and real-time requirements of ship trajectory prediction, the characteristics influencing the ship navigation trajectory are explored at the data level, the network input is determined through the correlation analysis, and a ship navigation trajectory prediction model combining the Recurrent Neural Networks - Long Short Term Memory (RNN-LSTM) is proposed. The network model is trained through the relevant data of ship Z-shape test and actual ship navigation data, and the prediction of future ship navigation trajectory is made. The prediction value of future trajectory is compared with the actual value. The results show that the error of model prediction is small, which verifies the practicability and effectiveness of the scheme in the ship trajectory prediction.