基于深度学习的舰船运动参数长时预测方法OA

To address the challenge of accurately predicting long-term ship motion due to its nonlinear and stochas-tic characteristics,this study investigates prediction methods for multi-dimensional ship motion parameters using various deep learning models.A dataset of six-degree-of-freedom(6-DOF)motion parameters was first constructed by collecting sensor data from a 20 t vessel.Subsequently,five deep learning models LSTM,Dlinear,PatchTST,FPT,and AutoTimes were employed to develop long-term prediction models for ship roll,pitch,and heave.The performance of these models was evaluated using mean squared error(MSE)and mean absolute error(MAE).The results demonstrate that among the five models,the one based on the AutoTimes method achieved the best perfor-mance,accounting for 61.11%of the minimum error indices.Further analysis revealed that increasing the length of the historical look-back window(which has an optimal value)and embedding textual timestamp tokens can both effectively enhance the model's prediction accuracy.The findings of this study offer a valuable reference for the se-lection and optimization of high-precision models for long-term prediction of ship motion.