基于耦合特征LSTM的船舶运动预报方法研究OA北大核心CSTPCD

Long short-term memory(LSTM)neural network model was used to predict the motion of ship advancing in waves under different sea states.Aiming at the characteristic that LSTM model was difficult to optimize,a coupled feature LSTM neural network model was proposed.First,the ship motion time series data were normalized.Then,the LSTM model with input layer,hidden layer and output layer was built based on the deep learning framework TensorFlow,and the original data were divided according to different features.Finally,LSTM models with different coupling features were used to predict the tested samples.Results show that the six-feature coupled LSTM neural network model has obvious advantages compared with other LSTM models for the prediction accuracy.The motion prediction error is decreased by 2.1%～12.9%in the level-4 sea state.In the level-5 sea state,the motion prediction error is decreased by 2.4%～12.3%.The six-feature coupled LSTM model can simultaneously output six degrees of freedom(six-DOF)motion with only one calculation,which can reduce the calculation time by 51.4%～82.7%and improve the calculation efficiency.