Study On Multiscale Ship Motion Extreme Short-term Prediction Model Based On Attention Mechanism

The ship motion caused by sea wind,waves and other factors in the process of sea navigation has an adverse effect on shipborne weapons,radar and other equipment.It is of great theoretical and practical value for the safe and efficient operation of shipboard equipment to apply the extreme short-term prediction of ship motion to the shipboard stable platform to compensate in advance.With this as the background,the ship motion extreme short-term prediction model based on attention mechanism is studied.Firstly,according to the frequency characteristics of ship motion signals,a ship motion prediction model based on the post-multiscale attention mechanism is proposed.The ship motion data is decomposed into different frequency scales by multiscale decomposition,and then LSTM network is used to predict the sub-signals obtained by decomposition,and the final prediction results are obtained by integrating all the LSTM network outputs by the post-multiscale attention mechanism.The experimental results show that the introduction of post-multiscale attention mechanism improves the ability of the model to process ship motion signals,and improves the ship motion prediction performance.Secondly,according to the prediction problem of ship motion under different sea conditions,a ship motion prediction model based on the pre-multiscale attention mechanism is proposed.The pre-multiscale attention mechanism is used to carry out adaptive weight on different frequency scales,so that the LSTM layer pays more attention to the important information and reduces the interference of noise signals,which improves the sensitivity of the model and enables the model to adapt to the prediction problem under different sea conditions.The experimental results show that the prediction model based on pre-multiscale attention can adapt to the ship motion prediction problem under different sea conditions.Finally,according to the increasing complexity of ship motion and the decreasing accuracy of prediction under complex sea conditions,a ship motion prediction model is proposed based on local attention mechanism.The local attention mechanism makes the model eliminate the scales irrelevant to the prediction and retain the important frequency scales,which enhances the ability of the model to extract useful features from the data with noise signals.In addition,in order to avoid falling into local optimum,a two-stage training algorithm is designed based on the model structure to train the model.The experimental results show that prediction model based on the local attention mechanism has better performance in predicting complex sea conditions.