Research On Prediction Method Of Ship Motion Attitude Based On Deep Learning

The marine environment is very complex,and phenomena such as sea breeze,waves and ocean currents often occur.Therefore,when a ship is operating on the sea,it will produce six-degree-of-freedom mutually coupled sway such as roll,pitch,and bow.Shake will affect the safety of personnel and cargo on ships.In addition,the electronic equipment and precision used on various ships also have different requirements for the stability of ships.If the ship’s motion attitude information can be predicted in advance for a period of time in the future,the stability and safety of the ship’s operations at sea will be improved,and it will have remarkable military meaning,social value and economic value.In this paper,the ship roll,pitch,sideslip angle and rotation angle velocity are taken as the research objects to study the prediction algorithm of ship motion attitude.The specific content is as follows:Firstly,preprocess the time series of ship motion attitude,and then divide the data set into two parts: training set and test set,which account for 85 percent and 15 percent of the total data set,respectively.Establish the Long-Short Term Memory Network(LSTM)model and set the parameters.According to the one-dimensional input,compare different window lengths in simulation experiments to select the best parameters.Due to the mutual influence between the ship motion attitude data,simulation experiments are used to explore the relationship between the data.Secondly,due to changes of sea breeze,sea waves and other factors,the ship attitude data fluctuates greatly.Under sudden weather and extreme weather conditions,only the LSTM network prediction results are often inaccurate,especially where the error is relatively large at the crests and troughs.This paper combines Complementary Ensemble Empirical Mode Decomposition(CEEMD)with LSTM.CEEMD can stabilize the non-stationary time series,and then use LSTM to predict and reconstruct.It is proved by experiment that the proposed CEEMD-LSTM model has higher prediction accuracy than the LSTM model,and has a better prediction effect on the positions of the peaks and troughs.Thirdly,due to the connection between the data of ship motion attitude,and the multi-dimensional input will bring additional errors,the attention mechanism module is added,and the LSTM is used to extract the temporal features of the ship’s motion posture,and the Convolutional Neural Networks(CNN)is used to extract the spatial features of the ship’s motion posture,so a ship motion attitude prediction model based on LSTM-CNN-Attention is proposed.The attention mechanism realizes the simulation of biological attention through algorithms,and dynamically adjust the weight of input features,so that the model can focus on the most effective information under limited resources,and extract the characteristics of different dimensions of data.It is proved by experiment that the model can get better prediction results.Finally,through the comparison of real ship data prediction experiments with different models,the multi-step prediction of roll,pitch,sideslip angle and rotation angle velocity in the ship’s attitude movement is completed,and the experimental curves and statistical results are given.It is proved by the experiment that the algorithm presented in this paper is effective.