Research On Very Short-Term Prediction Of Ship Motion Attitude Based On Deep Learning

When a ship operates at sea,it is affected by uncertainties such as wind,waves and currents,which generate very complex nonlinear motions in its six degrees of freedom directions.These motions are coupled with each other and act together on the ship,making the ship always in an irregular motion.The irregular motion will make the ship’s operation much less efficient,safe and comfortable.If the ship’s motion posture can be accurately predicted in advance and compensated,it will help to ensure the safety of operation and improve the operating environment.In this thesis,a very short-term combined prediction model of ship motion attitude based on self-attentive mechanism and long short-term neural networks is proposed for the characteristics that ship motion attitude is difficult to be accurately predicted.First,the ship’s surge,roll and heave data under 15 operating conditions are obtained through the mooring system built in ANSYS/AQWA,and these ship motion data will be used as the training set and test set of the model.Combining the advantages of both Self-Attention Mechanism(SA)and Long Short-Term Memory(LSTM)to build a combined forecasting model(SALSTM),the concept of attention gate is proposed in the combined model.The SALSTM model and the LSTM model were compared in three sets of representative ship motion data,and the attention scores of the SALSTM model and the LSTM model were also analyzed.The results show that in the SALSTM model,the LSTM network is dominated and the self-attentive mechanism is supplemented;the SALSTM model is able to improve the accuracy of the forecasting model by selecting the time period that needs attention through the attention gate and data information,while increasing the attention to that time period so that the model can obtain the key information and ignore the remaining unimportant information.Second,the effects of the magnitude of the environmental load and the variation of the incidence angle on the SALSTM prediction model are discussed to provide a reference for the practical application of the model.The results show that when the ship is in the top or cross wave condition,the change of the incident angle of the environmental load does not affect the prediction model of each degree of freedom of the ship much under the similar working conditions.The results show that when the ship is in the overhead wave or cross wave state,the change of the incident angle of the environmental load does not have much influence on the forecast model of each degree of freedom of the ship under the similar conditions;when the ship is in the overhead wave state,the change of the environmental load magnitude will have a greater influence on the forecast model of each degree of freedom of the ship under the similar conditions,and among the variables of the environmental load,the spectral peak period has the greatest influence on the forecast model,and the meaningful wave height,wind speed and current speed have less influence on the forecast model.Finally,the influence of forecast advance on the SALSTM forecast model is discussed to provide a reference for the practical application of the model.The results show that the forecast accuracy of the forecast model decreases with the increase of the forecast advance,and there is a sudden change in the process of decreasing,and the decreasing trend of the forecast accuracy shows a stepped pattern.The SALSTM forecasting model has a relatively high prediction accuracy for ship’s surge when the prediction advance of the model is within8 seconds,and the forecast accuracy of the sureg forecast model is relatively low when the forecast advance is more than 8 seconds.The SALSTM forecasting model has a relatively high prediction accuracy for ship heave when the forecast advance of the heave forecast model is within 4 seconds,and the forecast accuracy of the heave forecast model is not satisfactory when the forecast advance is greater than 4 seconds.