Research On On-line Prediction Of Floating Offshore Platform Motions Based On LSTM Network

With the continuous expansion of the scale and field of marine development,all marine resource development and marine transportation depend on marine floating structures such as ships and offshore platforms.The marine floating structures on the sea are always in a multi-degree-offreedom motion,which makes their operation and safety severely challenged.Accurate online prediction of the marine floating structure’s motion helps to ensure its safety and improve its operations.For example,the prediction of surge and sway can improve the dynamic positioning performance,the prediction of heave,pitch and roll can give riser motion compensation in advance.Therefore,the study of online prediction of the motion of marine floating structures is of great significance.Based on the LSTM network,the extreme short term on-line prediction method of offshore platform motions is developed for predicting the motions of platform using the wave sequence.A large number of wave sequence data and motion data were obtained by semi-submersible platform model test.The LSTM model is established and trained by using the experimental data,and the motion prediction and analysis were carried out for different test cases.The results show that this model produces excellent performance for the short-term on-line prediction of floating offshore platform motions.When the forecast duration is 12 s,the accuracy of the surge is higher than 95.3%,the accuracy of the sway is higher than95.6%.When the forecast duration is 8s,the accuracy of the heave(180 deg)is higher than 89.7%,the accuracy of the pitch is higher than 84.6%,the accuracy of the heave(90 deg)is higher than 87.3%,the accuracy of the roll is higher than 80.9%.The forecast results of the surge motion show that when the forecast duration is less than 9s,the forecast accuracy does not decrease significantly.When the forecast duration is greater than 9s,the forecast accuracy begins to decline,and when it is greater than 12 s,the decline is accelerated.The forecast results of the heave and pitch motion show that when the forecast duration is less than 8s,the forecast accuracy decreases slowly,and when the forecast duration is greater than 8s,the forecast accuracy begins to decline significantly.Moreover,with high computational efficiency,the calculation time per step is in the order of milliseconds,which is much smaller than the forecast duration,and the online prediction of motions can be realized.