Forecast And Application Of Pulsating Source And Translating Source Green Function Based On Machine Learning

Free-surface Green function is the basis to solve the marine hydrodynamic problems with boundary element method.For hydrodynamic calculation problem,it is a key issue to figure out the way of calculating Green function and its derivatives accurately and quickly.In this paper,for the radiation problem with zero forward speed,we calculate and establish a high-precision database for the pulsating point source Green function of dimensionless expression,together with the Kelvin source Green function in the steady wave-making problem.A deep learning function library Keras is used to train and study the database,a neural network forecasting model is further established,and both the accuracy and the efficiency of the global and local forecasting are discussed.It is shown that the accuracy of the pulsating point source Green function predicted by the machine learning model could be guaranteed.The efficiency of the neural network forecasting model is higher than that by the numerical integral method while it is lower when compared to the analytic functionbased polynomial approximation method.Based on the previous machine learning model of pulsating point source Green function,its application in hydrodynamic calculation is further discussed.In this paper,numerical simulation and prediction is performed to calculate the hydrodynamic coefficient and the first order wave force of the hemispherical buoys oscillation in regular waves,the improved Wigley ship and the S175 container ship.And the calculation results of hydrodynamic coefficient and motion response are compared with the results of WAMIT.The influence of the calculation accuracy of the Green function with the speed correction method is studied for the S7-175 container ship,and the calculation grid is densified to explore and analysis the influence of the amount of grid.The research shows that the prediction accuracy of the machine learning model can satisfy the calculation needs of the hydrodynamic coefficient with zero forward speed.The accuracy of the machine learning model with speed correction method can meet the calculation needs of the hydrodynamic coefficient.After calculation grid of the hull is densified,the calculation deviation of the hydrodynamic coefficient and the motion response is further reduced,especially the calculation results of the hydrodynamic coefficient are completely consistent,indicating that the influence of the accuracy of the Green function will be reduced after the calculation grid encryption.The neural network prediction model of the pulsating point source Green function with 3d-5d precision learned in this paper can meet the needs of actual hydrodynamic calculations,and has certain engineering application value.The neural network method above provides a new idea for improving the efficiency of hydrodynamic calculation and solving traditional calculation problems.