Research On The Dynamic Tension Prediction Method Of Single Mooring System Based On Deep Learning

The single point mooring system of Floating Production Storage & offloading(FPSO)provides horizontal restoring forces for the vessel,and realizes the positioning purpose of limiting the vessel motion in a certain range in the marine environment.However,the existing coupling analysis techniques of hydrodynamic and mooring are mostly used to predict the vessel motion and mooring dynamic tension under certain wave and current conditions.For the actual offshore engineering with random changes,the calculation of mooring forces can only rely on the static theory such as catenary formula.At the same time,it is difficult to maintain the service life of physical measurement equipments at sea for more than half a year.In order to solve the problem that mooring dynamic tensions can not be obtained accurately,this paper studies the characteristic relationship between vessel’s six degree of freedom(6-DOF)motion curves(surge,sway,heave,roll,pitch,yaw)and mooring dynamic tension curves based on deep learning to carry out real-time prediction of short-term mooring dynamic tensions.In this paper,the method of dynamic tension prediction of mooring system is studied based on deep learning.Firstly,the FPSO and single point mooring system are modeled and analyzed,then the sample database of vessel’s 6-DOF motion response and mooring dynamic tensions is obtained.Based on the three-dimensional potential flow theory,catenary formula and lumped mass method,the hydrodynamic characteristics,static stiffness horizontal restoring force characteristics and time domain coupling mooring dynamic tensions of FPSO and single point mooring system are analyzed by using software AQWA and SESAM.Then the post-processing automation program of time-domain coupled numerical samples for different wind,wave and current environmental parameters is compiled,which greatly shortens the calculation time of sample database and also provides training and testing datas for neural network model based on deep learning.Through the comparative analysis of numerical simulation results,it can be verified that the accuracy and feasibility of the sample database for the 6-DOF motion response and mooring dynamic tensions of FPSO and single point mooring system.Secondly,a long short term memory networks(LSTM)model is built based on deep learning.The data in the sample database are trained under single and multiple conditions respectively,and the optimal mooring dynamic tension coefficient matrix is saved.The LSTM model is used to process the time series data.The feature inputs are six degrees of freedom motion curves,and the outputs are mooring dynamic tension curves and yaw motion curves.During the training process,the LSTM model constantly adjusts the hyperparameter to obtain the minimum loss function and obtains the optimal mooring dynamic tension coefficient matrix.The results show that both single and multi conditions training can achieve small mean square error and mean absolute error.Finally,the mooring dynamic tension coefficient matrix and LSTM model are loaded to predict the mooring dynamic tensions respectively.The results show that the LSTM model can achieve good prediction results in single and multiple conditions,which indicates that the research on the dynamic tensions prediction method of mooring system based on deep learning is feasible.