Study On Dynamic Tension Estimation For The Underwater Soft Yoke Mooring System With LSTM-AM Neural Network

It is significant to monitor the mooring tension for the safety,operation and maintenance of single point mooring system of floating production,storage and offloading unit(FPSO).The existing mooring monitoring methods include: calculation of mooring tension by finite element analysis method,calculation of mooring tension by static methods such as static balance and catenary theory,and installation of tension sensor at the top of mooring line to monitor mooring tension.However,the calculation speed of finite element method is slow,and it is not suitable for the actual sea state with random changes.Many assumptions of static balance method affect the accuracy of mooring monitoring,and the sensor installed below the water surface seriously shorens its service life.In order to quickly and accurately monitor the mooring tension of FPSO,the mooring tension of underwater soft yoke mooring system was studied by using the attentionmechanism based long-short time memory neural network(LSTM-AM).Based on the combination of numerical simulation,attention mechanism and LSTM neural network,a deep learning algorithm is proposed to estimate the mooring tension of the underwater soft yoke mooring system through the motion response of the FPSO.The numerical simulation data of short-term sea state is used as the training and testing data of neural network.Because the motion response of FPSO and mooring tension are different in order of magnitude and unit,the maximum and minimum normalization method is used to process them.A feature extraction method of the first-order and second-order central moment is proposed to enable the neural network model to better learn the features of the input data.Since the motion response of the FPSO has different effects on the mooring tension,the characteristic attention mechanism is added to the LSTM neural network to make the neural network pay attention to surge and sway.The evaluation index of the estimation performance of the neural network model was defined,studied the length of the time window of the neural network model,the number of hidden layers,the number of neurons in each hidden layer and the optimization algorithm,and determined the optimal neural network model for the estimation of the underwater soft yoke mooring system.From the perspective of environmental factors,the estimation intervals of the neural network model for significant wave height,spectral peak period,wind velocity,current velocity,wave direction,current direction,wind direction were determined,and the sea state database of the training set was obtained by dividing the estimation intervals.In order to prove the accuracy of LSTM-AM neural network in estimating the mooring tension of underwater soft yoke mooring system,many different cases have been studied.All in all,the tension of the two mooring legs of the underwater soft yoke mooring system is different when the direction difference exists between the environmental loads.Adding attention mechanism can effectively improve the estimation accuracy of LSTM neural network.The training set is divided according to the estimated interval,which can completely cover the sea conditions that may be encountered in the whole operation cycle of FPSO.Through case analysis,it is proved that LSTM-AM neural network can be used to monitor mooring tension through FPSO motion response.