A new approach to predict dynamic mooring tension using LSTM neural network based on responses of floating structure

The prediction of dynamic mooring line tension using the responses of floating structures is of great significance for the safety monitoring of the station-keeping. Due to the mooring line tension usually has the typical lowfrequency and wave-frequency characteristics, this paper proposes a method of Low-frequency adds wavefrequency responses (LAWR) to predict the mooring line tension. Firstly, the optimal Long-short term memory (LSTM) neural network structure is determined through conducting the parameter sensitivity analysis on the performance of the LSTM model. Secondly, the low-frequency and wave-frequency tension are predicted using the corresponding low-frequency and wave-frequency responses as the input data, respectively. Then, the total mooring line tension is predicted by superposing the low-frequency tension and wave-frequency tension. Finally, the feasibility of the LAWR method is verified by considering the different data relevance between the training sets and validating sets, including the sea state condition contains different current direction, wave height and spectral peak frequency. The method proposed in this paper could further improve the prediction accuracy of mooring line tension more than 30%, which has great engineering significance.

Automated summary

This paper proposes a method to predict mooring line tension by using low-frequency and wave-frequency responses. By optimizing the neural network structure and using corresponding responses as input data, the low-frequency and wave-frequency tensions are predicted separately and then superposed to obtain the total tension. The feasibility is verified by considering different data relevance, and the method can significantly improve the accuracy of tension prediction.