The Combined Prediction of Wavelength and Wave Period Based on Wave Dispersion Relation and LSTM Algorithm

Disastrous waves often bring serious economic losses and casualties. Accurate and rapid prediction of sea conditions has an important impact on ship hedging, berthing and operation safety during disastrous waves. The existing wave prediction mostly takes wave height as the main index, and rarely consider the influence of wave period and wavelength on the navigation safety of offshore buildings and ships. Waves with larger wavelengths and periods have stronger penetration, which not only enter the port with more energy, but also may cause harbor resonance, affecting the mooring stability conditions and the number of days that berths that can be operated. In severe cases, it may even lead to the moored vessels accident.

In this study, based on the wavelength and period data simulated by the SWAN wave model, the LSTM-DR model is established to predict the wavelength and period by adding the wave dispersion relationship to the loss function of the LSTM algorithm.The loss function consists of three parts, which are the RMSE of wavelength and period and the error of dispersion relation. The model obtains the optimal simulation results by adjusting the proportion of the three in the loss function.The model was used to input 3 months, 6 months and 12 months of data ( 50 % for training and 50 % for verification ) for sensitivity experiments, and the calculation results were compared with the LSTM model. The results show that the shorter the input data, the more significant the accuracy of the time series prediction results, especially in the coastal water, the correlation coefficient, RMSE and MAPE are significantly improved. This shows that adding physical constraints to the artificial intelligence algorithm can effectively improve the accuracy of the prediction results under the condition of limited data.