Long-time prediction of sea wave trains by LSTM machine learning method

摘要

The present work proposes a novel way of long-time accurate prediction of sea wave trains using the Long ShortTerm Memory (LSTM) method. Typical kinds of regular waves (e.g. the Stokes wave), irregular waves (produced from the PM wave spectrum by either the equal frequency method or the equal energy method) and real sea wave trains are considered. The historical data of wave elevations are used in the training process of the LSTM model, based on which future wave evolutions are predicted. To improve the accuracy of long-time estimation of irregular waves, a multi-shot prediction method is proposed. The validity of the proposed approach is proved by comparing the performance of the single-shot prediction method and the multi-shot prediction method in multiple cases of irregular waves. In the meantime, the effects of the input and output lengths are investigated. Test results demonstrate that, although the standard LSTM scheme with the single-shot method presents appealing accuracy in forecasting regular waves, the multi-shot method is an essential component in enhancing the accuracy of long-time prediction of irregular waves.