Wave height forecast method with multi-step training set extension LSTM neural network

Abstract

As the main loading parameter in offshore facilities operations, the accurate prediction of wave height is of great importance. In order to solve the problem of wave height prediction with complex and changeable characteristics, the present paper establishes a wave height forecast model based on prototype monitoring data and multi-step training set extension Long-Short-Term Memory (LSTM) neural network. Firstly, the correlation analysis of wind speeds and wave heights is carried out based on prototype monitoring data. Then, a one-step-ahead wave height forecast method is established based on LSTM neural network. Different prediction models of different time intervals (t = 0.5h, 1h, 3h, 6h, 12h) are built to verify the accuracy. Prediction results indicate that with the increase in time intervals (such as 6h and 12h), the accuracy of the one-step-ahead wave height prediction method decreases rapidly. Hence, aiming at the low prediction accuracy for the large time interval, a multi-step training set extension LSTM model for wave height prediction is presented. Multiple sub-step wave height prediction data are used as the known quantities to improve the accuracy at the final prediction time. The influence of different sub-step step sizes is considered in the presented model. Simulated results show that the 6h prediction error is reduced by 45.83% compared with the one-step model and the 12h prediction error decreases by 52.83%.