A Novel Hybrid Model Based on Dual Attention Networks for Significant Wave Height Forecast

Abstract

Extreme waves pose a severe threat to human life and property. Timely and accurate wave forecasting can help humans take appropriate measures in advance to avoid the risks caused by extreme waves. However, it is challenging to accurately forecast ocean waves due to their non-linear and non-smooth characteristics. To overcome this difficulty, we propose a significant wave height prediction method based on feature engineering and dual attention networks. Specifically, in feature engineering, we first decompose the original wave signal by the discrete wavelet transform to obtain several wavelets, after which we add the decomposed wavelets to the original data set for data augmentation, and finally, we use feature selection to determine the features of the final input network. We construct a sequence-to-sequence network with a dual attention mechanism, including the attention at the input layer and the encoder-decoder layer. Extensive experiments are conducted to verify the effectiveness of our method on 24-h and 48-h predictions. The results show that the proposed method outperforms the other methods compared.