Significant Wave Height Prediction Using Adaptive Variational Mode Decomposition and Error-Compensation Model

Abstract

Significant wave height (Hs) is an essential parameter for offshore structure design and marine construction planning. Recently, the pre-processing techniques have been widely to improve Hs prediction performance. Among them, although the variational mode decomposition (VMD) is proved to be an effective tool, it is not parameter-adaptive and the manual selection of parameters has strong uncertainty. Therefore, an adaptive VMD, named GAVMD, is designed based on grey wolf optimizer (GWO), attention entropy (AE) and VMD. Moreover, the gate recurrent unit (GRU) and the extreme learning machine (ELM) are integrated as error-compensation model (GRU-ELM) to conduct prediction tasks. By combining the GAVMD and the error-compensation model, a novel hybrid model GAVMD-GRU-ELM is proposed to predict Hs. To validate the proposed model, the measured data of three buoys are used as datasets, and two single models ELM, GRU and a hybrid model GAVMD-GRU are adopted as baselines. The experiment results show that single models can provide suitable results for 3-h and 6-h predictions, but not for 12-h and 24-h. On the contrary, hybrid models can consistently predict well benefiting from GAVMD. In all prediction scenarios, GAVMD-GRU-ELM outperforms other models, which indicates that the error-compensation model can effectively improve the forecast accuracy.

Keywords: Significant wave height prediction, Deep learning, Variational mode decomposition (VMD), Error-compensation