Multi-step-ahead significant wave height prediction using a hybrid model based on an innovative two-layer decomposition framework and LSTM

Abstract

As waves are being developed as a renewable energy source, the development of new predictive algorithms to forecast wave height has garnered considerable interest. This study proposes an innovative hybrid model to predict the wave height, including a two-layer decomposition framework and long short-term memory (LSTM). First, the original wave height series is effectively and quickly decomposed into multiple sub-sequences using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). The refined composite multiscale entropy (RCMSE) is then applied to reconstruct these sub-sequences into high-frequency, medium-frequency, low-frequency, and trend components. The respective LSTM is adopted to predict the medium-frequency, low-frequency, and trend components and obtain the sub-results. Subsequently, a series of modes is obtained by the second decomposition of the high-frequency component with variational mode decomposition (VMD), and the sub-result is obtained by forecasting the modes with ensemble LSTM. Finally, we employ the ensemble LSTM to predict all sub-results and obtain the final wave height prediction result, rather than simply adding up all the sub-results linearly. The proposed hybrid model is tested geographically at two buoy stations in eastern New York and Gulf of Mexico. The results show that the proposed hybrid model is more accurate than other benchmark models.