Multivariate data decomposition based deep learning approach to forecast one-day ahead significant wave height for ocean energy generation

Abstract

Significant wave height is an average of the largest ocean waves, which are important for renewable and sustainable energy resource generation. A large significant wave height can cause beach erosion, and marine navigation problems in a storm. A novel data decomposition based deep learning modelling framework has been proposed where Multivariate Variational Mode Decomposition (MVMD) is integrated with Gated Recurrent Unit (GRU) to design the MVMD-GRU model. First, a correlation matrix is established to identify statistically important predictor lags. Next, the MVMD is employed to decompose the predictor lags into intrinsic mode functions (IMFs). The GRU model is then applied to the IMFs as inputs to design the MVMD-GRU framework to forecast one-day ahead significant wave height. Several other benchmarking deep learning models were hybridized with MVMD for comparison purposes. The outcomes suggest that the hybrid MVMD-GRU achieved better accuracy using goodness-of-fit metrics for Hay Point, Townsville, and Gold Coast stations in Queensland, Australia. The results show that MVMD significantly improved the forecasting accuracy of the GRU model in terms of WIE = 0.983, 0.918, 0.983, NSE = 0.932, 0.735, 0.934, LME = 0.978, 0.758, 0.752 for Hay Point, Townsville, and Gold Coast stations. This work is valuable to monitor and manage clean energy resources to optimize sustained energy generation.