Forecasting of significant wave height using Long Short-Term Memory and Multivariate Variational Mode Decomposition

Abstract

 forecasting the height of the significant wave (SWH) accurately and reliably is crucial for maritime and engineering purposes. This work designs a novel deep learning system to forecast the daily significant wave height (SWH). This work combines two techniques: Multivariate Variational Mode Decomposition (MVMD) and Long Short-Term Memory (LSTM). The oceanic time series data are decomposed into intrinsic mode (IMFs) functions using the MVMD technique. The LSTM uses the IMFs as inputs. The hybrid  MVMD-LSTM model produces is applied to  two stations Albatross Bay and Gladstone Queensland, Australia's stations. The outcome indicate that the proposed hybrid model  improved  the SWH forecasting, it scored acceptable performance  NS_E=0.09999,NS_E=0.99999, WI_E=0.9999 ,WI_E= 0.999998,LM_E=1.7995, LM_E= 2.3894 for Albatross Bay and Gladstone stations respectively. This work is essential for monitoring and managing clean energy resources to maximize sustainable energy production.