A hybrid EMD-SVR model for the short-term prediction of significant wave height

Short-term prediction of ocean waves is critical in oceanic operation-related activities. Statistical models have advantages in short-term wave prediction as complex phenomena are substantially simplified. However, conventional statistical models have limitations in forecasting nonlinear and non-stationary waves. This paper develops a hybrid empirical model decomposition (EMD) support vector regression (SVR) model designated as EMD-SVR for nonlinear and non-stationary wave prediction. Auto-regressive (AR) model, single SVR model and EMD-AR model were studied to validate the performance of the proposed model. The wavelet decomposition based SVR (WD-SVR) and EMD-SVR models have been investigated to compare the performances of the EMD and WD techniques. The model performances were evaluated by using time history comparison, root mean square error (RMSE), the correlation coefficient (R), the scatter index (SI) and coefficient of efficient (CE). Significant wave heights data used in the simulations were obtained from National Data Buoy Center (NDBC). Considerable improvements were found in the comparisons among the EMD-SVR and other models. The CE values indicate the EMD-SVR model shows good model performances and provides an effective way for the short-term prediction of nonlinear and non-stationary waves. (C) 2016 Elsevier Ltd. All rights reserved.