An EMD-SVR model for short-term prediction of ship motion using mirror symmetry and SVR algorithms to eliminate EMD boundary effect

Abstract

Short-term prediction technology has a vital role in improving the efficacy and safety of several offshore operations. Motivated by nonlinear learning ability of support vector regression model (SVR model) and non-stationary data processing ability of empirical mode decomposition (EMD), this study offers a hybrid EMD-SVR model for short-term prediction of ship motion using mirror symmetry and SVR algorithm to eliminate EMD boundary effect. This model is abbreviated as the MSEMD-SVR model in this study. Even though EMD is efficient in dealing with non-stationary data, its boundary effect decreases the prediction accuracy. Raw data are initially processed by improved EMD and then predicted by SVR in the MSEMD-SVR model. This study confirms the negative EMD boundary effect on the prediction accuracy of classical EMD-SVR model and validity of the mirror symmetry method using the rolling and pitching of ship motion data collected during sailing for experiments. Based on the results of contrast experiments, the MSEMD-SVR model is more feasible and reliable for short-term prediction of ship motion than the EMD-SVR model, which does not deal with EMD boundary effect or only applies the mirror symmetry method to deal with.