Improving prediction performance of significant wave height via hybrid SVD-Fuzzy model

Abstract

Significant wave height forecasting is an important aspect of many ocean related engineering and scientific phenomena. However, due to the complicated and chaotic nature of ocean waves, in general, data in hand has to be decomposed into its deterministic and stochastic features before feeding into the predictive model. The present study introduces a refined singular value decomposition (SVD) based algorithm for decomposing raw data into its hierarchally energetic pertinent features. SVD-fuzzy model is developed via hybridizing Adaptive Neuro-Fuzzy Inference Systems (ANFIS) model with the proposed algorithm. SVD-fuzzy model is compared with stand-alone fuzzy and combination of a widely used wavelet transformation and fuzzy logic (wavelet-fuzzy) models in predicting SWH in five distinctive Pacific Ocean locations for future lead times 1 h, 2 h, 3 h, 4 h, 5 h, 6 h, 12 h, 18 h and 24 h from two previous hourly SWH data. According to mean square error (MSE), the Nash-Sutcliffe coefficient of efficiency (CE), coefficient of determination (R2), and mean absolute error (MAE) model evaluation metrics, it is found that SVD-fuzzy model outperformed the stand-alone fuzzy model for all lead times and data (Maximum CE of SVD-Fuzzy model is 0.991). Further, SVD-fuzzy model results compare well with those of Wavelet-Fuzzy model. The outcomes of this study indicate that the integrated SVD-fuzzy modeling approach, with no underlying assumptions a priori, is a promising principled tool for its possible future applications not only for coastal and ocean problems but in engineering and scientific disciplines where future state space prediction of a complex and stochastic dynamical system is significant.