Near real-time significant wave height forecasting with hybridized multiple linear regression algorithms

Globally, major emphasis is currently being put in utilization and optimization of more sustainable and renewable energy resources, to overcome the future energy demand issues and potential energy crises due to many socioeconomic factors. A near-real-time i.e., half-hourly significant wave height (Hsig) forecast model is designed using a suite of selected model input variables where the multiple linear regression (MLR) model, considering the influence of several variables, is optimized by covariance-weighted least squares (CWLS) estimation algorithm to generate a hybridized MLR-CWLS model with a capability to forecast 30-min ahead Hsig values. First, a diagnostic statistical test based on the correlation coefficient is performed to determine relationships between inputs denoting historical behaviour and the target (Hsig) at one lag of 30-min (t – 1) scale. Subsequently, the data are split into training and testing subsets, following a normalization process, and the MLR-CWLS hybridized model is then trained and validated on the testing dataset adopted from eastern coastal zones of Australia that has a high potential for wave energy generation. Hybridized MLR-CWLS model is benchmarked against competing modelling approaches (multivariate adaptive regression splines-MARS, M5 Model Tree, and MLR) via statistical score metrics. The results show that the hybridized MLR-CWLS model is able to generate reliable forecasts of Hsig relative to the counterpart comparison models. The study ascertains the practical utility of the hybridized MLR-CWLS model for Hsig modelling with significant implications for its potential application in wave and ocean energy generation systems, and some of the other renewable and sustainable energy resource management.