Developing a novel hybrid Auto Encoder Decoder Bidirectional Gated Recurrent Unit model enhanced with empirical wavelet transform and Boruta-Catboost to forecast significant wave height

Major emphasis presently being made is on using and optimizing more sustainable and renewable energy resources to tackle the upcoming energy demand challenges and probable scarcity induced by several socioeconomic variables. In this research a new hybrid model: combination of empirical wavelet transform (EWT) and Auto Encoder Decoder Bidirectional Gated Recurrent Unit (AED-BiGRU) was used in forecasting daily significant wave height (Hs) in Emu Park and Townsville on the east coast of Australia. A newly developed CatBoost-Boruta algorithm was used to select important Intrinsic Mode Functions (IMFs) obtained from EWT decompositions. Three different machine learning models, including Random Forest (RF), Boosted Regression Tree (BRT), and Gene Expression Programming (GEP), were compaed with developed model. The input variables include lagged data of maximum wave height (Hmax), zero up-crossing wave period (Tz), peak energy wave period (T), direction (Dir_T), and sea surface temperature (SST). The comparison between single models showed that the AED-BiGRU model had a better performance than others. Decomposing the time series of the input data using empirical wavelet transform and entering them into the models significantly improved their performances compared to single models for both locations. Among the combined EWT-based models, the EWT-AED-BiGRU model performed better than other models (R = 0.9802, RMSE = 0.0815, MAPE = 8.6600 for Emu Park and R = 0.9735, RMSE = 0.0695, MAPE = 10.6596 for Townsville). The new developed model was used to forecast multi-step ahead significant wave height. Results showed that the EWT-AED-BiGRU model can forecast the significant wave height until 10 days with high accuracy