Wind speed prediction using empirical wavelet transform and bidirectional gated recurrent unit based hybrid model

Abstract

Accurate forecasting of wind speed is crucial for optimal extraction of energy, enabling integration of power grid, planning and management of renewable energy resources. To overcome the unpredictability of long-term trends and seasonal variation of wind, this study proposes a deterministic framework utilizing a hybrid model based on empirical wavelet transform (EWT), bidirectional gated recurrent neural network (BiGRU), and Bayesian optimization algorithm (BOA) for an hour-ahead wind speed prediction. Firstly, the EWT is used for preprocessing the wind speed data with enabling wavelet charaterticstics. Then, the BiGRU model is employed for regression using optimal values determined by the BOA method. The robustness of the proposed integrative framework is regressively evaluated over seven years (2015-2021) of hourly wind speed data across four locations in India. The evidence of numerical results of the proposed model demonstrates its effectiveness with a maximum improvement of 70%−80% in terms of RMSE values across all the studied regions. Furthermore, the model evaluation and pictorial results indicate that the proposed model is a potent tool for generating wind energy and its integration into the smart grid.