Wind power forecasting – A data-driven method along with gated recurrent neural network

Effective wind power prediction will facilitate the world’s long-term goal in sustainable development. However, a drawback of as an energy source lies its high variability, resulting challenging study forecasting. To solve this issue, novel data-driven approach is proposed for forecasting by integrating data pre-processing & re-sampling, anomalies detection treatment, feature engineering, and hyperparameter tuning based on gated recurrent deep learning models, which systematically presented first time. Besides, neural network Gated Recurrent Unit (GRU) successfully developed critically compared with algorithm Long Short-term Memory (LSTM). Initially, twelve features were engineered into predictive model, are speeds at four different heights, generator temperature, gearbox temperature. The simulation results showed that, terms forecasting, can capture degree accuracy lower computational costs. It also be concluded that GRU outperformed LSTM under all observed tests, provided faster training process less sensitivity to noise used Supervisory Control Data Acquisition (SCADA) datasets.