Wind Forecasting:Hybrid Statistical and Deep Neural Network Approaches

High penetration of renewable energies in the electrical grids require forecasting information for their proper integration. This necessitates the robust forecasting techniques in terms of faster speed and high accuracy. There is a large number of statistical and machine learning methods available for forecasting, but individually their speed and accuracy is lesser than the hybrid methods. This paper proposes two-hybrid methods ARIMA-SVR and ARIMA-RNN which are very fast and accurate. These methods are tested on a wind farm dataset of two zones, and hourly predictions are performed on these. Results obtained have shown that ARIMA-SVR hybrid architecture outperformed the ARIMA-RNN for the selected two wind farm zones.