Hybrid Approach Combining SARIMA and Neural Networks for Multi-Step Ahead Wind Speed Forecasting in Brazil

Abstract

This paper proposes a hybrid approach based on seasonal autoregressive integrated moving average (SARIMA) and neural networks for multi-step ahead wind speed forecasting using explanatory variables. In the proposed model, explanatory variables are first predicted, and wind speed forecasting is performed taking into account these forecasted values and wind speed historical series. The multi-step ahead forecasting is achieved recursively, by using the first forecasted value as input to obtain the next forecasting value. The proposed approach is tested using historical records of meteorological data collected from two real-world locations in Brazil. In order to demonstrate accuracy and effectiveness of the proposed approach, the results are compared with other techniques, such as neural networks, SARIMA, and SARIMA + wavelet. Simulation results reveal that the proposed hybrid forecasting method outperforms these popular algorithms for different forecasting horizons with higher accuracy.

Keywords

• SARIMA model
• explanatory variable selection
• multi-step ahead
• neural networks
• wind speed forecasting