Wind Speed Forecasting Using Statistical and Machine Learning Methods: A Case Study in the UAE

Abstract

Wind energy is a source of sustainable energy which is developing very quickly all over the world. Forecasting wind speed is a global concern and a critical issue for wind power conversion systems as it has a great influence in the scheduling of power systems as well as on the dynamic control of wind turbines. In this research, we deploy and study four forecasting models in order to forecast wind speeds in the city of Abu Dhabi, United Arab Emirates (UAE). Two of these models are conventional statistical methods, namely, (i) Auto Regression Integrated Moving Average (ARIMA) and (ii) Seasonal Auto Regression Integrated Moving Average (SARIMA) models, and the other two are drawn from the field of machine learning, namely, (i) Artificial Neural Networks (ANN) and (ii) Singular Spectrum Analysis (SSA) models. We compare the performances of these four models in order to determine the model which is most effective for forecasting wind speed data. The results show that the forecasting model SSA provides, on average, the most accurate forecasted values compared to the other three models. However, those three models, ARIMA, SARIMA and ANN, offer better results for the first few hours (around 24 h), which indicates that ARIMA, SARIMA, and ANN models are suitable for short-term forecasting, while SSA is suitable for long-term forecasting. The findings of our research could contribute in defining the fitting forecasting model in terms of short-term forecasting or long-term forecasting.