Seasonal ARIMA and LSTM Models for Wind Speed Prediction at El-Oued Region, Algeria

The Saharan state of El-Oued becomes a hub for agricultural activity, and leads the production of potatoes and dates in Algeria. However, the region faces challenges due to limited access to rural electricity and wind erosion. Given the cube relationship between wind power and wind drift potentials with wind velocity, it is imperative to obtain accurate wind speed predictions. These forecasts are essential to optimize the feasibility of off-grid electricity powered by wind and mitigate wind erosion risks. One major challenge in predicting wind speed is its stochastic, intermittent, and non-dispatchable characteristics. This study focuses on short-term wind speed prediction using statistical methods like seasonal ARIMA and deep learning techniques such as LSTM. Hourly wind speed data measured at 10 m AGL at El-Oued was utilized to compare the models. While SARIMA's performance improved significantly with rolling prediction, LSTM exhibited superior performance, demonstrated by achieving the lowest RMSE and MAE values for the deployed dataset.