Wind Speed Forecasting by Conventional Statistical Methods and Machine Learning Techniques

Abstract

Intermittency is the main challenge for wind power integrated in power grids. The intermittent nature of wind speed gives rise to fluctuations of output power from a wind turbine that poses serious concerns over power system stability and reliability. Therefore, accurate wind speed forecasting is essential for planning and operation of wind power generation. In this paper, short term wind speed forecasting methods are investigated using one-year historical data. Conventional time series methods (Autoregressive Moving Average (ARMA) and Autoregressive Integrated Moving Average (ARIMA)) and machine learning methods (Artificial Neural Network (ANN) and three Support Vector Machine (SVM) algorithms (Linear SVM, Polynomial SVM and radial basis function (RBF) SVM)) are considered in this study. The forecasted wind speed data are compared with historical wind speed data in terms of Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Results obtained in this paper show that machine learning methods outperformed conventional time series methods in short-term wind speed prediction.