An Effective Very Short-Term Wind Speed Prediction Approach Using Multiple Regression Models

As one of the dominant forms of renewable energy sources, wind power generation plays an increasingly important role in modern energy landscape. A very short-term wind speed prediction is essential for monitoring and control of power systems with high wind power penetration to improve system stability and reliability. In this article, an accurate five-minute horizon wind speed prediction method is proposed by integrating and comparing four machine learning regression algorithms, including multiple-layer perception regressor (MLPR), random forest regressor (RFR), K-nearest neighbors regressor (KNNR), and decision tree regressor (DTR). Twenty minutes historical data of wind speed in a one-minute interval is used for wind speed predictions, which are actual wind speed data provided by the National Renewable Energy Laboratory (NREL), Golden, CO, USA. The proposed method is intended to offer an effective and low-cost way for very short-term wind speed prediction. Pearson’s correlation coefficient (PCC) is adopted for feature selection. The four algorithms are evaluated through statistic error indices and Bland–Altman method, and the MLPR algorithm shows the best performance.