Short-term wind power prediction using Least-Square Support Vector Machines

This paper presents a short-term prediction scheme of wind power from wind speed data using Least-Square Support Vector Machines (LS-SVM). The paper develops different LS-SVM models that make use of atmospheric temperature and take advantage of the periodicity of the wind speed data. Results show that atmospheric temperature and using the periodic trend improves the predictions accuracy over the persistence model. The proposed models predict wind power within an error margin of 20% of rated power, 85% of the time.