Search

Abstract Wind speed and power forecast is an essential component to ensure grid stability and reliability. The traditional forecasting methods fail to address the non-linearity in the wind speed time-series, thus paving way for machine intelligent algorithms. This paper discusses a hybrid machine intelligent wind forecasting model utilizing different variants of Support Vector Regression (SVR) built on wavelet transform. Various performance indices are evaluated to identify the possible best one among four different machine learning regressors for wind forecasting application. Apart from standard ε-SVR and LS-SVR, two new regression models, namely, ε-Twin Support vector regression (ε-TSVR) and Twin Support vector regression (TSVR) are used to forecast short-term wind speed, and are compared with Persistence model for four wind farm sites. The effect of larger dataset on forecasting performance is evaluated for two wind farm sites from USA and India. Further, wind power ramp events are investigated at different hub heights and the forecasting performance of different variants of SVR is compared for five wind farm sites. Highlights • Machine intelligent SVR variants are discussed for short-term wind forecasting. • Overfitting and underfitting scenarios of variants of SVR are assessed. • Forecasting performance of SVR variants is assessed during wind power ramp event. • Wind power ramp events are analyzed at different hub heights for 5 wind farm sites. [ABSTRACT FROM AUTHOR]