Wavelet transform and variants of SVR with application in wind forecasting

Abstract

Accurate wind prediction plays an important role in grid integration. In this paper, we analyze the performance of a hybrid forecasting method comprising of wavelet transform and different variants of Support Vector Regression (SVR) like ε -SVR, Least Square Support Vector Regression (LS-SVR), Twin Support Vector Regression (TSVR) and ε -Twin Support Vector Regression (ε -TSVR). Each of these methods is trained and tested for a wind farm Sotavento, Galicia, Spain. Wavelet transform is used to filter the raw wind speed data from any kind of stochastic volatility. Among the different variants of SVR, the forecasting results of ε -TSVR and TSVR are compared with ε -SVR and LS-SVR to evaluate various quantitative measures like RMSE, MAE, SSR/SST and SSE/SST.