A method for short-term wind speed time series forecasting using Support Vector Machine Regression Model

Wind speed forecasting has drawn a lot of research interests around the globe as it plays a key role in wind power plant operation. Accurate wind speed forecasting is vital for the integration of wind energy conversion system into existing electric power grids. The important factor of wind speed forecast is the choice of accurate prediction algorithm. Support Vector Machine Regression Model (SVM-R), the most widely used algorithm for classification and forecasting measures, has shown extraordinary performance in various fields for short-term forecasting. Different SVM kernels including polynomial, linear and Gaussian have been explored. The performance of each kernel function has investigated on real time-series wind speed data for the site located at coastal areas of Sindh, Pakistan. The algorithm converts original training data into a higher dimension using nonlinear mapping. Optimal linear hyper-plane (LHP) is examined for separating data of one class from another one within this new dimension. The trend of root mean square error (RMSE) due to variation in various parameters, i.e., size of training sample, kernel parameters and regularization parameter has been presented. The LIBSVM software has been used in R environment to implement SVM-R model. The results of minimum wind speed prediction error in SVM linear kernel reveal that better selection of kernels can improve the performance of SVM-R.