A Multivariate Wind Power Forecasting Model Based on LS-SVM

There are many multivariate forecasting models which incorporate weather indicators and other information for wind farm power output forecasting. In most situations, performance of these individual models is problem-dependent. Thus, it is difficult for forecasters to choose the right technique for unique situations. In this paper, firstly, indicators such as wind speed, and wind direction are analyzed and selected. Then, a new multivariate LS-SVM model and some classical linear and nonlinear multivariate models are presented. Finally, wind power output data from 78 wind parks for a period of 1 year from America wind data Pool are used to test and compare the models. The results show that the multivariate LS-SVM model can outperform other models such as multivariate linear models and multivariate NN model on all the four measures, i.e. MAPE, large error, average rank and performance score.