Forecasting Short-Term Wind Speed Using Support Vector Machine with Particle Swarm Optimization

High precision forecasting is a prerequisite and guarantee for the operation of grid-connected wind farms. Affected by various environmental factors, wind speed exhibits high fluctuations, autocorrelation and stochastic volatility. Therefore it remains great challenges for short-term wind speed forecasting. To capture its non-stationary property and its tendency, a forecasting model using support vector machine (SVM) with particle swarm optimization (PSO) is proposed for quantitative analysis. PSO is exploited to determine the optimal regularization and kernel parameters for selecting SVM parameters. The present model employes not only the small learning ability and simple calculation of SVM, but also strong global search ability of PSO. The addressed model was tested using real wind speed data. Experimental results show that, the proposed model has the best forecasting accuracy, comparing with classical SVM model and back propagation neural network model.