Short-term wind speed forecasting using wavelet transform and support vector machines optimized by genetic algorithm

Affected by various environment factors, wind speed presents characters of high fluctuations, autocorrelation and stochastic volatility; thereby it is hard to forecast with a single model. A hybrid model combining with input selected by deep quantitative analysis, Wavelet Transform (WT), Genetic Algorithm (GA) and Support Vector Machines (SVM) was proposed. WT was exploited to decompose the wind speed signal into two components, an approximation signal to maintain the major fluctuations and a detail signal to eliminate the stochastic volatility. SVM were built to model the approximation signal. Autocorrelation and partial correlation were applied to analyze the inner ARIMA Autoregressive Integrated Moving Average (ARIMA) relationship between the historical speeds thus to select the input of SVM from them, and Granger causality test was applied to select input from environment variables by checking the influence of temperature with different leading lengths. The parameters in SVM were fine-tuned by GA to ensure the generalization of SVM. A case study of a wind farm from North China demonstrates that this method outperforms the comparison models.