Short-term wind speed forecasting using STLSSVM hybrid model

With the rapid growth of wind power, wind speed forecasting becomes more and more significant to ensure stable and efficient operations of wind power system. This paper proposes an improved hybrid methodology for short-term wind speed forecasting. After data preprocessing, MI algorithm is used to select proper wind speed features, then Ensemble Empirical Mode Decomposition (EEMD) is utilized to decompose the original wind speed series in order to make the chaotic series more stable. A novel model named ST-LSSVM is proposed to forecast the decomposed sub-series, which combines the Least Squares Support Vector Machine (LSSVM) and State Transition method (ST). In order to further enhance the model performance, Particle Swarm Optimization (PSO) is utilized to fine-tune the parameter values of the ST-LSSVM. Finally, real world wind speed data are used to estimate the proposed hybrid forecasting model. The results demonstrate that proposed ST-LSSVM hybrid model has the best prediction accuracy in one to six step’s forecasting, compared with Persistence, Autoregressive Integrated Moving Average (ARIMA), Back-Propagation Neutral Network (BPNN) and Least Squares Support Vector Machine (LSSVM) models.