Probabilistic Forecasting of Wind Power Generation Using Online LASSO VAR and EGARCH Model

Wind power generation has uncertainty due to the high fluctuation of wind speed. In traditional wind power prediction models, the uncertainty is measured by normal distribution with zero mean and constant variance. However, the variance may vary with time, which means the variance has heteroscedasticity. To improve the prediction accuracy, this paper proposes a new integrated probabilistic wind power prediction model for wind farm considering heteroscedasticity based on online least absolute shrinkage and selection operator and vector autoregression (LASSO VAR) and the exponential generalized autoregressive conditional heteroskedasticity (EGARCH) model. First, online LASSO VAR is used to forecast power output. Then, heteroscedasticity of residuals is validated by autoregressive conditional heteroskedasticity test. Considering heteroscedasticity, the news impact curve and dynamic significance line verify that positive and negative residuals affect future volatility asymmetrically. Thus, the EGARCH model is used to forecast the residuals to obtain the conditional variance of point prediction results. Finally, the probabilistic result of total power output is obtained by summing the power output of turbines in the wind farm considering the correlation of the active wind power of wind turbines. This method is applied to forecast the power output of a wind farm in East China and is proved effective in improving the prediction accuracy.