Short term load forecasting based on improved ESTAR GARCH model

Taking into account the outlier effect in volatility of load time series, the impacts of extra-large shocks on load time series are investigated, and a novel load forecasting method based on improved ESTAR (IESTAR) structure is established. On the second order moment level, the IESTAR GARCH model is proposed for short term load forecasting. Furthermore, the proposed model with fat-tail distribution is generalized and estimated. Mechanism of regimes against different magnitudes of shocks is analyzed with the help of large shock parameter in the proposed models. Moreover, the outlier effect is further investigated by the News Impact Curve (NIC) and the involute test from two different viewpoints. Case study on a practical example for short term load forecasting clearly validates the feasibility of the proposed method. Forecast performance comparison of all the IESTAR GARCH models is provided, and IESTAR GARCH model with fat-tail distribution is proved to be a prospective model for short term load foresting in this study.