Short Term Load Forecasting using Regime-Switching GARCH Models

Modeling the volatility in load time series can contribute to improving the performance of short-term load forecasting (STLF). In this work, to capture the nonlinear characteristics of volatility, regime switching in the volatility of load time series is investigated. By combining regime-switching models with Generalized Auto-Regressive Conditional Heteroscedastic (GARCH) models, two types of regime-swithcing GARCH models, Threshold Auto-Regressive GARCH (TAR-GARCH) and Logistic Smooth Transition Auto-Regressive GARCH (LSTAR-GARCH) load forecasting models, are studied. In addition, LSTAR is effectively used to handle the discontinuity point problem of TAR near the threshold. Furthermore, the fat-tail effect in load time series is examined, and the regime switching GARCH models with fat-tail distribution are proposed for generalization. Case study on a practical sample for STLF clearly validates the feasibility and effectiveness of the proposed methods. The slope structure of News Impact Curve (NIC) is proposed to depict the behavior of TAR-GARCH and LSTAR-GARCH type models near the threshold. Forecasting results by all the presented regime-switching GARCH type models are provided. It is concluded that LSTAR-GARCH model with fat-tail distribution is a promising method for STLF.