Hybrid Momentum TAR-GARCH models for short term load forecasting

Short term load forecasting plays a fundamental role in providing input to economic dispatch and secure operation of power systems. Investigating the volatility of the load time series is essential to improving the performance of short term load forecasting. In this work, based on the threshold characteristics in the volatility of load time series, a novel Hybrid Momentum TAR-GARCH model is proposed for short term load forecasting. It is also demonstrated as a generalized form of the TAR-GARCH and Momentum TAR-GARCH models. Moreover, taking into account the fat-tail effect in electric load time series, the Hybrid Momentum TAR-GARCH model with a fat-tail distribution is proposed. Specifically, the models involving Laplace or generalized error distribution are studied. By means of the Conditional Maximum Likelihood Estimation (CMLE) method, the parameters are estimated for all the proposed models. With the help of the hybrid news impact surface, the impact of different shocks on load forecasting is analyzed. Load forecasting based on the practical load data of a representative city in China is performed. Results of the case study clearly validate the feasibility and effectiveness of the proposed models. Also, the model comparison for forecasting performance demonstrates that the Hybrid Momentum TAR-GARCH model with a generalized error distribution outperforms others based on several statistical criteria.