Short-term user load forecasting based on GARCH-M family model with different distributions

Abstract

Power load forecasting is one of the basic tasks power system research,and time series analysis is currently the most widely used forecasting method. Aiming at the fluctuation and the characteristics of peak and thick tail of user daily load time series,the generalized autoregressive conditional heteroskedasticity-in-mean (GARCH-M) family model is proposed to predict user load. Firstly,the autoregressive conditional heteroskedasticity (ARCH) effect of load series is examined by using the Lagrange multiplier (LM) test according to the distribution of user daily load time series. Secondly,under three different distributions of Gaussian distribution,t-distribution and generalized error distribution (GED),the GARCH-M family model is established according to the different forms of fluctuation compensation terms. Finally,combined with the loss function,the prediction analysis results show that the GARCH-M family model with different distributions improves the accuracy of short-term user load prediction compared with the traditional time series analysis model.

Keywords

• time series analysis
• short-term user load forecasting
• autoregressive conditional heteroskedasticity (arch) effect
• generalized autoregressive conditional heteroskedasticity-in-mean (garch-m) family model
• fat tail effect
• loss function