Modeling and Forecasting Daily Electricity Load Curves: A Hybrid Approach

• Activity Number: 639
• Type: Invited
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: IMS
• Author(s): Haeran Cho*+ and Yannig Goude and Qiwei Yao
• Companies: London School of Economics and EDF France and London School of Economics
• Keywords: electricity load ; factor modelling ; curve time series ; generalised additive model ; information criterion

Abstract

In this paper, we propose a method for modelling the amount of electricity usage which is of particular importance in making one-day-ahead prediction of electricity load. The proposed procedure consists of two stages, which are: (i) fitting a generalised additive model to the weekly average of electricity loads to take into account the overall trend, seasonality and the effect of meteorological factors, and (ii) modelling the dependence structure between any two consecutive daily loads in a curve time series framework. Specifically, the second stage involves the identification of a low-dimensional structure that well-describes the serial dependence between daily curves, which reduces the problem of predicting a curve to that of predicting a low-dimensional coefficient vector. The performance of the proposed methodology is demonstrated in a comparative simulation study using the French electricity load dataset between 1996 and 2009.