Year: 2015
Type: conference-paper
Abstract: Generalized Additive Models (GAM) are a widely pop-ular class of regression models to forecast electricity demand, due to their high accuracy, flexibility and inter-pretability. However, the residuals of the fitted GAM are typically heteroscedastic and leptokurtic caused by the nature of energy data. In this paper we propose a novel approach to estimate the time-varying conditional vari-ance of the GAM residuals, which we call the GAM2 algorithm. It allows utility companies and network op-erators to assess the uncertainty of future electricity de-mand and incorporate it into their planning processes. The basic idea of our algorithm is to apply another GAM to the squared residuals to explain the depen-dence of uncertainty on exogenous variables. Empiri-cal evidence shows that the residuals rescaled by the estimated conditional variance are approximately nor-mal. We combine our modeling approach with online learning algorithms that adjust for dynamic changes in the distributions of demand. We illustrate our method by a case study on data from Réseau de transport d’électricité, the operator of the French transmission grid. 1
Authors: Tri Kurniawan Wijaya, Mathieu Sinn, Bei Chen
Institutions: École Polytechnique Fédérale de Lausanne, IBM (United States)
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W564250221