Article ID: | iaor2006505 |
Country: | Netherlands |
Volume: | 21 |
Issue: | 3 |
Start Page Number: | 425 |
End Page Number: | 434 |
Publication Date: | Jul 2005 |
Journal: | International Journal of Forecasting |
Authors: | Souza R.C., Bunn D.W., Hippert H.S. |
Keywords: | forecasting: applications, neural networks |
Neural networks have apparently enjoyed considerable success in practice for predicting short-term daily electricity load profiles. Most of these applications have utilised very large neural network specifications, which raises the methodological question of over-fitting. This paper examines this issue by comparing several forecasting methods on a sample of hourly electricity demands, including both large neural networks and conventional regression-based methods. We find good performance for the large neural networks, and offer some analysis of why forecasting the 24 element vector of daily electricity demands may be particularly conducive to this approach.