Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
}
TY - GEN
T1 - The impact of preprocessing on forecasting electrical load: an empirical evaluation of segmenting time series into subseries
AU - Crone, Sven F.
AU - Kourentzes, Nikolaos
PY - 2011
Y1 - 2011
N2 - Forecasting future electricity load represents one of the most prominent areas of electrical engineering, in which artificial neural networks (NN) are routinely applied in practice. The common approach to overcome the complexity of building NNs for high-frequency load data is to segment the time series into simpler and more homogeneous subseries, e. g. seven subseries of hourly loads of only Mondays, Tuesdays etc. These are forecasted independently, using a separate NN model, and then recombined to provide a complete trace forecast for the next days ahead. Despite the empirical importance of load forecasting, and the high operational cost associated with forecast errors, the potential benefits of segmenting time series into subseries have not been evaluated in an empirical comparison. This paper assesses the accuracy of segmenting continuous time series into daily subseries, versus forecasting the original, continuous time series with NNs. Accuracy on hourly UK load data is provided in a valid experimental design, using multiple rolling time origins and robust error metrics in comparison to statistical benchmark algorithms. Results indicate the superior performance of NN on continuous, non-segmented time series, in contrast to best practices in research, practice and software implementations.
AB - Forecasting future electricity load represents one of the most prominent areas of electrical engineering, in which artificial neural networks (NN) are routinely applied in practice. The common approach to overcome the complexity of building NNs for high-frequency load data is to segment the time series into simpler and more homogeneous subseries, e. g. seven subseries of hourly loads of only Mondays, Tuesdays etc. These are forecasted independently, using a separate NN model, and then recombined to provide a complete trace forecast for the next days ahead. Despite the empirical importance of load forecasting, and the high operational cost associated with forecast errors, the potential benefits of segmenting time series into subseries have not been evaluated in an empirical comparison. This paper assesses the accuracy of segmenting continuous time series into daily subseries, versus forecasting the original, continuous time series with NNs. Accuracy on hourly UK load data is provided in a valid experimental design, using multiple rolling time origins and robust error metrics in comparison to statistical benchmark algorithms. Results indicate the superior performance of NN on continuous, non-segmented time series, in contrast to best practices in research, practice and software implementations.
U2 - 10.1109/IJCNN.2011.6033657
DO - 10.1109/IJCNN.2011.6033657
M3 - Conference contribution/Paper
SN - 978-1-4244-9636-5
SP - 3285
EP - 3292
BT - The 2011 International Joint Conference on Neural Networks (IJCNN)
PB - IEEE
CY - New York
T2 - International Joint Conference on Neural Networks (IJCNN)
Y2 - 31 July 2011 through 5 August 2011
ER -