Input Variable Selection for RBFN-based Electric Utility's CO2 Emissions Forecasting

This study investigates the performance of radial basis function networks (RBFN) in forecasting the monthly CO2 emissions of an electric power utility. We also propose a method for input variable selection. This method is based on identifying the general relationships between groups of input candidates and the output. The effect that each input has on the forecasting error is examined by removing all inputs except the variable to be investigated from its group, calculating the networks parameter and performing the forecast. Finally, the new forecasting error is compared with the reference model. Eight input variables were identified as the most relevant, which is significantly less than our reference model with 30 input variables. The simulation results demonstrate that the model with the 8 inputs selected using the method introduced in this study performs as accurate as the reference model, while also being the most parsimonious.

• I. Falconett
• K. Nagasaka

• Correlation analysis
• CO2 emissions forecasting
• electric power utility
• radial basis function networks.

[1] M. Kainuma, Y. Matsuoka, and T. Morita, "The AIM/end-use model and
its application to forecast Japanese carbon dioxide emissions," European
Journal of Operational Research, Vol. 122, Issue 2, 16 April 2000, pp.
416-425.
[2] S. Kasahara, S. Paltsev, J. Reilly, H. Jacoby, and A. D. Ellerman,
"Climate Change Taxes and Energy Efficiency in Japan," Journal of
Environmental and Resource Economics, Springer, Vol. 37, Number 2,
June 2007, pp.377-410.
[3] C. Yang and S. Schneider, "Global carbon dioxide emissions scenarios:
sensitivity to social and technological factors in three regions," Mitigation
and Adaptation Strategies for Global Change, Springer, Vol. 2, Number
4, 1998, pp. 373-404.
[4] R. Schmalensee, T. M. Stoker, and R. Judson, "World carbon dioxide
emissions: 1950-2050," Review of Economics and Statistics, Vol. 80,
issue 1, 1998, pp. 15-27.
[5] D. Holtz-Eakin, and T. M. Selden, "Stoking the fires? CO2 emissions and
economic growth," Journal of Public Economics, Vol. 57, Issue 1, May
1995, pp. 85-101.
[6] A. Islam, and D. Wang, "Forecasting an Electric Utility's CO2 Emissions
Using SAS/AF and SAS/STAT Software: A Linear Analysis," SAS
SUGI22 conference proceedings, San Diego - California, March 16-17,
1997.
[7] A. Lendasse, J. Lee, E. de Bodt, V. Wertz, and M. Verleysen,
"Approximation by radial-basis function networks: application to option
pricing", ACSEG 2002 conference proceedings Connectionist
Approaches in Economics and Management Sciences, Boulogne-sur-Mer,
France, 2002, pp. 201-212.
[8] D. Wedding, K. Cios, "Time series forecasting by combining RBF
networks, certainty factors and Box-Jenkins model", Neurocomputing,
Vol. 10, Issue 2, 1996, pp. 149-168.
[9] B. Whitehead, T. Choate, "Cooperative-competitive genetic evolution of
radial basis function centers and widths for time series prediction", IEEE
Transactions on Neural Networks, Vol. 7, 1996, pp. 869-880.
[10] M. J. Orr, "Introduction to Radial Basis Functions Networks," Technical
Report, Edinburgh University, Edinburgh, Scotland, UK, April 1996.