Regression Technique for Electricity Load Modeling and Outlined Data Points Explanation
Hundreds of explanatory variables such as historical consumption data, climate variables, socioeconomic and demographics parameters, etc. are used to forecast major forms of commercial energy consumptions. The scientists increasingly face the problems of big data: huge amount of data which grows in time. This article presents the analysis of the relationships between electricity amount to deliver and factors having more or less significant impact on electricity consumptions. The linear regression algorithm is used to reduce the set of explanatory variables and to evaluate their importance. A reduction of number of variables without significant loss of accuracy of the model is presented. Next, regression decision trees are used both to evaluate the quality of the modeled energy consumption, as well as to further explanatory variables reduction. The last part of the article deals with the outlining data points and explains reasons they come off model predictions.
KeywordsElectricity load modeling Regression decision tree Electricity short-term forecasting
- 1.Singh, A.K, Ibraheem, S.K., Muazzam, Md.: An overview of electricity demand forecasting techniques. In: Proceedings of National Conference on Emerging Trends in Electrical, Instrumentation & Communication Engineering, vol. 3, No. 3, pp. 38–48 (2013)Google Scholar
- 9.Gori, F., Takanen, C.: Forecast of energy consumption of industry and household and services in Italy. Heat Technol. 22(2), 115–121 (2004)Google Scholar
- 16.Gajowniczek, K., Nafkha, R., Ząbkowski, T.: Electricity peak demand classification with artificial neural networks. Comput. Sci. Inf. Syst., ACSIS 11, 307–315 (2017)Google Scholar
- 17.Marcjasz, G., Uniejewski, B., Weron, R.: Importance of the long-term seasonal component in day-ahead electricity price forecasting revisited: Neural network models. HSC Research Reports HSC/17/03 (2017)Google Scholar
- 18.Polish power system dataset. http://www.pse.pl/index.php?dzid=77. Accessed 12 Aug 2017