A novel data-driven approach for residential electricity consumption prediction based on ensemble learning. (1st May 2018)

• Record Type: Journal Article
• Authors: Chen, Kunlong; Jiang, Jiuchun; Zheng, Fangdan; Chen, Kunjin
• Is Part Of: Energy. Volume 150(2018)
• Page Start: 49
• Page End: 60
• Publication Date: 2018-05-01
• Subjects: Household electricity consumption -- Ensemble learning -- Neural network -- Extreme gradient boosting; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2018.02.028
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: With the development of smart grid as well as the electricity market, it is of increasing significance to predict the household electricity consumption. In this paper, a novel data-driven framework is proposed to predict the annual household electricity consumption using ensemble learning technique. The extreme gradient boosting forest and feedforward deep networks are served as base models. These base models are combined by ridge regression. What is more, the importances of input features are estimated. A subset of features is selected as the important features to feed into the model to increase its accuracy. A comparison of the proposed ensemble framework against classical regression models indicates that the former can reduce by 30 % of the prediction error. The results of this study show that ensemble learning method can be a convenient and accurate approach to predict household electricity consumption. Highlights: An electricity consumption prediction framework is proposed. Ensemble learning method is introduced to increase the prediction accuracy. The feature importance is estimated based on information gain. The relationship between important features is illustrated.