Understanding the characteristics of residential electricity consumption in Korea: Applying machine learning techniques to household-level energy survey data
Understanding the characteristics of residential electricity consumption in Korea: Applying machine learning techniques to household-level energy survey data
Moon, Jongwoo
Moon, Jongwoo
Demand-side approaches become an important pillar for energy analysis, and their roles for achieving climate targets have been increasingly emphasized globally. Particularly, Korea is one of the countries experiencing a rapid transition of demographic and household structures, and accordingly, the current and future energy demand could be significantly affected. As per the importance of the understanding the energy demand characteristics, this study contributes to understanding the electricity consumption of households by analyzing how the various household characteristics can be used to understand the household’s electricity consumption with household-level survey and machine learning techniques. This study utilizes the Household Energy Standing Survey published in 2022 and selects key household, housing, and appliance ownership and usage characteristics from the entire dataset. Afterward, the study applies Support Vector Machine, Random Forest, and Decision Tree classifiers to classify the household’s monthly electricity consumption. The results suggest that the Random Forest classifier provides slightly better performances in general compared to the other models. Moreover, the feature importance suggests that the housing characteristics, such as the size of housing, and appliance usage information, and some household characteristics, such as the number of household members and household income, are relatively important features for classification. Although the study finds some evidence of the importance of household and behavioral information in understanding the household’s electricity consumption, the study also identifies the limitation of the survey dataset in extracting the behavioral information.
1405_04.pdf
