An ensemble clustering based framework for household load profiling and driven factors identification

Abstract

A two-stage framework is presented for household electricity consumption pattern mining, in which a concurrent k-means and spectral clustering (CKSC) method is used in the first stage. In the second stage, different household electricity consumption (HEC) patterns are first analyzed. Then, driven factors behind each electricity consumption pattern are discovered based on multi-nominal logistic regression. Experiment is carried out using the Irish open data, including both detailed smart meter measurements and the survey data. CKSC method efficiently divides the 4181 households into six groups. Based on that, different HEC patterns in weekdays and in weekends are extracted and discussed. Meanwhile, abnormal HEC patterns over Christmas are also discovered. The results show that the identified six groups are representative, given that the discovered HEC patterns are varied both in consumption scales and intraday volatilities. Especially, potential household groups for peak load shifting and shaving are found and further discussed. Besides, the established multi-nominal logistic regression model is statistically significant. Considering the dwelling characteristics, family characteristics and household appliances of households without smart meter readings, their HEC patterns can be inferred based on the proposed model in this study.