Flexibility characterization of residential electricity consumption: A machine learning approach

In this paper, we propose a methodology based on machine learning techniques to characterize the flexibility of electricity consumption in the residential sector. The main challenge is that the characterization of flexibility requires to know which and when appliances are being used and how available are users to change its utilization: However, this type of data is not generally available. In this work, we propose a full-stack methodology to solve this problem: we start by processing total electricity consumption data with feature engineering; then we use Non-Intrusive Load Monitoring (NILM) or Interactive Learning (IL) to identify the use of the appliances with higher flexibility (water heating, space heating and clothes drier); then we apply a Random Forest classifier to identify when the flexible appliances are being used; and finally we apply a K-means clustering algorithm to evaluate the flexibility of such appliance. We compare the results using accuracy, recall, and f-score indicators. The results show that the proposed approach can be used to characterize with high accuracy the use of flexible appliances just based on aggregated electricity consumption collected by smart meters with a low sampling rate. Further, we also demonstrate that Interactive Learning is a viable alternative approach to NILM to disaggregate electricity consumption. (c) 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).