Leveraging Smart Meter Data for Adaptive Consumer Profiling

Abstract

The growing availability of smart meter data from households is contributing to the digitalization of the energy sector and driving its transformation into new business concepts, such as flexibility markets. This data allows for more precise consumer profiling than ever before, which can be beneficial for demand response modeling, dynamic tariff models… While profiling customers based on time-series consumption data has been extensively studied in the literature, this article offers a different approach by examining the feasibility of profiling new customers using their early data, with varying amounts of smart meter data available (from one month to one year) and considering different approaches of frequency of data (daily, weekly, bi-weekly or monthly). By using an existing dataset (Low Carbon London project, spearheaded by UK Power Networks) for modeling validation and test purposes, the data pipeline in our study employs Dynamic Time Warping K-Means clustering. We compare the similarity of clusters obtained with different data lengths against ground-truth clusters built on all the data by using Rand Index. Results indicate that, on average, one month of data is enough for accurate profiling for a small number of consumer types, although for a larger number it is recommended to use a slot of 6 months. The approach enables effective profiling even on limited data for new customers.