A data-driven methodology for deriving electricity consumption typologies from smart meters

We present a data-driven methodology to identify residential electricity consumption typologies from large-scale smart meter data. The proposed approach combines seasonal feature extraction, clustering via Self-Organizing Maps (SOMs), and expert-in-the-loop validation to ensure both statistical robustness and operational relevance. Although the methodology is tailored to residential consumption, it also captures other non-residential load patterns – such as commercial, industrial, and public-sector profiles – present in the analyzed data. The methodology was applied to more than 23,000 time series from five international datasets, resulting in 40 distinct consumption patterns. These clusters were grouped into five behavioral categories – primary residences, holiday homes, equipment-intensive households, offices, and public lighting – each capturing characteristic daily and seasonal load signatures. The resulting typology enables a more realistic and interpretable representation of household electricity use, improving the design of demand-side strategies, tariff schemes, and forecasting models. In particular, it offers a practical foundation for energy planning and policy targeting across diverse regions, and can inform real-time classification tools or adaptive services under conditions of data scarcity or external disruptions. • A new data-driven methodology for residential electricity consumption typologies. • Integration of time series feature extraction, Self-Organizing Maps, and expert validation. • Identification of 40 operationally relevant consumption patterns across multiple countries.