Leveraging Electricity Load Data for Household Demographic Size Characterization: HouseMix

Household socio-demographic factors are essential for understanding and modeling electricity consumption patterns, particularly in the context of renewable energy integration and smart grid development. However, accurately inferring such information from electricity load data remains challenging due to the complex and variable nature of consumption behaviors, which are shaped by latent factors such as lifestyle, appliance usage, and occupancy patterns. In this paper, we propose HouseMix, a novel framework for predicting socio-demographic information from electricity consumption data. By leveraging an attention-guided mixture-of-experts architecture, HouseMix effectively captures diverse temporal dynamics and extracts informative representations from load profiles, providing a cost-effective and scalable solution for downstream applications such as personalized load forecasting and energy policy development. We validate HouseMix on real-world smart meter data through the task of predicting household demographic size, a representative and widely relevant socio-demographic factor. Experimental results demonstrate that HouseMix achieves high prediction accuracy and significantly improves the performance of load forecasting models, offering valuable insights for grid management and demand-side strategies.