Behaviour-Conditioned Neural Processes for Heterogeneous Household Load Forecasting

arXiv·medium signal

Ramin Soleimani, Andrea Visentin and Dirk Pesch (arXiv 2607.16168, cs.LG) tackle residential short-term load forecasting by conditioning neural processes on behavioural profiles, addressing the core difficulty that household demand is heterogeneous across homes and variable within one. Neural processes are a good fit here because they handle few-shot adaptation to a new household without retraining. Relevant beyond energy: this is the general shape of per-user personalization with limited history.