Climate-informed residential electricity demand modelling: a comparative study of inland and coastal dwellings

Understanding climate-sensitive residential electricity demand is important for data-driven load modelling and digital energy management, yet dwelling-scale comparative evidence remains limited. This study analyzes synchronized hourly smart-meter and weather measurements from two monitored residential dwellings in contrasting climatic settings in Karnataka, India: an inland dwelling in Bangalore and a coastal dwelling in Mangalore. Descriptive, correlation, multivariate, lagged, and forecasting-oriented feature analyses were used to characterize weather-demand relationships and to evaluate whether climate-aware engineered features improve predictive performance relative to raw weather inputs. The inland dwelling showed weak climate-demand coupling, whereas the coastal dwelling exhibited stronger sensitivity to solar-related variables, modest humidity effects, and exploratory delayed weather-response patterns. Climate-only baseline models explained limited variance, particularly for the inland dwelling, indicating the importance of non-climatic drivers and richer contextual data. At the same time, lagged, interaction, and rolling climatic features improved predictive performance for several benchmark models, with gains that were model- and dwelling-dependent. Although the findings are case-specific, the study provides dwelling-scale evidence on how the two monitored dwellings responded to contrasting climatic exposures during the observation period and demonstrates the value of interpretable climate-aware feature construction for residential load modelling and smart-building analytics.