Clustering Open Data for Predictive Modeling of Residential Energy Consumption across Variable Scales: A Case Study in Andalusia, Spain

Abstract

The energy budget of households, linked to residential energy consumption (REC), serves as a critical indicator of quality of life and economy trends. Despite the lack of widely available accurate statistics at regional or smaller scales, they are of crucial interest for a better understanding of the features influencing REC and its impact on energy poverty, wellbeing, and the climate crisis. This research aims to present a new information model for predictive parameters and REC forecasting through an innovative use of available open data. Geoprocessing, data mining, and machine learning clustering algorithms were applied to open datasets of location, population, and residential building stock parameters highly correlated with their REC, on the ensemble of 785 municipalities of Andalusia, Spain. The model identified 65 clusters of towns sharing the same potential REC, with 73% of the population concentrated in 10 of these. The resulting data-driven bottom-up model of provincial REC had a mean absolute error of only 0.63%. Furthermore, it provided the territorial distribution, with local resolution, of the identified clusters of cities with similar characteristics. This methodology, with a flexible regional- to city-scale analysis, provides knowledge generation that offers numerous practical applications for energy policy planning. Its future implementation would assist stakeholders and policymakers in enhancing the performance and decarbonization of the residential building stock.