Machine learning-based localized predictive modeling of household energy consumption in the Netherlands

Abstract

This study explores annual energy consumption by postal code through the impact of localized climatic factors, demographic traits, and building features on Household Energy Consumption (HEC) in the Netherlands over 11 years. Aggregating annual data at the level of postal codes and street level, the research provides an inclusive analysis of the impact of temperature, urbanization, household size, building age, and building surface on electricity and gas consumption patterns. Using machine learning methods, specifically the XGBRegressor Time Series Model, the study finds that previous years’ consumption, building age, and building surface at the postal code level are the primary factors in predicting annual energy consumption at the street level; demographic factors and climatic factors have a relatively lesser impact at this level of aggregation. The study’s findings highlight the importance of energy policies and infrastructure planning that are sensitive to local needs and increase energy resilience in environmental change. This research adds to the current debates on climate change adaptation by providing actionable recommendations to policymakers seeking to improve energy resilience in the context of environmental change.