AttGRU-based intra-site and cross-site building energy consumption prediction using meteorological features with uncertainty-aware CO2 scenario analysis

Building energy consumption varies with meteorological conditions, which indirectly affect CO₂ emissions through energy demand. However, existing studies on overall energy demand often overlook two key aspects: the influence of meteorological factors on energy consumption across multiple sites and how variations in fossil and renewable energy use affect CO₂ emissions and reductions in the building sector. We utilize meteorological data from the Building Data Genome Project-II dataset to predict building energy consumption. Our model, AttGRU, is a gated recurrent unit enhanced with soft attention to generate context vectors, which are further processed using dropout-distributed masks to avoid over-reliance on dominant patterns. Based on these energy consumption predictions, we then analyze various fossil-renewable energy mix scenarios and evaluate their impacts on CO₂ emissions and reductions in the building sector. The AttGRU model demonstrates superior performance for building energy consumption prediction compared to existing methods, validated through cross-site and intra-site experiments. In cross-site evaluation, the proposed model achieved a mean squared error (MSE) of 0.0212, a mean absolute error (MAE) of 0.0640, and a coefficient of determination (R²) of 0.930 for a temporal horizon of T=24. In intra-site experiments, the model obtained an MSE of 0.106, an MAE of 0.240, and an R² of 0.668 for short-term prediction (T=1). When longer temporal sequences considered (T=24), the prediction performance improved significantly, yielding an MSE of 0.0105, an MAE of 0.0582, and an R² of 0.9671. The analysis further identifies air temperature and dew temperature as the most influential meteorological drivers of building energy demand. Building on these predictions, the proposed uncertainty-aware scenario-based analysis quantitatively establishes relationships between energy and carbon outcomes, showing a strong positive association between fossil energy use and CO₂ emissions and an inverse relationship with renewable energy penetration.