Energies, Vol. 19, Pages 21: Integrating Machine Learning Temporal Disaggregation and Physics-Based Simulation for Lifecycle Assessment of Buildings

Energies, Vol. 19, Pages 21: Integrating Machine Learning Temporal Disaggregation and Physics-Based Simulation for Lifecycle Assessment of Buildings

Energies doi: 10.3390/en19010021

Authors:
Giannis Iakovides
Renos Rotas
Petros Iliadis
Stefanos Petridis
Nikos Nikolopoulos
Elias Kosmatopoulos

This study presents an integrated framework for lifecycle assessment (LCA) and lifecycle costing (LCC) of buildings and districts that combines machine learning-based temporal disaggregation, physics-based simulation, and holistic environmental evaluation. The methodology addresses a key limitation of conventional LCA practice: the reliance on temporally aggregated energy data, which obscures daily and seasonal variability affecting environmental and economic indicators. A hierarchical disaggregation algorithm was used to reconstruct hourly electricity profiles from monthly totals and was coupled with the INTEMA building energy performance simulator and the VERIFY LCA/LCC platform. The disaggregation algorithm was validated on an office building in Cardiff, UK, supported by cross-validation across multiple UK office buildings, and achieved strong agreement with measured hourly consumption (R2 = 0.81, RMSE = 3.71 kWh). In the Cardiff case, the reconstructed hourly profiles reproduced lifecycle greenhouse gas emissions and costs within 0.5% of the reference hourly measurement approach, compared with deviations of 44.1% and 2.9% under conventional monthly aggregation. The complete hybrid framework was then applied to a district in Massagno, Switzerland, encompassing eight buildings with heterogeneous typologies, for which only aggregated energy data were available (monthly for the office building and annual for the others). Over a 20-year horizon, total emissions reached 9429 tCO2-eq and primary energy demand approached 226 GWh, equivalent to 41 kgCO2-eq·m−2·yr−1. The results illustrate the framework’s applicability to multi-building systems and its ability to support LCA and LCC in contexts with limited temporal data availability.