Artificial neural networks to forecast building heating/cooling demand and climate resilience based on envelope parameters and new climatic stress indices

Accurate prediction of building space conditioning demand is essential for energy efficiency, cost reduction and climate resilience. In this study, artificial neural networks (ANNs) are applied to forecast the thermal energy demand (TED) of buildings, using as inputs envelope parameters and two new climatic stress indices, i.e., HSI (heating stress index) and CSI (cooling stress index). To validate the proposed approach, three Italian residential case studies are analyzed: the EMAR (EnergyPlus + Matlab® for Residential) building model previously investigated by some of the authors, a multi-family residential building and an apartment block, both from the TABULA project. The results demonstrate that the method is reliable, with an average absolute percentage error in ANNs outputs compared to EnergyPlus simulations of 4.6 % for TED heating vs HSI and 10.4 % for TED cooling vs CSI. Furthermore, climate change scenarios based on the IPCC (Intergovernmental Panel on Climate Change) assessments are implemented to forecast the future TED of the case studies using ANNs. In the 2050 climate scenario, with re-evaluated HSI and CSI, ANNs forecasts show an expected decrease in TED heating, almost entirely offset by the increase in TED cooling, resulting in an approximately stable overall TED for space conditioning. The findings highlight the advantages of using the proposed ANNs-based approach to assess building resilience and adaptability to changing climate conditions.