Machine learning to predict building energy performance in different climates

Abstract Digitalization is sweeping the world of buildings. Notably, use machine and deep learning techniques to develop buildings’ digital twins becoming crucial foster energy transition construction sector a sustainable urban growth. Digital can ensure user-friendly, fast reliable prediction building loads demands, thereby enabling comprehensive optimization planning, design operation. Accordingly, this study investigates predict heating in Rome (Italy, Mediterranean conditions, “Csa” climate Köppen Geiger classification) Berlin (Germany, European backcountry, “Cfb”). Firstly, real building, located Benevento, used artificial neural networks (ANNs), then implemented MATLAB® achieve meta-models behavior. NARX (nonlinear autoregressive model with exogenous inputs) are trained based on simulated data, provided by well-known simulation tool EnergyPlus using software DesignBuilder® as interface. The meta-model inputs related weather while required outputs concern thermal load for space heating. analysis performed reference annual forecasts demands. In all cases, ANNs architecture optimized best fitness outputs. results show that be precious support operation different buildings climates. Nonetheless, meta-modeling procedure needs properly conducted experts set suitable frameworks hyperparameter values ANNs, well right interpretation results.