Estimating energy savings of ultra-high-performance fibre-reinforced concrete facade panels at the early design stage of buildings using gradient boosting machines

The selection of an energy-efficient building facade system, as one of the most important early-stage design decisions, plays a crucial role in reducing building energy use by controlling heat transmission between outdoor and indoor environments. This paper aims to evaluate the feasibility and applicability of gradient boosting machines in estimating the energy savings of different facade alternatives in the early-stage design of building facades. The energy performance of two competing facade systems was estimated for different scenarios using building energy simulations (i.e. EnergyPlus (TM)). Three gradient boosting machines were developed based on the data collected from the simulation of thirteen building types in fifteen different locations (i.e. 195 scenarios). The prediction performance of gradient boosting models was compared with the building energy simulation results of two new building models that were not used in the database development phase to validate the models. Moreover, the prediction power of the trained gradient boosting models was compared with three common prediction models (i.e. Artificial Neural Networks, Random Forest, and Generalized Linear Regression) based on several performance metrics. The results showed the superiority of gradient boosting machines over other models in estimating total site energy savings, heating energy savings, and buildings' cooling energy savings.