Deep learning for estimating energy savings of early-stage facade design decisions

Abstract

The selection of high-performance building facade systems is essential to promote building energy efficiency. However, this selection is highly dependent on early-stage design decisions, which are extremely challenging considering numerous design parameters with early-stage uncertainties. This paper aims to evaluate the applicability of deep learning networks in estimating the energy savings of different facade alternatives in the early-stage design of buildings. The energy performance of two competing façade systems (i.e., Ultra-High-Performance Fiber-Reinforced-Concrete and conventional panels) was estimated for different scenarios through building energy simulations using EnergyPlus™. Three deep learning networks were trained using the collected data from the simulation of fourteen buildings in fourteen different locations to estimate the heating, cooling, and total site energy savings. The accuracy of trained deep networks was compared with the accuracy of three common data-driven prediction models including, Gradient Boosting Machines, Random Forest, and Generalized Linear Regression. The results showed that the deep learning network trained to predict building total site energy savings had the highest accuracy among other models with a mean absolute error of 1.59 and a root mean square error of 3.48, followed by Gradient Boosting Machines, Random Forest, and last Generalized Linear Regression. Similarly, deep networks trained to predict building cooling and heating energy savings had the lowest mean average error of 0.20 and 1.17, respectively, compared to other predictive models. It is expected the decision support system developed based on this methodology helps architects and designers to quantify the energy savings of different facade systems in early stages of design decisions.