Artificial neural network models for building energy prediction

Abstract

There is a national need for a quick and easy building energy performance assessment system of existing buildings, without resorting to dynamic building energy simulation tools which usually require significant cost, time and expertise. In this study, the authors report the development of a building energy profiling system which is based on Artificial Neural Network (ANN) models. The ANN models were made by a series of EnergyPlus pre-simulations sampled by a Monte Carlo technique. The MBE and CVRMSE between EnergyPlus and ANN models are 1.53% and 7.82%, respectively. It is concluded that the profiling system requires minimalistic inputs and provides accurate energy performance assessment of a given building.