Prediction of energy performance of residential buildings using regularised neural models

Abstract

Human habitats are one of the major consumers of energy. Therefore, in the current age of increasing carbon dioxide footprints, analysing energy efficiency of a building is important and is the subject of the current study. Machine-learning-based artificial neural network (ANN) approaches are used in the current study to investigate building energy performance. Eight parameters – relative compactness, surface area, wall area, roof area, overall height and orientation of the building, as well as the glazing area and its distribution – are selected as the input parameters and heating and cooling loads (CLs) as the output parameters. The network prediction capability was checked by comparing the predictions of the ANN architecture with the benchmark test case. A well-trained and validated ANN is used to predict 96 conditions by varying glazing area and glazing area distribution. The ANN is found to capture the physics efficiently. This study revealed that there is a significant potential to improve the energy efficiency of a building and the maximum saving in the CL can be as high as 20.67% for a fraction of the glazing areas equal to 0.15 if the glazing area distribution is 32.5% in the north and 22.5% each in the east, south and west.