Benchmarking energy performance of building envelopes through a selective residual-clustering approach using high dimensional dataset

Benchmarking energy performance of existing residential buildings' envelopes remains a challenge due to the complex physical and non-physical interacting factors of buildings. Regression analysis with sufficient data samples can be attractive for benchmarking application due to its capability in neutralizing the effects of noise variables. However, multicollinearity effects among explanatory variables often lead to unreliable regression models, especially in cases of high-dimensional variables. Principal Component Regression can transform co-linear variables via principal component analysis to orthogonal components and simultaneously has the neutralization function of linear regression analysis of high dimensional dataset. A new benchmarking method is developed using multivariate linear regression analysis with principal component analysis to address the multicollinearity risk with high dimensional dataset. The method was applied to datasets of a real project. The results indicate that Principal Component Regression is able to address multicollinearity risk, through using fewer orthogonal principal components that are linear combinations of original variables. The benchmarking outcome using this method is validated through infrared thermography validation. The benchmarking result is superior to that of the traditional statistical rating method using average energy consumption of buildings.