Spatial Variation of Urban Building Energy Analysis

Abstract

Building energy use in various parts of a city can be visualized spatially using GIS (geographical information system) programs. When doing so, one would expect spatial variations due to socioeconomic and/or physical factors. This paper investigates the local spatial variations of domestic energy use intensity (both gas and electricity) across London. Three local regression methods are considered: geographically weighted regression (GWR), mixed model, and Bayesian hierarchical computation. The results indicate that the Bayesian hierarchical model can produce reliable results compared to the other two frequentist methods. The ‘Stan’, a new full Bayesian statistical inference with Hamiltonian Monte Carlo (HMC), can reduce computational cost compared to the commonly used sampling methods (random walk Metropolis or Gibbs sampling). The Stan Bayesian method is expected to have more widely applications in energy assessment of urban buildings.