Many urban energy use modeling tools and methods have been developed to understand energy use in cities, but often have limitations in aggregating across multiple scales and end-uses, which adversely affects accuracy and utility. Increased data availability and developments in machine learning (ML) provide new possibilities for improving the accuracy and complexity of urban energy use models. This paper presents an integrated framework for urban energy use modeling (UEUM) that localizes energy performance data, considers urban socio-spatial context, and captures both urban building operational and transportation energy use through a bottom-up data-driven approach. The framework employs ML techniques for building operational energy use modeling at the urban scale with a travel demand model for transport energy use prediction. The framework is demonstrated using Chicago as a case study because it has significant variations in urban spatial patterns across its neighborhoods and it provides publicly available data that are essential for the framework. Results for Chicago suggest that, among the tested algorithms, k-nearest neighbor shows the best overall performance in terms of accuracy for a single-output model (i.e., for building or transportation energy use separately) and artificial neural network algorithm is the most accurate for the integrated model (i.e., building and transportation energy use combined). Exploratory analysis demonstrates that the urban attributes examined herein explain 41% and 96% of the variance in building and transportation energy use intensity, respectively. The UEUM framework has the potential to aid designers, planners, and policymakers in predicting urban energy use and evaluating robust theories and alternative scenarios for energy-driven planning and design.
All Science Journal Classification (ASJC) codes
- Building and Construction
- Mechanical Engineering
- Management, Monitoring, Policy and Law