Explainable Machine Learning Reveals Urban Morphology Impact on Residential Energy Consumption

Strategic Insights -

Researchers at the University of Hong Kong have developed an explainable machine learning framework to understand the relationship between urban morphology and building energy consumption in Dongguan, China. The study integrates large-scale smart meter records with 3D building footprints to construct high-resolution datasets, demonstrating superior predictive performance over comparative methods. The analysis reveals significant correlations between building shape coefficients, floor area ratios, and building height, as well as threshold effects for key parameters, providing evidence-based strategies for climate-responsive urban planning.

Key Takeaways:

  • The study employs an explainable machine learning framework to analyze the impact of urban morphology on building energy consumption in Dongguan, China.
  • The research integrates large-scale anonymized smart meter records with 3D building footprints to construct high-resolution datasets.
  • The analysis demonstrates superior predictive performance (R-2 = 0.646) over comparative machine learning methods using XGBoost modeling.
  • SHapley Additive exPlanations (SHAP) analysis reveals significant correlations between building shape coefficients and energy consumption.
  • Threshold effects are identified for six key parameters: SCB, FAR, BCR, BH, PD, and NDVI, delineating inflection points between positive and negative impacts on energy consumption.
  • Urban design strategies optimizing building shape coefficients (SCB ≤ 0.22) are most effective in reducing energy consumption.
  • The study provides a quantifiable assessment framework and evidence-based strategies for climate-responsive urban planning to advance sustainable energy transitions.

Statistics:

  • The research demonstrates superior predictive performance (R-2 = 0.646) over comparative machine learning methods.
  • The analysis reveals significant correlations between building shape coefficients (SCB) and energy consumption (BECI).
  • Threshold effects are identified for six key parameters: SCB (0.38), FAR (1.5), BCR (0.32), BH (15 m), PD (8 poi/km2), and NDVI (0.22).
  • The study provides a quantifiable assessment framework for climate-responsive urban planning.

Sources:

  • Assessing Urban Morphology Effects On Residential Building Electricity Consumption Via Explainable Machine Learning: Evidence From China's Hot Summer and Warm Winter Zone. Energy and Buildings, 2025;345.
  • University of Hong Kong, Faculty of Architecture, Division of Landscape Architecture, Future Urban & Sustainable Environm Fuse Lab.
  • National Science Fund for Distinguished Young Scholars, National Natural Science Foundation of China (NSFC).
  • Elsevier Science Sa, PO Box 564, 1001 Lausanne, Switzerland (Energy and Buildings - www.journals.elsevier.com/energy-and-buildings/).
  • Xinxin Wu, University of Hong Kong, Faculty of Architecture, Division of Landscape Architecture, Future Urban & Sustainable Environm Fuse Lab, Hong Kong, People's Republic of China.