As cities like Philadelphia strive to meet ambitious greenhouse gas
reduction goals, predicting how zoning and development decisions will impact
building energy use and emissions is a major challenge. A new study led by
Simi
Hoque, PhD, professor of civil, architectural and environmental
engineering, uses machine learning to address this problem.
The model developed by Hoque’s team leverages a deep-learning program,
called Extreme Gradient Boosting (XGBoost), to forecast neighborhood-level
energy use based on housing features, demographics and socioeconomics. A
Shapley analysis then pinpoints which factors most influenced the
predictions.
“Machine learning is well equipped to handle this challenge because the
models can iteratively learn and improve through training despite data
limitations,” Hoque explained.
In a hypothetical scenario forecasting energy use through 2045, the
interpretable model suggested zoning policies like upzoning may increase
residential energy use in some areas. For commercial buildings, square
footage and employee count were key drivers.
While further testing is required, the research demonstrates machine
learning’s potential to inform zoning and development decisions that align
with emissions goals. By surfacing high-impact variables, the approach could
help cities like Philadelphia customize energy policies as they chart a path
toward carbon neutrality.