Back to search

Georgia Institute of Technology

A Machine Learning Workflow for Localized Weather Integration in Building Energy Modeling

Abstract

dc:description.abstract

This thesis develops an ML-based workflow for bringing localized weather data into building energy modeling at the campus scale. Conventional BEM uses a single airport-based Typical Meteorological Year (TMY3) file to represent the weather across an entire city, missing the spatial variation that occurs within dense urban areas. The workflow addresses this by replacing the temperature and humidity fields in a standard TMY3 file with ML-predicted values at specific campus locations, generating localized EnergyPlus Weather (EPW) files at approximately 4\,m spatial resolution. The workflow is tested on the Georgia Tech campus. Five weather scenarios are built for a shoebox energy model: the conventional airport TMY3 file, on-site observations and ML predictions at a weather station location (used as a validation check), and ML predictions at two thermally extreme points on campus, one hotspot and one cool-spot. Five decarbonization strategies are simulated under each scenario to see how localized weather changes their predicted effectiveness. The results show that the effect of localized weather data is not uniform across strategies. HVAC plant optimization reduces cooling energy by 20 to 30\% depending on the scenario, but the absolute savings are largest at the hotspot. Adaptive setpoint control and daylighting are more effective at the cool-spot than at the hotspot, because internal gains contribute more to cooling demand in cooler locations. The east-west WWR reduction and envelope retrofits produce nearly identical savings across all scenarios, because their effects depend on geometry and conduction physics rather than on local temperature. A retrofit analysis based on TMY3 alone would miss these differences and treat every building on the campus as thermally identical.

Degree

thesis:*
Name thesis:degree_name
Architecture, MS
Department thesis:degree_department
Architecture
Grantor
Georgia Institute of Technology
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Shih, Han-Syun
Advisor dc:contributor.advisor
  • Kastner, Patrick
Committee members dc:contributor.committeemember
  • Milz, Johannes
  • Marshall, Tyrone

Rights

Language dc:language.iso
English

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/81611

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Shih, Han-Syun. A Machine Learning Workflow for Localized Weather Integration in Building Energy Modeling. Georgia Institute of Technology, 2026. https://hdl.handle.net/1853/81611