AI-Based Urban Air Impact Analysis System for Factory and Power Plant Emissions (GAIA Project)
GAIA is a university capstone prototype for analysing the potential urban air-quality effects of factory and power-plant emissions through a hybrid physical and machine-learning approach.
The team used seven years (2018β2024) of Tele-Monitoring System (TMS) time-series data and combined the AERMOD physical-diffusion model with a Transformer-based architecture.
The prototype pipeline combines TMS, Korea Meteorological Administration, and AirKorea data; applies cleaning, gap-filling, and temporal feature engineering; and uses PyTorch-based modelling to examine relationships among weather and pollutant variables.
A Streamlit dashboard presents scenario inputs, spatial plume views, and observed-versus-predicted comparisons using Folium, GeoPandas, and Xarray.
The prototype was designed to explore decision-support use cases. IoT integration and nationwide or climate-scenario coverage remain proposed extensions rather than completed features.
Key Components
- Hybrid model: AERMOD physical diffusion + Transformer deep learning
- Data pipeline: TMS/KMA/AirKorea APIs β cleaning, gap filling, feature engineering
- Uncertainty modeling: Gaussian Diffusion for plume dynamics
- Visualization: Streamlit dashboard, Folium/GeoPandas/Xarray spatial views