An AI-Driven Digital Twin for Predictive Energy Monitoring and Hierarchical Model Predictive Control of Smart Buildings
Honours Individual Project - Yiğit Sayar · University of Glasgow, School of Computing Science · March 2026
Supervised by Dr. Awais Shah and Harsh Vivek Shah
EquiTwin is a digital twin for smart buildings that monitors energy use, forecasts future building conditions, and controls heating and ventilation in a simulated environment. It was built and built for deployment at the Sir Alwyn Williams Building (SAWB), University of Glasgow.
Special thanks for my supervisors.
| Layer | Description |
|---|---|
| Home | Interactive BIM viewer with live sensor overlays and telemetry status cards |
| Dashboard | Historical time-series monitoring, anomaly review, and database inspection |
| Forecast | Model training, per-horizon accuracy rankings, and artefact inspection |
| Controller | Closed-loop MPC thermal model simulation with solver diagnostics and HVAC output |
Repeated sensor disruptions at SAWB (LoRaWAN timeouts, firmware issues) motivated three design responses: synthetic data generation preserving cross-variable physical structure; PSI drift detection flagging distribution shift before each training session; and graceful degradation ensuring monitoring remains operational under missing artefacts, solver failure, or sensor outage.
For setup instructions, see setup.md