Harnessing Machine Learning for optimal solar-energy deployment in the UK.
This is previous project I developed with other the engineering teammembers at the amazing Alten Innovation Lab!
Solar Savvy is an interactive web application that predicts how much solar power can be generated on rooftops, car parks, or open fields anywhere in the United Kingdom.
It combines state-of-the-art computer-vision models with LiDAR elevation data, NASA & Copernicus irradiance feeds, and Google Maps imagery to give users precise, actionable estimatesβempowering individuals, businesses, and local authorities to accelerate renewable-energy adoption.
| Capability | How it helps |
|---|---|
| User-friendly web UI (Flask + HTML/CSS/JS) for address search and interactive mapping | Zero-install, intuitive experience |
| Three specialised ML pipelines (rooftops, car parks, fields) | Tailored accuracy for each land-use type |
| Real-time navigation & polygon drawing tools | Quickly explore large areas or focus on custom regions |
| Automatic roof-segment selection | Compare individual roof facets for optimal panel placement |
| Panel-type comparison (e.g. Thin-film vs Monocrystalline) | Instant what-if analysis by adjusting efficiency Ξ· |
- Imagery Fetch β 512 Γ 512 static satellite tile via Google Static Maps.
- Semantic β Instance Segmentation β U-Net model masks buildings; fine-tuned Bi-YOLO isolates individual roof planes.
- LiDAR Pitch Correction β UK EA 1 m DSM provides height values; PCA sets panel orientation.
- Panel-Placement Algorithm β
shapelypacks standard 1.7 m Γ 1 m rectangles, scaled by pitch. - Solar-Yield Calculation β Copernicus CAMS & NASA POWER irradiance feeds plugged into
(P = A \times \eta \times G \times T).
| Task | Architecture | Dataset(s) / Source |
|---|---|---|
| Roof mask | U-Net | Massachusetts Buildings, Inria Aerial Imagery |
| Roof plane | Bi-YOLO fine-tune | Johann Lussange 3D roof planes repo |
| Field segmentation | Segment Anything (SAM) | Custom field masks + SAM prompt-tuning |
| Car-park | SAM + post-filter | Augmented 4k dataset, IoU = 87% |
- Vision Transformers to replace Bi-YOLO for crisper boundaries
- Batch tile processing API for utility-scale feasibility studies
- Extend to wind-turbine siting with mesoscale weather models
This project is owned by Alten Group
- Project conceived and led at ALTEN UK Innovation Lab by
Daniel Ennis (Team Lead), Jonathan McMurtry (PM), and the engineering team. - Built on open-source work from Meta, NASA, ESA CAMS, and the UK Environment Agency.