Using AI for cheaper and more efficient solar power
summary
End-to-end pipeline that turns a single aerial image into panel layouts and yearly yield estimates using rot net for roof alignment and multiple segmentation CNNs for facets, obstacles and shadow masks trained on about 100k labeled images (training ~18 hours for 500 epochs) and with an estimated 4% manual labeling error. Energy is validated via PVWatts data for latitude/longitude. Built product automates quotes and matching but still needs UX work, marketplace integration and text/speech analytics to improve conversion.
AI-written summary · about
chapters (10)
- 0:00AI in solar power solutions
- 6:18Renewable energy and AI integration
- 9:28Solar power and AI's role
- 11:39Manual solar panel installation challenges
- 14:13Rooftop solar analysis algorithms
- 15:40Rooftop segmentation and alignment
- 18:04Shadows and obstacles in rooftop analysis
- 20:36AI model postprocessing and validation
- 22:44Energy production estimation with AI
- 24:55AI-driven solar challenges
related talks
11mUsing AI to Accelerate the Smart Green Transition
18mDetection of solar panels based on aerial images of the city of...
37mEstimating solar production w/ distributed deep learning from smart meter data
36mAI-Powered Real-Time Quality Control
28mGeoAI: Evolution of AI-Based Tree/Vegetation Detection Models Using Imagery
50mAzure AI, Power new possibilities for every organization
19mHow we leverage AI to build great maps at massive scale
40mVertex AI: Pipelines for your MLOps workflows