Stanford University researchers created machine learning algorithm Deep Solar to count current number of solar panel installations in U.S.

Briefing

Stanford University researchers created machine learning algorithm Deep Solar to count current number of solar panel installations in U.S.

January 15, 2019

Briefing

  • Deep Solar – Stanford University scientists developed machine learning algorithm called Deep Solar that was able to count how many solar panel installations are in U.S. by analyzing one billion satellite images in one month
  • 47 Million Installations – Identified 1.47 million solar panels were installed in country, including households, businesses, and utility-owned solar power plants, more than latest estimate of 1.02 million, with researchers excluding sparsely populated areas and estimating 5% installation rate in areas not covered
  • AI Training – Trained algorithm to identify solar panels in 370,000 images, with each image measuring 100 feet by 100 feet
  • More Accurate – Achieved 93% accuracy in spotting solar panels, better than previous models
  • Applications – Data could be useful for utility companies helping them estimate and balance supply and demand, as well as regulators, solar panel sellers, and others
  • Next Steps – Include estimating solar installations in rural areas and other countries, as well as adding feature that will enable algorithm to calculate solar panel’s angle and orientation, which can help measure amount of power generated

Accelerator

Business Model and Practices

Business Model
and Practices

Sector

Energy, Information Technology, Utilities

Organization

Stanford University

Source

Original Publication Date

December 19, 2018

Leave a comment