DeepSolar

The deployment of solar energy on the energy supply side plays a critical role in decarbonizing the energy sector and mitigating climate change. Compared to conventional fossil fuel energy, solar energy is more intermittent and decentralized. Individual households, traditionally defined as energy consumers, can now install solar photovoltaic (PV) panels on their rooftops to produce electricity on their own and thus become energy “prosumers”. Furthermore, the electricity generated by solar PV panels can flow back to the electrical grid and, if not well managed, potentially cause instability to the entire power system.Ssolar PV systems, therefore, are more challenging to document, coordinate, and manage than conventional centralized power generators.

In this project, we developed DeepSolar, a deep learning framework that identifies geolocations, sizes, and subtypes of solar PV panels from satellite and aerial images. Relying on its high accuracy and scalability, we further constructed a comprehensive high-fidelity solar deployment database for the contiguous U.S. Due to the scarcity of ground truth annotations of solar panel sizes, we developed a weakly-supervised approach to both identify solar panels and estimated their sizes with a single neural network using only image-level class labels as supervision. We deployed the model to over 1 billion satellite image tiles across the contiguous U.S. in mid-2017 and constructed a geospatial database containing 1.47 million solar systems which have been further classified into different subtypes including residential PVs, commercial PVs, and utility-scale PVs.

We demonstrated its value by discovering that residential solar deployment density peaks at a population density of 1,000 capita/mile2, increases with annual household income asymptoting at $150k, and has an inverse correlation with the Gini index representing income inequality. We uncovered a solar radiation threshold (4.5 kWh/m2/day) above which the solar deployment is “triggered”. Furthermore, we also built an accurate machine learningbased predictive model to estimate the solar deployment density at the census tract level. We offer the DeepSolar database as a publicly available resource for researchers, utilities, solar developers, and policymakers to further uncover solar deployment patterns, build comprehensive economic and behavioral models, and ultimately support the adoption and management of solar electricity.

We are now working on updating this database to 2022. Stay tuned!

Publication:

  • Jiafan Yu*, Zhecheng Wang*, Arun Majumdar, and Ram Rajagopal (2018). DeepSolar: A Machine Learning Framework to Efficiently Construct a Solar Deployment Database in the United States. Joule. (* Equal contribution)
    (link)

(project website) (code - Tensorflow legacy) (code - PyTorch)

Media coverage:

(Stanford News) (MIT Technology Review) (TechCrunch) (Gizmodo) (Earth.com) (Fast Company) (PBS NewsHour) (La Vanguardia (Barcelona daily))