Researchers from Stanford University have developed a deep learning system, called DeepSolar, to detect solar panel installation rates to understand socioeconomic factors relating to solar panel deployment.
The team of researchers included Stanford engineers Arun Majumdar and Ram Rajagopal, together with grad students Jiafan Yu and Zhecheng Wang.
The team integrated machine learning (ML) technology, to develop the deep learning and predictive model, DeepSolar. The new system was provided with approximately 370,000 satellite images – each covering about 100ft by 100ft. Each image was categorized into two classes – of having a solar panel installed or not. Following from that the DeepSolar system learned to identify features associated with solar panels (e.g., color, texture and size).
Visit DATAx Singapore on March 5–6, 2019
The researchers fed the system data, from the US census data, such as environmental and socioeconomic factors to identify trends and factors of solar panel deployment.
The solar installation report of the "contiguous US with exact locations and sizes of solar panel" was carried out over a month. DeepSolar was fed with more than "one billion image tiles covering all urban areas as well as locations with reasonable nighttime lights to construct the first complete solar installation profile".
The report found 1.47 million installations.
According to the researchers, the model has a precision rate of 93.1% in residential areas and 93.7% in non-residential areas and the new system provides a more enhanced accuracy rate than previous reports.
The collated data is the first publicly available, accurate solar panel installation database in the contiguous US, according to the researchers.
"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," the team added.