Project Name: Deciphering Degradation: Machine Learning on Real-World Performance Data
Funding Opportunity: Solar Energy Technologies Office Fiscal Year 2018 Funding Program (SETO FY2018)
SETO Research Area: Photovoltaics
Location: San Francisco, CA
SETO Award Amount: $1,250,000
Awardee Cost Share: $500,000
Principal Investigator: Adam Shinn

-- Award and cost share amounts are subject to change pending negotiations --

This project team is measuring degradation rates for a range of solar energy systems using a type of artificial intelligence called machine learning. The rates at which solar cells degrade, and thus lose efficiency, under different conditions is very difficult to discern, but this information is necessary in order to finance solar energy systems. An estimated degradation rate of 0.5% is currently used for financing purposes based on experience to date, but the uncertainty in this value leads to higher costs.

APPROACH

kWh Analytics will expand its data collection of residential solar energy systems to at least 100 new systems and add utility-scale systems to diversify the types and conditions of systems it tracks. The team will work with utilities to record system performance data from each system every 15 minutes, versus the usual once daily or monthly. This will allow the team to use machine learning to decouple the actual degradation rate, which manifests as a slow reduction in efficiency and therefore voltage over time, from other more sudden or short-lived effects, such as intermittent clouds or rainy days.

INNOVATION

Using machine learning to measure solar cell degradation rates is a new application of this powerful approach. By separating the cell degradation from other effects, this team will obtain the first reliable measurements of solar cell degradation rates. The reduced uncertainty in degradation rates will result in lower costs associated with financing of solar systems, driving down installation costs.