Overview of the Project
A recent trend in PV research is to study and learn about PV system performance based
on real-world datastreams of distributed and utility scale PV systems, instead of only relying on
lab-based experimentation. Two major PV research challenges are to determine and control the
performance loss rate (PLR) of PV systems over their lifetimes, and to forecast their power
output up to 6 hours or 1 to 5 days.
A novel class of adaptive Graph Neural Network (GNN) models will enable us to devise
and develop a spatiotemporal-based learning framework for PV power plants that addresses these two challenges while simultaneously enriching the real-world datastreams automatically, using underutilized data sources for data imputation. The GNN models will automatically curate
spatiotemporal weather, irradiance, and power data from multiple PV systems and data sources
into a system topology-aware network. The GNN will determine PLR for specific systems and
module brands and forecast PV power output for specific systems and regions. In addition to
addressing PLR and power forecasting, the GNN model will be able to diagnose localization of
anomalies, such as hurricane or tornado tracts, and reliably evaluate robustness to malicious data
manipulation that can occur in cyber attacks.
Members and Collaborators
Roger H. French (Principal Investigator and Kyocera Professor, CWRU)
Laura S. Bruckman (Research Associate Professor, CWRU)
Yinghui Wu (Assistant Professor, CWRU and Staff Scientist, Pacific Northwest National Laboratory)
Mehmet Koyutürk (Andrew R. Jennings Professor of Computing Sciences, Electrical
Engineering & Computer Science, CWRU)
Jennifer L. Braid (Visiting Researcher, Sandia National Laboratories)
Jean-Nicolas Jaubert (Director, Product Reliability, Certification & System Performance, CanadianSolar)
David H. Meakin (Senior Manager Module Reliability, SunPower)
Mohamed Arafath Nihar (Department of Computer and Data Sciences, CWRU)
Alan Curran (Department of Materials Science and Engineering, CWRU)
Nick Steffan
Benjamin Spurgeon
Unfunded Team Members
SunPower
Brookfield Renewables
C2 Energy
SolarEdge
CanadianSolar
Acknowledgement
This material is based upon work supported by the U.S. Department of Energy’s Office of Energy Efficiency and Renewable Energy (EERE) under Solar Energy Technologies Office (SETO) Agreement Number DE-EE0009353.
