A new forecasting model that predicts nuclear reactor outages using machine learning has been developed by researchers from the U.S. Department of Energy’s (DOE) Argonne National Laboratory and Purdue University. The model predicts weekly outages of commercial light water reactors, with the aim of increasing accuracy to estimate operations and maintenance costs of nuclear facilities by predicting unplanned shutdowns, which are a major source of revenue loss.
Another goal of this research is to help developers of small modular reactors with projection of their operations and maintenance costs, allowing for mitigation strategies to be integrated into the designs. When commercialized, small modular reactors are expected to deliver carbon-free electricity at a lower cost compared to existing light water reactors.
Konstantinos Prantikos, a Purdue University graduate research assistant and an Argonne visiting graduate student; Alexander Heifetz, an Argonne principal electrical engineer; and Lefteri H. Tsoukalas, professor of nuclear engineering and founding director of Purdue’s AI Systems Lab, presented their findings at the American Nuclear Society Winter Meeting & Expo 2022 and were published in the Winter Meeting Transactions of the American Nuclear Society.
This work was supported in part by DOE, Advanced Research Projects Agency-Energy Generating Electricity Managed by Intelligent Nuclear Assets program, and by a Goldman Sachs Gives donation to Purdue University AI Systems Lab.