AI-based model makes predicting fusion profiles faster

June 28, 2021, 7:00AMNuclear News

PPPL physicist Dan Boyer. (Photo: Amber Boyer/Kiran Sudarsanan)

Researchers at the Department of Energy’s Princeton Plasma Physics Laboratory are using machine learning to predict electron density and pressure profile shapes on the National Spherical Torus Experiment-Upgrade (NSTX-U), the flagship fusion facility at PPPL that is currently under repair.

The hope is that such predictions, generated by artificial neural networks, could improve the ability of NSTX-U researchers to optimize the components of experiments that heat and shape the fusion plasma.

“This is a step toward what we should do to optimize the actuators,” said PPPL physicist Dan Boyer, author of the paper, “Prediction of electron density and pressure profile shapes on NSTX-U using neural networks,” published by Nuclear Fusion, a journal of the International Atomic Energy Agency. “Machine learning can turn historical data into a simple model that we can evaluate quickly enough to make decisions in the control room or even in real time during an experiment.”

The model: Boyer and coauthor Jason Chadwick, an undergraduate student at Carnegie Mellon University who was a participant in the Science Undergraduate Laboratory Internship program at PPPL last summer, tested neural network forecasts using 10 years of data for NSTX, the forerunner of NSTX-U, and the 10 weeks of operation of NSTX-U. The machine learning tests correctly predicted the distribution of pressure and density of the electrons in fusion plasmas, two critical but difficult-to-forecast parameters.

“While physics-based models for predicting electron pressure and density exist,” Boyer said, “they are not appropriate for real-time decision-making. They take way too long to calculate and are not as accurate as we need them to be.”

Addressing both issues, the new model, once it is trained, takes less than one thousandth of a second to evaluate. The speed of the resulting model could make it useful for many real-time applications, Boyer said.

Limitations: Because the model is trained on historically observed data, however, it cannot make predictions about new operating points with high accuracy, Boyer added. This limitation will be addressed by adding the results of physics-based model predictions to the training data and developing techniques of adapting the model as new data become available.

Related Articles

High-energy physics gets DOE funding

August 6, 2021, 7:01AMNuclear News

The Department of Energy’s Office of Science (DOE-SC) on August 2 announced a plan to provide $100 million over the next four years for university-based research on a range of high-energy...