Accelerated Deep Learning Discovery in Fusion Energy Science

PI William Tang, Princeton Plasma Physics Laboratory
Tang Aurora ESP

The team’s Fusion Recurrent Neural Network uses convolutional and recurrent neural network components to integrate both spatial and temporal information for predicting disruptions in tokamak plasmas. Image: Julian Kates-Harbeck, Harvard University; Eliot Feibush, Princeton Plasma Physics Laboratory

Project Summary

Machine learning and artificial intelligence can demonstrably accelerate scientific progress in predictive modeling for grand challenge areas such as the quest for clean energy via fusion power. This project seeks to expand modern convolutional and recurrent neural net software to carry out optimized hyperparameter tuning on exascale supercomputers to make strides toward validated prediction and associated mitigation of large-scale disruptions in burning plasmas such as ITER.

Project Description

Building the scientific foundations needed to deliver accurate predictions in key scientific domains of current interest can best be accomplished by engaging modern big-data-driven statistical methods featuring machine/deep learning a and artificial intelligence. An especially time-urgent problem facing the development of a fusion energy reactor is the need to reliably predict and avoid large-scale major disruptions in magnetically-confined tokamak systems such as the burning plasma ITER device. Significantly improved methods of prediction with better than 95% predictive capability are required to provide sufficient advanced warning for disruption avoidance and mitigation strategies to be effectively applied before critical damage can be done to ITER. This formidable task demands accuracy beyond the near-term reach of hypothesis-driven /first-principles extreme-scale computing simulations that dominate current research and development in the field. This project aims to leverage leadership-class computers to greatly enhance the training of neural nets to predict the onset of a disruption at least 30 milliseconds or more in advance of the actual event—the time required to implement mitigation strategies in a real experiment after receiving an alarm.

Related Articles

Ensuring the portability of deep learning software to explore fusion energy on Aurora

Preparing for exascale: Eliminating disruptions on the path to sustainable fusion energy

Project Type