Abstract:Nuclear fusion power delivered by magnetic-confinement tokamak reactors holds the promise of sustainable and clean energy<a href="/articles/s41586-019-1116-4#ref-CR1">1</a>. The avoidance of large-scale plasma instabilities called disruptions within these reactors<a href="/articles/s41586-019-1116-4#ref-CR2">2</a>,<a href="/articles/s41586-019-1116-4#ref-CR3">3</a> is one of the most pressing challenges<a href="/articles/s41586-019-1116-4#ref-CR4">4</a>,<a href="/articles/s41586-019-1116-4#ref-CR5">5</a>, because disruptions can halt power production and damage key components. Disruptions are particularly harmful for large burning-plasma systems such as the multibillion-dollar International Thermonuclear Experimental Reactor (ITER) project<a href="/articles/s41586-019-1116-4#ref-CR6">6</a> currently under construction, which aims to be the first reactor that produces more power from fusion than is injected to heat the plasma. Here we present a method based on deep learning for forecasting disruptions. Our method extends considerably the capabilities of previous strategies such as first-principles-based<a href="/articles/s41586-019-1116-4#ref-CR5">5</a> and classical machine-learning<a href="#ref-CR7">7</a>,<a href="#ref-CR8">8</a>,<a href="#ref-CR9">9</a>,<a href="#ref-CR10">10</a>,<a href="/articles/s41586-019-1116-4#ref-CR11">11</a> approaches. In particular, it delivers reliable predictions for machines other than the one on which it was trained—a crucial requirement for future large reactors that cannot afford training disruptions. Our approach takes advantage of high-dimensional training data to boost predictive performance while also engaging supercomputing resources at the largest scale to improve accuracy and speed. Trained on experimental data from the largest tokamaks in the United States (DIII-D<a href="/articles/s41586-019-1116-4#ref-CR12">12</a>) and the world (Joint European Torus, JET<a href="/articles/s41586-019-1116-4#ref-CR13">13</a>), our method can also be applied to specific tasks such as prediction with long warning times: this opens up the possibility of moving from passive disruption prediction to active reactor control and optimization. These initial results illustrate the potential for deep learning to accelerate progress in fusion-energy science and, more generally, in the understanding and prediction of complex physical systems.

Parsimonious statistical techniques for the detection of drifts toward dangerous operational conditions in tokamaks

Dynamic Optimization of Open-Loop Input Signals for Ramp-Up Current Profiles in Tokamak Plasmas

Plasma Models for Real-Time Control of Advanced Tokamak Scenarios

A hybrid physics/data-driven logic to detect, classify, and predict anomalies and disruptions in tokamak plasmas

Receding-horizon optimal control of the current profile evolution during the ramp-up phase of a tokamak discharge

Dynamic Optimization Of Tokamak Plasmas Via Control Parameterization And The Time-Scaling Transformation

Optimal Tracking Control of Current Profile in Tokamaks

Ramp-Up-Phase Current-Profile Control of Tokamak Plasmas Via Nonlinear Programming

Data-Driven Modeling and Feedback Tracking Control of the Toroidal Rotation Profile for Advanced Tokamak Scenarios in Diii-D

Vertical instability forecasting and controllability assessment of multi-device tokamak plasmas in DECAF with data-driven optimization

Investigation of Machine Learning Techniques for Disruption Prediction Using JET Data

Predicting disruptive instabilities in controlled fusion plasmas through deep learning

A statistical approach for the automatic identification of the start of the chain of events leading to the disruptions at JET

Recent Progress on Deep Learning-Based Disruption Prediction Algorithm in HL-2A Tokamak

Real-time disruption prediction in multi-dimensional spaces leveraging diagnostic information not available at execution time

Adaptive Anomaly Detection Disruption Prediction Starting from First Discharge on Tokamak

Automated experimental design of safe rampdowns via probabilistic machine learning

Tokamak divertor plasma emulation with machine learning

MHD spectrogram contribution to disruption prediction using Convolutional Neural Networks

ST40 electromagnetic predictive studies supported by machine learning applied to experimental database

Active Disruption Avoidance and Trajectory Design for Tokamak Ramp-downs with Neural Differential Equations and Reinforcement Learning