Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning

Abstract The tokamak offers a promising path to fusion energy, but disruptions pose a major economic risk, motivating solutions to manage their consequence. This work develops a reinforcement learning approach to this problem by training a policy to ramp-down the plasma current while avoiding limits...

Full description

Saved in:
Bibliographic Details
Main Authors: Allen M. Wang, Cristina Rea, Oswin So, Charles Dawson, Darren T. Garnier, Chuchu Fan
Format: Article
Language:English
Published: Nature Portfolio 2025-06-01
Series:Communications Physics
Online Access:https://doi.org/10.1038/s42005-025-02146-6
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849469913417121792
author Allen M. Wang
Cristina Rea
Oswin So
Charles Dawson
Darren T. Garnier
Chuchu Fan
author_facet Allen M. Wang
Cristina Rea
Oswin So
Charles Dawson
Darren T. Garnier
Chuchu Fan
author_sort Allen M. Wang
collection DOAJ
description Abstract The tokamak offers a promising path to fusion energy, but disruptions pose a major economic risk, motivating solutions to manage their consequence. This work develops a reinforcement learning approach to this problem by training a policy to ramp-down the plasma current while avoiding limits on a number of quantities correlated with disruptions. The policy training environment is a hybrid physics and machine learning model trained on simulations of the SPARC primary reference discharge (PRD) ramp-down, an upcoming burning plasma scenario which we use as a testbed. To address physics uncertainty and model inaccuracies, the simulation is massively parallelized on GPU with randomized physics parameters during policy training. The trained policy is then run in feedback on a transport simulator as a demonstration. We also directly address the crucial issue of control validation by demonstrating that a constraint-conditioned policy can be a trajectory design assistant that designs a library of feed-forward trajectories to handle different physics conditions and user constraint settings, a promising approach for the sensitive context of burning plasma tokamaks. Finally, we demonstrate that the training environment can be a useful platform for feed-forward optimization approaches by optimizing feed-forward trajectories that are robust to physics uncertainty.
format Article
id doaj-art-0ca3738deb254c2b99fa24a037449f55
institution Kabale University
issn 2399-3650
language English
publishDate 2025-06-01
publisher Nature Portfolio
record_format Article
series Communications Physics
spelling doaj-art-0ca3738deb254c2b99fa24a037449f552025-08-20T03:25:19ZengNature PortfolioCommunications Physics2399-36502025-06-018111310.1038/s42005-025-02146-6Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learningAllen M. Wang0Cristina Rea1Oswin So2Charles Dawson3Darren T. Garnier4Chuchu Fan5Plasma Science and Fusion Center, Massachusetts Institute of TechnologyPlasma Science and Fusion Center, Massachusetts Institute of TechnologyLaboratory for Information and Decision Systems, Massachusetts Institute of TechnologyLaboratory for Information and Decision Systems, Massachusetts Institute of TechnologyPlasma Science and Fusion Center, Massachusetts Institute of TechnologyLaboratory for Information and Decision Systems, Massachusetts Institute of TechnologyAbstract The tokamak offers a promising path to fusion energy, but disruptions pose a major economic risk, motivating solutions to manage their consequence. This work develops a reinforcement learning approach to this problem by training a policy to ramp-down the plasma current while avoiding limits on a number of quantities correlated with disruptions. The policy training environment is a hybrid physics and machine learning model trained on simulations of the SPARC primary reference discharge (PRD) ramp-down, an upcoming burning plasma scenario which we use as a testbed. To address physics uncertainty and model inaccuracies, the simulation is massively parallelized on GPU with randomized physics parameters during policy training. The trained policy is then run in feedback on a transport simulator as a demonstration. We also directly address the crucial issue of control validation by demonstrating that a constraint-conditioned policy can be a trajectory design assistant that designs a library of feed-forward trajectories to handle different physics conditions and user constraint settings, a promising approach for the sensitive context of burning plasma tokamaks. Finally, we demonstrate that the training environment can be a useful platform for feed-forward optimization approaches by optimizing feed-forward trajectories that are robust to physics uncertainty.https://doi.org/10.1038/s42005-025-02146-6
spellingShingle Allen M. Wang
Cristina Rea
Oswin So
Charles Dawson
Darren T. Garnier
Chuchu Fan
Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
Communications Physics
title Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
title_full Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
title_fullStr Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
title_full_unstemmed Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
title_short Active ramp-down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
title_sort active ramp down control and trajectory design for tokamaks with neural differential equations and reinforcement learning
url https://doi.org/10.1038/s42005-025-02146-6
work_keys_str_mv AT allenmwang activerampdowncontrolandtrajectorydesignfortokamakswithneuraldifferentialequationsandreinforcementlearning
AT cristinarea activerampdowncontrolandtrajectorydesignfortokamakswithneuraldifferentialequationsandreinforcementlearning
AT oswinso activerampdowncontrolandtrajectorydesignfortokamakswithneuraldifferentialequationsandreinforcementlearning
AT charlesdawson activerampdowncontrolandtrajectorydesignfortokamakswithneuraldifferentialequationsandreinforcementlearning
AT darrentgarnier activerampdowncontrolandtrajectorydesignfortokamakswithneuraldifferentialequationsandreinforcementlearning
AT chuchufan activerampdowncontrolandtrajectorydesignfortokamakswithneuraldifferentialequationsandreinforcementlearning