Reconstruction-free magnetic control of DIII-D plasma with deep reinforcement learning
Authors:
G. F. Subbotin,
D. I. Sorokin,
M. R. Nurgaliev,
A. A. Granovskiy,
I. P. Kharitonov,
E. V. Adishchev,
E. N. Khairutdinov,
R. Clark,
H. Shen,
W. Choi,
J. Barr,
D. M. Orlov
Abstract:
Precise control of plasma shape and position is essential for stable tokamak operation and achieving commercial fusion energy. Traditional control methods rely on equilibrium reconstruction and linearized models, limiting adaptability and real-time performance. Here,the first application of deep reinforcement learning (RL) for magnetic plasma control on the mid-size DIII-D tokamak is presented, de…
▽ More
Precise control of plasma shape and position is essential for stable tokamak operation and achieving commercial fusion energy. Traditional control methods rely on equilibrium reconstruction and linearized models, limiting adaptability and real-time performance. Here,the first application of deep reinforcement learning (RL) for magnetic plasma control on the mid-size DIII-D tokamak is presented, demonstrating a nonlinear approach that improves robustness and flexibility across plasma scenarios. Using the Soft Actor-Critic algorithm, this method eliminates the need for equilibrium reconstruction, enabling high-speed control execution and scalability on larger fusion devices. NSFsim, a 2D Grad-Shafranov equilibration solver with a circuit equation and a 1D transport solver, is used to train the agent. Its capability of reproducing the kinetic parameter evolution alongside magnetic equilibria evolution appears to be an essential factor significantly affecting control quality. RL-based controllers demonstrated robust magnetic control in experimental application at DIII-D, preserving control performance in transient events during plasma discharges, and reaching target parameters from the first discharge without additional tuning or modifications. The approach itself has significant generalization potential across devices and targets. This work represents a step toward AI-driven, real-time plasma control, advancing the feasibility of next-generation fusion reactors.
△ Less
Submitted 16 June, 2025;
originally announced June 2025.
Reconstructing the Plasma Boundary with a Reduced Set of Diagnostics
Authors:
M. S. Stokolesov,
M. R. Nurgaliev,
I. P. Kharitonov,
E. V. Adishchev,
D. I. Sorokin,
R. Clark,
D. M. Orlov
Abstract:
This study investigates the feasibility of reconstructing the last closed flux surface (LCFS) in the DIII-D tokamak using neural network models trained on reduced input feature sets, addressing an ill-posed task. Two models are compared: one trained solely on coil currents and another incorporating coil currents, plasma current, and loop voltage. The model trained exclusively on coil currents achi…
▽ More
This study investigates the feasibility of reconstructing the last closed flux surface (LCFS) in the DIII-D tokamak using neural network models trained on reduced input feature sets, addressing an ill-posed task. Two models are compared: one trained solely on coil currents and another incorporating coil currents, plasma current, and loop voltage. The model trained exclusively on coil currents achieved a mean point displacement of 0.04 m on a held-out test set, while the inclusion of plasma current and loop voltage reduced the error to 0.03 m. This comparison highlights the trade-offs between input feature complexity and reconstruction accuracy, demonstrating the potential of machine learning algorithms to perform effectively in data-limited environments, such as those expected in Fusion Power Plants (FPP) due to diagnostic constraints imposed by the presence of blankets and shielding.
△ Less
Submitted 15 May, 2025;
originally announced May 2025.
