From Forecasting to Feedback: Evaluating Learned Dynamics for Tokamak Control
Abstract
Learned temporal models are moving beyond offline forecasting toward interactive environments for decision making, yet logged-trajectory accuracy need not predict closed-loop behavior under policies optimized in the learned model. This forecasting-control gap matters in tokamak magnetic control, where coupled plasma-circuit dynamics and actuator constraints can amplify errors through feedback. Using a calibrated free-boundary Grad-Shafranov solver, we collect four datasets in a 2x2 design over physics-parameter and current-reference coverage. At matched budgets of 500, 1k, 3k, and 5k rollouts, we train 16 action-conditioned Transformers; PPO policies optimized in the 5k surrogates are then evaluated without adaptation on six reference-solver conditions. Only joint-coverage D4 shows monotonically decreasing cross-condition mean error, whereas D1 improves in distribution but degrades out of distribution. Binary completion masks substantial closed-loop differences: D2-D4 each complete 83.3% of conditions, but reach on average 99.0%, 94.7%, and 90.5% of the evaluation horizon, while their worst conditions reach 94.2%, 68.2%, and 43.2%. Policies trained with action noise complete 5/24 matched conditions, compared with 19/24 under nominal training, and have a lower mean horizon-capped episode length. Based on one trained policy per configuration and one evaluation realization per condition, the results support combining prediction error, episode completion, condition-level duration, and tracking quality.
Related
- Eliminating Tokamak Disruptions with Feedback
- Extrapolating learned dynamics across bifurcations based on generalized synchronization
- The energetics of force errors in machine-learned molecular dynamics
- How I Learned to Stop Worrying and Love the Redfield Equation: Completely-Positive Resummation of Non-Markovian Dynamics
- Smite: A quasiclassical trajectory (QCT) program for bimolecular collisions and unimolecular dynamics on ab-initio, and machine-learned potential energy surfaces
BibTeX
@misc{shi2026from,
title = {{From Forecasting to Feedback: Evaluating Learned Dynamics for Tokamak Control}},
author = {Guoyang Shi and Siqi Ding and Xuanhe Wang and Pei Guo and Changquan Yu and Jiale Hu and Jiayi Zhi and Zhengyuan Chen and Jianguo Chen and Tianyuan Liu},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.03690},
eprint = {2611.03690},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.03690},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - From Forecasting to Feedback: Evaluating Learned Dynamics for Tokamak Control
AU - Guoyang Shi
AU - Siqi Ding
AU - Xuanhe Wang
AU - Pei Guo
AU - Changquan Yu
AU - Jiale Hu
AU - Jiayi Zhi
AU - Zhengyuan Chen
AU - Jianguo Chen
AU - Tianyuan Liu
PY - 2026
DA - 2026/11/01
PB - arXiv
DO - 10.48550/arxiv.2611.03690
UR - https://arxiv.org/abs/2611.03690
AB - Learned temporal models are moving beyond offline forecasting toward interactive environments for decision making, yet logged-trajectory accuracy need not predict closed-loop behavior under policies optimized in the learned model. This forecasting-control gap matters in tokamak magnetic control, where coupled plasma-circuit dynamics and actuator constraints can amplify errors through feedback. Using a calibrated free-boundary Grad-Shafranov solver, we collect four datasets in a 2x2 design over physics-parameter and current-reference coverage. At matched budgets of 500, 1k, 3k, and 5k rollouts, we train 16 action-conditioned Transformers; PPO policies optimized in the 5k surrogates are then evaluated without adaptation on six reference-solver conditions. Only joint-coverage D4 shows monotonically decreasing cross-condition mean error, whereas D1 improves in distribution but degrades out of distribution. Binary completion masks substantial closed-loop differences: D2-D4 each complete 83.3% of conditions, but reach on average 99.0%, 94.7%, and 90.5% of the evaluation horizon, while their worst conditions reach 94.2%, 68.2%, and 43.2%. Policies trained with action noise complete 5/24 matched conditions, compared with 19/24 under nominal training, and have a lower mean horizon-capped episode length. Based on one trained policy per configuration and one evaluation realization per condition, the results support combining prediction error, episode completion, condition-level duration, and tracking quality.
ER -