Go to ICLR 2025 Conference homepageData-driven plasma equilibrium forecasting in magnetic fusion tokamak
Keywords: nuclear fusion, plasma equilibrium, tokamak, video prediction
TL;DR: This work addresses the plasma equilibrium forecasting problem as a video prediction task.
Abstract: The most promising approach to achieving nuclear fusion is through tokamaks, which confine plasma using magnetic fields.
Understanding the current plasma equilibrium state in tokamaks is critical for effective plasma control.
Unlike previous studies, which reconstruct equilibrium from magnetic field information, our work forecasts future equilibrium based on past equilibrium states.
Specifically, we formulate the plasma equilibrium prediction task as a video prediction task, a well-explored problem in the machine learning community.
This formulation allows us to capture the spatio-temporal dynamics of plasma states and provides a foundation for multimodal modeling of data streams from tokamak operations.
Our methodology, incorporating a physics-inspired learning technique for physically reliable predictions, achieved plausible results in forecasting future plasma equilibrium up to 200 ms ahead compared to baselines.
This approach holds promise for predicting plasma instabilities and preventing disruptions, marking a significant step towards developing stable fusion reactors.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3629
Loading