Abstract Details
Abstracts
Author: Yashika Ghai
Requested Type: Poster
Submitted: 2026-03-22 22:14:47
Co-authors: Donald A. Spong, Arpan Biswas, J. Varela, Luis Garcia
Contact Info:
Oak Ridge National Laboratory
1 Bethel Valley Road
Oak Ridge, Tennessee 37830
USA
Abstract Text:
We have developed machine learning–based surrogate models for energetic particle (EP) transport in ITER that are significantly faster and reasonably accurate compared to existing high-fidelity nonlinear simulation codes. Such reduced-order models are essential for optimizing ITER plasma scenarios, particularly for steady-state burning plasma operation, where confinement of fusion-born alpha particles is critical for sustained self-heating. Energetic alpha particles produced in D–T fusion reactions can resonantly interact with and destabilize Alfvén eigenmodes (AEs), leading to enhanced EP transport and potential alpha losses. These losses reduce plasma self-heating efficiency and may increase heat loads on plasma-facing components. Accurately quantifying the impact of EP-driven transport on burning plasma performance remains a key challenge for reactor-scale devices such as ITER. Our approach begins with the generation of a comprehensive nonlinear simulation database using the FAR3d code for two ITER steady-state scenarios [1]. These simulations capture EP-driven transport under varying equilibrium and profile conditions. The resulting dataset is used to train surrogate models of alpha particle flux using neural networks (NNs) and Gaussian Process (GP) regression. We assess and compare these models in terms of prediction accuracy, generalization capability, and associated confidence bounds. The development of this surrogate framework provides a computationally efficient tool for EP transport prediction in ITER and establishes a pathway toward constructing more general surrogate models applicable to reactor-level fusion devices beyond ITER. Such models can be incorporated into integrated modeling platforms, including the IPS-FASTRAN framework, to enable self-consistent evaluations of burn performance in next-generation burning plasma scenarios.
Ref: [1] D.A. Spong, Y. Ghai, J. Varela, L. Garcia et al, Nucl. Fusion, 65, 112004 (2025).
Characterization: 6.0
Comments: