Cycle-Consistent and Uncertainty-Aware Neural Surrogates for Tokamak Edge Plasmas

2026-07-23 · Source: Artificial Intelligence · Field: Science & Research — Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences, Physical Sciences & Chemistry · Depth: Expert, quick

Summary

A new cycle-consistent and uncertainty-aware neural surrogate addresses the computational bottleneck of tokamak edge plasma simulations, which are critical for predicting heat fluxes and detachment onset but are too slow for real-time applications. This model combines a conditional U-Net forward model, mapping five control parameters to two-dimensional plasma-state fields on the SOLPS-ITER mesh, with an optimization-based inverse method. The inverse method employs cycle-consistency for self-supervised quality checks, eliminating the need for ground-truth labels during inference. An ensemble of multilayer perceptrons also provides electron temperature and density profiles with uncertainty estimates. The forward model achieves normalized root-mean-square errors below 2.6% and Pearson correlations above 0.95. Cycle-consistency regularization boosts the average cyclical R^2 from 0.59 to 0.99, enabling recovery of all five control parameters with Pearson r ≥ 0.97. A k-d tree warm start ensures over 95% database completion. With approximately 4×10^6 parameters, the model generates full 2D predictions in milliseconds, 5-6 orders of magnitude faster than SOLPS-ITER, supporting real-time control and digital twin applications.

Key takeaway

For research scientists developing digital twins or real-time control systems for complex physical phenomena like tokamak plasmas, this cycle-consistent neural surrogate offers a critical acceleration. You can achieve 5-6 orders of magnitude faster predictions than traditional simulations, enabling extensive parameter scans and robust uncertainty analysis. Consider integrating such self-supervised, uncertainty-aware models to enhance both forward and inverse prediction capabilities in your high-fidelity simulation workflows.

Key insights

Cycle-consistent neural surrogates with uncertainty awareness significantly accelerate tokamak edge plasma simulations.

Principles

• Cycle-consistency improves inverse model accuracy without degrading forward.
• Uncertainty estimates guide where more simulation data is needed.
• Warm-starting significantly boosts inverse method success rates.

Method

Combines a conditional U-Net forward model with an optimization-based inverse method built on the frozen forward network, enforcing consistency. An MLP ensemble adds uncertainty estimates for profiles.

In practice

• Enable real-time control of tokamak edge plasmas.
• Facilitate rapid parameter scans for optimization.
• Develop digital twins for fusion devices.

Topics

• Neural Surrogates
• Tokamak Plasmas
• Cycle-Consistency
• Uncertainty Quantification
• SOLPS-ITER
• Computational Physics

Best for: AI Scientist, Research Scientist

Related on AIssential

• Towards Data-Efficient Cross-Device Generalization of Grad-Shafranov Equilibria via Transfer Learning Neural Operator — Neural operator pretraining enables data-efficient, real-time, cross-device generalization of Grad-Shafranov equilibria for magnetic confinement fusion.
• Neural network surrogates with uncertainty quantification for inverse problems in partial differential equations — DeepGaLA uses uncertainty-aware neural network surrogates for scalable, reliable Bayesian inference in inverse problems, outperforming traditional methods in efficiency.
• Accelerate Clean, Modular, Nuclear Reactor Design with AI Physics — AI-augmented simulations with surrogate models significantly accelerate nuclear reactor design by overcoming computational bottlenecks.
• Offline Reinforcement Learning for Plasma Control in Nuclear Fusion: Codebase and Benchmark — RL4F provides a standardized offline RL benchmark for multi-actuator, long-horizon nuclear fusion plasma control.
• Accounting for Hysteresis and Eddy Currents in Finite Element Simulations of Ferromagnetic Laminated Cores using a Recurrent Neural Network — A recurrent neural network can efficiently surrogate complex hysteresis and eddy current effects in finite element simulations.
• TokaMark: A Comprehensive Benchmark for MAST Tokamak Plasma Models — TokaMark provides a unified, open benchmark with 14 tasks and a baseline for AI models in fusion plasma modeling.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.