Neural operator reconstructs unseen electron density in tokamaks fast and with uncertainty
Reconstruction of Unobserved Core Electron Density in Tokamak Plasmas with a Neural Operator Transport Surrogate and Uncertainty Quantification
Machine Learning
Summary
Measuring the full electron density inside fusion devices called tokamaks is difficult because sensors can only see parts of it. The authors show a way to fill in the invisible parts by using a physics-based equation combined with a fast neural network model that predicts how electrons move. This method quickly estimates the full electron density and includes an assessment of how confident those estimates are. It performs better than standard curve-fitting methods on both simulated and real fusion data.
What this means in practice
- •For tokamak control engineers: Generate full electron density profiles in real time from partial sensor data to improve plasma control in fusion reactors.
- •For plasma diagnostic developers: Combine limited measurements with physics-based models and neural surrogates to reconstruct unmeasurable plasma properties with quantified uncertainty.
Authors
D. Sorokin, A. Granovskiy, M. Stokolesov, E. Adishchev, M. Nurgaliev, G. Subbotin
Abstract
Real-time control of tokamak plasmas requires the electron density profile. However, in a reactor-relevant device, diagnostic port access is limited and the profile can be observed only partially. In the present work we reconstruct the density profile by inferring the coefficients of a physics-based one-dimensional particle transport equation, using a Fourier Neural Operator (FNO) surrogate of the transport solver as a fast differentiable forward model. The surrogate enables realtime-compatible gradient-based reconstruction and uncertainty quantification from the reflectometry and interferometry measurements. We compare four uncertainty quantification methods - multi-seed ensembles, the Laplace approximation, Metropolis-Hastings, and a Gaussian method independent of the reconstruction fit. The resulting uncertainty is scored with CRPS and coverage calibration, separately on the observed and unobserved regions. We further demonstrate advantage of our approach over standard analytical function fit on synthetic and experimental DIII-D discharges.