New Surrogate Model Enhances Understanding of Plasma Behavior

A new research paper titled "EuroPED-NN: Uncertainty aware surrogate model" has been published by A. Panera Alvarez and collaborators. The study introduces an uncertainty-aware surrogate model for the EuroPED plasma pedestal model, utilizing a Bayesian neural network with a noise contrastive prior (BNN-NCP) technique. This model was trained using data from the JET-ILW pedestal database and has undergone evaluations consistent with the EuroPED-NN framework.

The BNN-NCP technique not only matches the output of a conventional neural network but also provides confidence estimates for its predictions, highlighting areas of uncertainty. This capability is crucial for assessing the robustness and reliability of the model. The EuroPED-NN model has been validated through two analyses: one focusing on electron density in relation to increasing plasma current, and another examining the relationship between the pressure gradient and the EuroPED model.

Additionally, the researchers developed an experimental model that incorporates uncertainty, which is operational within the JET database. Both models have been tested across approximately 50 AUG shots, further confirming their effectiveness.

This work contributes to the field of plasma physics by enhancing the understanding of plasma behavior under various conditions, which could have implications for future research and applications in fusion energy. The full paper can be accessed at arXiv:2402.00760.