Data-Based Prediction and Analysis of the Plasma Pedestal in Tokamak Fusion Experiments

Abstract

In tokamak fusion reactors, the pedestal is a steep pressure drop at the plasma edge in high confinement mode (H-mode). It is an important factor in both generating fusion power and machine wall deterioration, and understanding and predicting pedestal behavior can enable future fusion performance optimization. In this work, we present data-driven analysis on and prediction for the pedestal layer from experimental data. We introduce three different novel datasets, a neural network model (mNN) for calculating 5 key properties of the pedestal from basic machine parameter inputs, and an algorithm for data-cleaning based on the K-NN model.