The application of classification methods in a data driven investigation of the JET L–H transition

Abstract

Understanding the processes which establish the H-mode edge transport barrier (ETB) and the scaling of those processes with the plasma properties local to the plasma edge is of critical importance for optimizing the performance of power-station scale fusion plasmas. In this paper, data from 67 JET pulses were assembled and classified by confinement mode. A neural network classification technique was applied to identify the nature of the dependence of the L–H boundary on plasma parameters local to the plasma edge. Strong dependences on Te, ne, ⟨B⟩ and a weak dependence on q80 were found. In applying the neural network model as a confinement mode identification tool, the correct operating mode was identified for 98.86% of the time slices in a test data set. Using an extended data set the boundary predictions from five theoretical ETB models were evaluated. These models consider a range of different physical processes for the ETB. One model was clearly rejected. While other models proved more favourable, none of the tested models robustly described the JET L–H boundary.