Quasilinear turbulent particle and heat transport modelling with a neural-network-based approach founded on gyrokinetic calculations and experimental data

Abstract

A novel quasilinear turbulent transport model DeKANIS has been constructed founded on the gyrokinetic analysis of JT-60U plasmas. DeKANIS predicts particle and heat fluxes fast with a neural network (NN) based approach and distinguishes diffusive and non-diffusive transport processes. The original model only considered particle transport, but its capability has been extended to cover multi-channel turbulent transport. To solve a set of particle and heat transport equations stably in integrated codes with DeKANIS, the NN model embedded in DeKANIS has been modified. DeKANIS originally determined turbulent saturation levels semi-empirically based on JT-60U experimental data, but now it can also estimate them using a theory-based saturation rule. The new saturation model is still partly connected to experimental data, but it offers the potential for applying DeKANIS independently of the device.