Development of a semi-empirical particle and heat transport model and improvement in its turbulent saturation rule

The gyrokinetic-based turbulent transport models are essential to predict density and temperature profiles, but introducing detailed descriptions of turbulence physics tends to increase the computational cost. To accelerate the profile predictions, a neural-network (NN) based approach has been undertaken. Our study is also developing a NN-based turbulent transport model DeKANIS. A turbulent saturation rule employed in DeKANIS was based on experimental particle fluxes estimated for JT-60U H-mode plasmas, and it was apt to overestimate temperatures. To reduce the overestimation, a different saturation rule is built including experimental heat fluxes.