Neural Network Computed Bootstrap Current for Real Time Control in DIII-D

In an effort to provide a fast and accurate calculation of the bootstrap current density for use as a constraint in real-time equilibrium reconstructions, we have developed a neural network (NN) non-linear regression of the NEO code calculated bootstrap current $j_{BS}$. A new formulation for $j_{BS}$ in NEO allows for a determination of the coefficients on the density and temperature scale lengths. The new formulation reduces the number of inputs to the NN, and the number of output coefficients is 2 times the number of species (including electrons). The NN can reproduce the NEO and Sauter coefficients to a high degree of accuracy ($<1\%$ error). The toroidal (not parallel) component of the bootstrap current density calculated in NEO has been used as a constraint in an offline equilibrium reconstruction for comparison to the NN calculation. The computational time of this method ($\mu$s) makes it ideal for real time calculation in DIII-D.