Learning plasma equilibria from scratch with deep neural network Grad-Shafranov solver

Reconstructing plasma equilibria has played an important role in tokamak performance. While the reconstruction can be routinely performed by solving Grad-Shafranov (GS) equation, it is mostly reconstructed based on magnetic measurements and still challenging to produce kinetic equilibria in real time due to unavailability of real time kinetic profiles as well as demands on long computation time. We deal with these issues by developing a deep neural network capable of directly solving GS equation based on unsupervised learning. We demonstrate that the proposed network does not require any precalculated database such as EFIT data and is possible to reconstruct kinetic equilibrium in real time. The proposed method can be a promising approach for ITER and beyond to achieve high quality operations and enhance plasma performance.