Estimation of 2D profile dynamics of electrostatic potential fluctuations using multi-scale deep learning

Dynamics in magnetically confined plasmas are dominated by turbulence driven by spatial inhomogeneities in density and temperature. Simultaneous measurement of velocity field and density fluctuations is necessary to observe the particle transport, but the measurement of the velocity field fluctuations is often challenging. Here, we propose a method to estimation velocity field fluctuations from density fluctuations by using plasma turbulence simulations and a deep technique learning. In order to take multi-scale characteristics into account, the several number of spatial filters are used in the convolutional neural network. The velocity field fluctuations are successfully predicted, and the particle transport estimated from the predicted velocity field fluctuations is within 93.1% accuracy. The deep learning could be used for the prediction of physical variables which are difficult to be measured.