2025 Volume 20 Article ID: 1203035
We demonstrate the estimation of electrostatic potential fluctuations in dynamically varying Kelvin-Helmholtz turbulence using multi-scale convolutional neural network. The turbulence field is obtained from simulations based on a reduced fluid model in cylindrical magnetized plasmas. The target turbulence shows limit-cycle oscillations, and coherent and spiral structures are generated and annihilated repeatedly. High accuracy of the prediction is realized for the electrostatic potential field, and the estimation of the particle flux calculated from the predicted potential agrees with the answer with 98.4% accuracy. Behavior of the prediction accuracy is also discussed by changing the hyper parameters, such as the number of filters and the size of the training data.
