Prediction of Turbulence Temporal Evolution in PANTA by Long-Short Term Memory Network

Abstract

Prediction of time evolution of multi-scale turbulence is performed by using Long-short term memory networks. The time series data is obtained by Langmuir probes in a linear magnetized plasma device, PANTA. The simultaneous prediction of high and low frequency components of turbulence is shown to be possible within several tens percent accuracy. The prediction accuracy depends on the initial network, which can be controlled by reducing the learning rate.