Disruption prediction for future tokamaks using parameter-based transfer learning

Abstract Tokamaks are the most promising way for nuclear fusion reactors. Disruption in tokamaks is a violent event that terminates confined plasma and causes unacceptable damage to device. Machine learning models have been widely used predict incoming disruptions. However, future reactors, with much higher stored energy, cannot provide enough unmitigated disruption data at high performance train predictor before damaging themselves. Here we apply deep parameter-based transfer method prediction. We model on J-TEXT tokamak it, only 20 discharges, EAST, which has large difference size, operation regime, configuration respect J-TEXT. Results demonstrate reaches similar trained directly EAST using about 1900 discharge. Our results suggest proposed can tackle challenge predicting disruptions like ITER knowledge learned from existing tokamaks.