Neural Network Prediction Of Disruptions Caused By Locked Modes On J-TEXT Tokamak

In tokamaks, disruption is a dramatic event in which the plasma confinement issuddenly destroyed. There is the huge electromagnetic force with disruptions as well asthe release of the heat effect on the supporting structures of tokamak to damage. As thephysics of disruptions are not understood perfectly, to forecast disruption accurately andreduce the damage to devices by the external control is a burning desire for scienceresearcher.Locking is one of keys of plasma disruptions. Locked modes are closely related toplasma MHD instability and lead to disruptions. Since the spring2011experiment, alarge area of damage of Mirnov coils seriously affect the observation of the behavior ofplasma magnetic activation. Then in the summer of2011, we re-install the Mirnov coilsarrays of J-TEXT to repair damaged probe and improve structure. At the same time,radial magnetic field (Br) will rise when plasma mode is locking. In order to observe thisphenomenon, locked mode coils are constructed as new diagnostic signals to measureradial magnetic field on J-TEXT.This paper is focus on the prediction of disruptions caused by locked modes as theopening on disruption forecast work. Back-Propagation neural network is used as thecomputational tools. The network, which is based on Back-Propagation(BP) neuralnetwork, use Mirnov coils and locked mode coils signals as input data and output a signalincluding information of prediction of locked mode.On J-TEXT, The rate of successful prediction of disruption caused by locked modesis more than90%. For intrinsic locked mode disruptions, the network can give a warningsignal ahead of the locking-time about1ms. For the disruption caused by RMPs lockedmodes, the network can give a warning signal ahead of the locking-time about10ms.