Neural Network Approach for Estimation and Prediction of Time to Disruption in Tokamak Reactors

This paper deals with the problem of predicting the onset of a disruption on the basis of some known precursors possibly announcing the event. The availability in real time of a large set of diagnostic signals allows us to collectively interpret the data in order to decide whether we are near a disruption or during a normal operation scenario. In this work, a database of disruptive discharges in Joint European Torus (JET) have been analyzed for the purpose. Neural Networks have been investigated as suitable tools to cope with the prediction problem. The experimental database has been exploited aiming to gain information about the mechanisms which drive to a disruption. The proposed processor will operate by implementing a classification of the shot type, and outputting a real number that indicates the time left before the disruption will take place.