A NEURAL NETWORK-BASED DISRUPTION ALARM FOR PLASMA CONTAINMENT

DISRUPTIONS (SUDDEN LOSSES OF PLASMA ENERGY TO THE SURROUNDING STRUCTURE) DEGRADE OPERATING CONDITIONS AND MUST BE AVOIDED IN FUTURE LARGE TOKAMAKS. IT IS PLANNED TO STUDY THE FEASIBILITY OF A DISRUPTION ALARM SYSTEM THAT WOULD ALLOW CONTROL MEASURES TO BE INVOKED BEFORE A DISRUPTION OCCURS. IN PHASE I, A DATA SET OF PARAMETER MEASUREMENTS FROM THE TOKAMAK FUSION TEST REACTOR (TFTR) WILL BE CREATED, AND THESE DATA WILL BE USED TO TRAIN A NEURAL NETWORK TO DISTINGUISH BETWEEN DISCHARGES THAT DISRUPT OPERATION NEAR THE HIGH-BETA LIMIT OF OPERATION AND THOSE THAT REMAIN STABLE. EXAMINATION OF THE NETWORK STIMULUS-RESPONSE RELATIONSHIP WILL REVEAL WHICH PARAMETERS PROVIDE A SIGNATURE FOR AN IMPENDING DISRUPTION. ADAPTIVE SIGNAL PROCESSING TECHNIQUES WILL BE USED TO EXTRAPOLATE PARAMETER EVOLUTION. IT IS PLANNED TO DETERMINE WHETHER HIGH-BETA DISRUPTIONS CAN BE PREDICTED FROM PARAMETRIC DATA, AND TO IDENTIFY THE PARAMETRIC RELATIONSHIPS THAT PROVIDE A SIGNATURE OF IMPENDING DISRUPTION. THE RESULTS WILL BE INCORPORATED INTO THE PHASE II PROJECT OF DEVELOPING A SYSTEM CAPABLE OF PROVIDING A DISRUPTION ALARM SUITABLE FOR TOKAMAK CONTROL APPLICATIONS.