Development of an efficient real-time disruption predictor from scratch on JET and implications for ITER

Abstract

Prediction of disruptions from scratch is an ITER-relevant topic. The first operations with the new ITER-like wall constitute a good opportunity to test the development of new predictors from scratch and the related methodologies. These methodologies have been based on the Advanced Predictor Of DISruptions (APODIS) architecture. APODIS is a real-time disruption predictor that is in operation in the JET real-time network. Balanced and unbalanced datasets are used to develop real-time predictors from scratch. The discharges are used in chronological order. Also, different criteria to decide when to retrain a predictor are discussed. The best results are obtained by applying a hybrid method (balanced/unbalanced datasets) for training and with the criterion of retraining after every missed alarm. The predictors are tested off-line with all the discharges (disruptive/non-disruptive) corresponding to the first three JET ITER-like wall campaigns. The results give a success rate of 93.8% and a false alarm rate of 2.8%. It should be considered that these results are obtained from models trained with no more than 42 disruptive discharges.

Runaway electron mitigation by 3D fields in the ASDEX-Upgrade experiment

“…In the experiments reported here a scan of ∆φ in steps of 45 • is performed and the perturbations are characterized by a dominant n = 1 toroidal mode number; a toroidal field of 2.5T limits the Clearly, such an approach requires disruption prediction methods currently in development and investigation [42][43][44][45].…”

Advanced Disruption Predictor Based On The Locked Mode Signal: Application To Jet

“…Trying to avoid the use of huge amount of discharges in the training process, a recent alternative has been the development of disruption predictors from scratch [2,3]. These are high learning rate predictors whose learning process starts with the first disruption.…”

“…It should be noted that PBOD has to perform better than the standard predictor based only on triggering an alarm when the locked mode amplitude crosses a threshold. Following the experience for feature selection in [1,2,3], PBOD also uses time windows to extract relevant information for the predictor. The locked mode signal is sampled at 1 kSamples/s (same sampling rate than APODIS) and the basic time window is 32 ms long (again the same temporal length used in the APODIS windows).…”

Upgrading Model Selection Criteria with Goodness of Fit Tests for Practical Applications

“…This approach seems quite promising and worth further investigations. The proposed criteria could also be further improved by implementing more advanced metrics, such as the geometric distance and the Venn definition of probability [29][30][31]. In this perspective, a combination of the proposed upgrades with recently developed extensions of the AIC criterion, using a class of pseudo distances instead of the Kullback-Leiber divergence, should also be seriously pursued [32].…”