Overview of Disruption Event Characterization and Forecasting (DECAF) Research

Physics-based disruption event characterization and forecasting (DECAF) research determines the relation of events leading to plasma disruption and aims to provide early warning for disruption avoidance. Offline analysis accesses data from several tokamaks (e.g. KSTAR, MAST/-U, NSTX/-U, ASDEX-U, DIII-D) to best understand, validate, and extrapolate models and to consider the key question of event and disruption correlation vs. causality. Fully automated analysis of large datasets is possible with initial results showing true positive rates over 99%. Real-time (r/t) DECAF has started on KSTAR. Experiments produced over 50 plasmas that are forecast with 100% accuracy in r/t, some triggering controlled plasma shutdown or disruption mitigation. Warnings were issued well before (>0.5s) the expected disruption. R/t magnetics, T_e profiles from electron cyclotron emission (ECE), 2D T_e fluctuation data from ECE imaging, and velocity profile acquisition are installed. An r/t MSE system has been built. Research supporting DECAF is shown including resistive stability evaluation at high non-inductive current fraction and innovative counterfactual machine learning application to MHD stability limits. *This research is supported by U.S. DOE grants DE-SC0020415, DE-SC0018623, and DE-SC0021311.