For long-pulse tokamaks, one of the main challenges in control strategy is to simultaneously reach multiple control objectives and to robustly handle in real-time (RT) unexpected events (off-normal-events -- ONEs) with a limited set of actuators. We
have developed in our previous work a generic architecture of the plasma control system (PCS) including a supervisor and an actuator manager to deal with these issues. We present in this paper recent developments of real-time decision-making by the supervisor to switch between different control scenarios (normal, backup, shutdown, disruption mitigation, etc.) during the discharge, based on off-normal-event states. We first standardize the evaluation of ONEs and thereby simplify significantly the supervisor decision logic, as well as facilitate the modifications and extensions of ONE states in the future. The whole PCS has been implemented on the TCV tokamak, applied to disruption avoidance with density limit experiments, demonstrating the excellent capabilities of the new RT integrated strategy.
During a tokamak discharge, the plasma can vary between different confinement regimes: Low (L), High (H) and, in some cases, a temporary (intermediate state), called Dithering (D). In addition, while the plasma is in H mode, Edge Localized Modes (ELM
s) can occur. The automatic detection of changes between these states, and of ELMs, is important for tokamak operation. Motivated by this, and by recent developments in Deep Learning (DL), we developed and compared two methods for automatic detection of the occurrence of L-D-H transitions and ELMs, applied on data from the TCV tokamak. These methods consist in a Convolutional Neural Network (CNN) and a Convolutional Long Short Term Memory Neural Network (Conv-LSTM). We measured our results with regards to ELMs using ROC curves and Youdens score index, and regarding state detection using Cohens Kappa Index.
