Overview of SPARC disruption prediction and avoidance research

SPARC disruption prevention strategies will leverage first-principle and machine learning-based solutions to ensure that SPARC will accomplish its mission goals. This poster will provide an overview of research advances addressing disruption prediction and avoidance on SPARC. An off-normal warning (ONW) system for asynchronous control actions is under development, with a particular focus on first campaign operations. The ONW software will include physics-based thresholds beyond machine learning-driven ones, with particular focus on radiative limits and vertical instabilities. The most extreme off-normal event being a plasma disruption, the poster will present progress made in developing potential candidate algorithms for the Disruption Mitigation System trigger based on deep learning and pre-trained architectures. Typically, these algorithms provide a classification probability for the current plasma state. However, time-to-event predictions allow the control system to better prioritize actuators' responses and tailor them to the plasma trajectory and expected margin to stability. Progress will be discussed in this area leveraging deep survival analysis (C. Nagpal et al 2021 IEEE JBHI) and a more classical statistical definition of disruptivity (in units of 1/time) derived from available databases.