Tuesday, March 20, 2018
Tokamaks are subject to sudden and unexpected events called disruptions which terminate the plasma current, causing a loss of plasma confinement. Since disruptions also produce large forces and strong loads that can damage the device, it is important to predict them in advance in order to avoid them or mitigate their damage. Although a thorough physical understanding of the mechanisms that drive a plasma into an unstable disruptive regime has not yet been reached, disruptions have been addressed using advanced statistical analysis and, recently, machine learning techniques. In this presentation, an exploratory machine learning study using a random forest algorithm for disruption prediction will be discussed, and a comparison of the algorithm’s performance on DIII-D and Alcator C-Mod will be analyzed.