A RL-based Vertical Stabilization System for the EAST tokamak

Abstract

The Vertical Stabilization (VS) is one of the most challenging control problems in tokamak fusion reactors. Existing solutions are usually tailored for the specific device. In this work, a model-free agent trained with the Q-learning algorithm is proposed as a possible alternative solution to the VS problem. The EAST tokamak is then considered as case-study to show the effectiveness of the proposed approach. The robustness of the VS agent is assessed by taking into account different plasma conditions and by including it into simulations of the overall EAST plasma control system.