Differential Rotation Control for the DIII-D Tokamak via Model-Based Reinforcement Learning*

Differential Rotation Control for the DIII-D Tokamak via Model-Based Reinforcement Learning

Reinforcement learning has recently proven itself as a technique that is able to learn sophisticated controllers. Not only has it been leveraged to achieve super-human performance on games such as Go, but it also has been used for shape control on TCV. In this work, we wish to use these techniques to develop a controller that uses the beams to maintain a β_N value while achieving specified differences in rotation between the q=1 and q=2 boundary. Prior works suggest that achieving a high difference between these boundaries can prevent NTMs, and development of such a controller would aid in testing this claim. Although developing this controller is important in its own right for investigating how rotation shear affects NTM seeding, a successful implementation of this controller will serve as an example of how arbitrary feedback controllers can be learned solely through data.

Acknowledgements

This work was supported by DE- SC0021275 (Machine Learning for Real-time Fusion Plasma Behavior Prediction and Manipulation) and DE-FC02-04ER54698.

This work is supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE1745016 and DGE2140739. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.