Abstract:

Maintaining stability of magnetically confined plasma is a central obstacle to practical nuclear fusion. Modeled as kinetic Vlasov–Poisson equations, the control problem is notably challenging due to non-linearity, sensitivity to initial conditions, and partial observability. Recent development of advanced AI technologies shows promise in control of plasma systems, while theoretical understanding and principled methodologies are still under-explored.

In this talk, we discuss recent advances in machine learning for plasma control. Starting from an expert controller constructed from a fully observed model, we develop algorithms that learn a feedback policy that operates only on experimentally available measurements. We prove that the learned policy stabilizes the plasma dynamics over long time horizons, and provide non-asymptotic sample complexity guarantees for the learning algorithm. The theories demonstrate the advantage of learning-based control in terms of adaptivity to unknown initial conditions and long-term stability. Empirical results on simulated plasma systems also validate the efficacy of our methods in stabilizing plasma over long time horizons.

This work builds a bridge between statistical learning theory and control of complex physical systems, and represents a step toward theoretically grounded, AI-assisted control strategies for fusion energy. Joint work with Xiaofan Xia and Qin Li.