Full-Discharge Tokamak Scenario Planning on DIII-D by Neural-Network-Enabled Actuator Trajectory and Plasma-Shape Optimization

K. Shabbir, B. Leard, T. Rafiq, E. Schuster

68th Division of Plasma Physics (DPP) Annual Meeting of the American Physical Society (APS)

Chicago, IL, USA, November 2-6, 2026

Scenario optimization is essential for designing actuator trajectories and plasma-shape evolution that achieve desired plasma performance. In this work, feedforward scenario optimization is performed within the COTSIM (Control Oriented Transport SIMulator) integrated modeling framework using Sequential Quadratic Programming (SQP) with finite-difference gradients. Because gradient-based optimization requires numerous predictive simulations, a suite of fast neural-network surrogate models is integrated to accelerate the workflow, including MMMnet for transport, NUBEAMnet for neutral beam heating and current drive, TORAYnet for electron cyclotron heating and current drive, and PEDESTALnet for pedestal prediction. The framework determines optimized actuator trajectories and plasma-shape evolution that drive the plasma toward prescribed scalar and profile targets throughout the ramp-up, flat-top, and ramp-down phases of the discharge. These neural-network surrogates reduce optimization time from hours to minutes while maintaining high predictive accuracy, enabling offline scenario planning and rapid iterative design. Planned applications include high-qmin internal transport barrier discharges and extended high-βN, high-βp scenarios with high noninductive current fraction on the DIII-D tokamak.

*Supported by the U.S. DOE under Awards DE-SC0010661 and DE-FC02-04ER54698.