2025 – Use RL to harness and control ITBs in high-qmin steady-state scenario

Use RL to harness and control ITBs in high-qmin steady-state scenario

2025 Research Campaign, Steady State and Pulsed Fusion Core

Purpose of Experiment

This experiment aims to utilize the internal transport barrier (ITB) to enhance the performance of high-qmin scenario. Early heating is applied during the current ramp-up phase to raise the q profile and delay current penetration into the core region. However, in high-qmin scenarios, this early heating can sometimes trigger ITB formation during the ramp-up phase, leading to a sudden increase in betaN, followed by a collapse, even when nearly identical heating waveforms are used in otherwise stable discharges. To achieve high-qmin scenarios more reliably, a neural-network-based q profile and betaN control system, developed through reinforcement learning, will be employed to adapt to sudden changes in transport characteristics when ITB is triggered.

Experimental Approach

The controller was trained using integrated transport simulations designed to reproduce the experimental results of high-qmin scenario discharges from the last experimental campaign. To enhance adaptability to uncertainties in transport characteristics, randomized transport model parameters were used during training. The trained control system consists of two stages. The first-stage neural network (Analyzer NN) is trained to infer ITB strength from 1D profile data, while the second stage consists of multiple neural networks (Controller NNs), trained to control q profile and betaN using NB and EC as actuators. Each Controller NN is trained on simulations covering different ranges of ITB strength. To achieve stable q profile and betaN control, we will optimize the selection of Controller NNs. Additionally, we will evaluate the effectiveness of the two-stage control system for high-qmin plasmas with ITB by disabling the adaptive switching of Controller NNs. Once stable control is demonstrated, we will attempt to modify the betaN target to further enhance the performance of high-qmin scenario discharges.

Interested in a behind-the-scenes look at DIII-D? Join us for a virtual OR in-person tour during Fusion Energy Week (May 5-9)! Sign up for a tour here.

Got it!
X