Abstract Real-time control of electron cyclotron heating (ECH) is critical for optimizing plasma performance and preventing instabilities in advanced fusion devices. We present a comprehensive real-time ECH control system developed on the Large Helical Device (LHD) that enables real-time adjustment of microwave heating in response to evolving plasma conditions. A field-programmable gate array (FPGA) controller actuates the ECH launchers, enabling millisecond-level adjustments of both injection angle and polarization to maximize absorption as plasma density and temperature profiles change. To compute suitable settings rapidly, a generative adversarial network (GAN) model was trained on thousands of past LHD discharges and ray-tracing simulations of ECH to generate control parameters for the deposition position and polarization. In high-density LHD experiments, another machine-learning-based prediction framework enabled the first active avoidance of radiative collapse. The predictor identified an impending collapse about 65 milliseconds in advance, triggering automated ECH power re-targeting and a cutoff of fueling that stabilized the plasma beyond the conventional density limit. These developments demonstrate how real-time ECH control, together with machine-learning-assisted prediction and inference, can sustain stable, high-performance plasmas, highlighting a pathway toward long-duration, steady-state fusion operations.
Kenmochi et al. (Mon,) studied this question.