Speaker
Description
The real-time distribution of density profiles serves as valuable data for monitoring plasma density and position. Presently, devices rely on magnetic measurement diagnostics to obtain reference data for these purposes. The microwave reflectometer serves as a widely employed density profile diagnostic system on magnetic confinement fusion devices across various nations. It is also planned to be utilized on the future ITER facility for measuring plasma density distribution, aiming to determine the plasma boundary position and density.
Traditional physics-based profile inversion algorithms require manual extraction of time-delayed data and suffer from slow computation speeds, rendering them incapable of providing real-time density profiles during experimental discharges for plasma position and density feedback control. To achieve real-time density profile data acquisition and enable online data processing, there is an urgent need to reduce the computation time for profile inversion. This work presents a rapid density profile inversion algorithm based on a deep neural network model. Leveraging a large volume of density profile data for training and inference, this approach circumvents the intricate physical processes involved in microwave propagation within magnetized plasmas, as encountered in traditional physical algorithms. The described method has been successfully applied on the EAST device, achieving rapid and accurate density profile inversion.
The next step involves integrating this data-driven real-time density profile algorithm directly into the deployed profile reflectometer data acquisition system on EAST to enable parallel computation of data acquisition and processing. This integration aims to provide real-time density profile distribution, particularly during density feedback control experiments, such as gas or pellet injection experiments.
