Overcoming challenges: leveraging machine learning for efficient modeling of divertor plasmas

Modeling divertor plasma is challenging and time-consuming due to the inherent multi-scale, multi-physics nature of the problem. Consequently, application of high-fidelity divertor plasma model is often limited. The emerging machine learning technique offers an alternative solution to this challenge. A fast and fairly accurate data-driven surrogate model for complex physics is possible by leveraging the latent feature space concept as the intermediate step. This idea was first tested in the inertial fusion research [1], while the consequent study with simplified 1D flux-tube setup further demonstrated that complicated divertor plasma state has a low-dimensional representation in latent space [2]. Following the same methodology, application-specific surrogate models for divertor plasma (e.g., initial solution prediction for code acceleration, integrated tokamak divertor design, and divertor plasma detachment control) are constructed, trained, and tested based on more realistic 2D axisymmetric transport simulations. These models appear to be able to fulfill the designated tasks, indicating that machine learning could be a powerful tool for divertor plasma physics and fusion energy research.