An artificial intelligence has been developed that can simulate the plasma state inside a nuclear fusion reactor 1,000 times faster than before.

UNIST Department of Nuclear Engineering Professor Lee Ji-min and Professor Yoon Eui-seong's team announced on the 17th that they have developed 'FPL-net', a deep learning-based artificial intelligence model that can accelerate the solution of mathematical equations describing plasma states.

In nuclear fusion power generation, also known as artificial sun technology, the inside of the generator must be maintained in a high-temperature plasma state like the actual sun. Plasma is a state in which matter is separated into negatively charged electrons and positively charged ion particles, and accurately predicting collisions between particles in this state is a key factor in maintaining a stable nuclear fusion reaction.

The plasma state is represented by a mathematical model, one of which is the Fokker-Planck-Landau equation (FPL). The Fokker-Planck-Landau equation predicts collisions between positive and negative charged particles, that is, Coulomb collisions. Originally, to solve this equation, an iterative method was used to gradually find a solution, which required a lot of calculations and took a long time.

The FPL-net developed by the research team can solve equations in one go, unlike the existing iterative method. It can solve problems 1,000 times faster than before, and has shown high accuracy with a prediction error of 10⁻⁵.

The Fokker-Planck-Landau collision process has the characteristics of conserving density, momentum, and energy, and the explanation is that the accuracy was improved by defining a function so that these physical quantities are conserved during the artificial intelligence model learning process.

The accuracy of the AI model was verified by thermal equilibrium simulation. If errors accumulate in the continuous simulation process, accurate thermal equilibrium cannot be obtained.

The joint research team said, “While maintaining accuracy, we have shortened the calculation time by 1,000 times compared to the existing code using CPU by utilizing deep learning using GPU,” and “It will become the cornerstone of turbulence analysis code that simulates the entire area of a nuclear fusion reactor and digital twin technology that implements a real tokamak in the virtual space of a computer.” A tokamak is a special structure that confines plasma.

The research team added, “However, this study is limited to electron plasma, and for application, research is needed to expand it to a complex plasma environment of various particles containing impurities.”

This research was conducted with the support of Ulsan National Institute of Science and Technology (UNIST), the National Research Foundation of Korea, and the Korea Institute of Energy Technology Evaluation and Planning, and was published on February 15 in the Journal of Computational Physics, an international academic journal.