Machine Learning Surrogate Model for EP Transport in ITER Steady State Case

Efficient confinement of energetic particles (EPs) is crucial for achieving steady state burning plasma conditions in fusion pilot plants. FAR3d is a gyro fluid simulation code, which is used to analyze the energetic particle transport fluxes due to Alfvén eigenmode (AE) instability in a fusion device. However, these calculations are computationally intensive and time-consuming, necessitating the development of faster, but reasonably accurate surrogate models. We aim to leverage machine learning tools to develop a surrogate model for predicting EP transport fluxes, using the data generated by FAR3d simulations for a DIII-D experiment. We will generate a comprehensive dataset from FAR3d simulations of EP transport fluxes as a function of key input parameters such as fast ion density and safety factor (q) profiles, etc., for a DIII-D case and preprocess the data to train the surrogate model using Gaussian regression techniques. The model’s performance will be then validated against FAR3d simulations to determine its accuracy, uncertainties, and limitations, which will guide further improvements. This study will help us in precluding some challenges in building a surrogate model for calculations of EP fluxes in reactor-relevant devices such as ITER.