High-Performance Prediction of Thermodynamic Property of NaNO3–NaCl–NaF Molten Salt by Deep Potential Molecular Dynamics Simulation for High-Temperature Thermal Application

Abstract Nitrate molten salts are widely applied in concentrating solar power due to their excellent thermodynamic properties. However, conventional experimental methods struggle to accurately measure the thermodynamic properties of molten salts and thoroughly analyze their microscopic structures. The deep potential molecular dynamics (DPMD) method provides a novel approach for investigating the structures and thermodynamic characteristics of molten salts. By integrating machine learning (ML) with the deep potential GENerator (DP-GEN) active learning algorithm, we developed a high-fidelity interatomic potential function of NaNO3–NaCl–NaF molten salt. The structure and thermodynamic properties of the molten salt were systematically explored, and the influence of temperature on these characteristics was elucidated. The results show that the elevated temperature changes the interionic distances and the interionic forces. The increase in temperature leads to a decrease in the coordination number and a reduction in the energy barrier, making it easier for ions to detach from the central ion, thereby enhancing the ion migration ability in molten salts. The simulated density and specific heat capacity show excellent consistency with the experimental results, with relative errors of 2.11% and 0.45%, respectively. When the temperature rises from 573 to 773 K, the simulated viscosity reduces from 2.803 to 1.526 mPa·s, while thermal conductivity declines from 0.576 to 0.504 W/(m·K). Combining deep potential molecular dynamics with the DP-GEN active learning algorithm, this study provides an efficient approach to predict the thermodynamic properties of molten salt thermal energy storage materials.