Molecular dynamics study on structure and thermal properties of high-performance chloride molten salt
The rapid development of renewable energy technologies and improvements in deep load adjustment and frequency regulation in thermal power systems have placed higher demands on the operating temperature range and thermophysical properties of molten salt storage materials.MgCl2-NaCl-KCl(MgNaK)molten salt stands out among the potential candidates.However,complete thermophysical property data for MgNaK molten salt is lacking.In this study,we developed a machine learning potential function to accurately describe the microscopic particle interactions in MgNaK molten salt(with a composition of 45.4％mol MgCl2,33％mol NaCl,and 21.6％mol KCl).This model is based on energy and atomic force information obtained from first-principles molecular dynamics(FPMD)simulations.The reliability of this machine learning potential function was validated by comparing the partial radial distribution function(PRDF)and coordination number(CN)with FPMD results,showing excellent agreement.We explored how the local structure and thermal properties of MgNaK vary with temperature from atomic and electronic perspectives.The introduction of Na+or K+ions disrupts the tightly connected MgClx network structure,thereby affecting transportation properties.The density(ρ)and constant-pressure specific heat capacity(Cρ)obtained from machine learning potential function simulations were found to closely match the experimental data,with deviations of less than 2％.Furthermore,we examined the thermal conductivity(λ)of MgNaK molten salt using kinetic theory,we examined a negative linear correlation with temperature,which aligns with observations in other chloride molten salts.Based on molecular simulations and experimental measurements,we provide recommended values for λ and viscosity(η)of MgNaK molten salt across the entire operating temperature range.
high performance chloride molten saltenergy storage technologymicrostructures and thermophysical propertiesmachine learning potential
万方数据
维普
