Machine learning model to efficiently predict the structure and properties of MgCl2–NaCl–KCl melts

Abstract

MgCl2–NaCl–KCl ternary melts hold great promise as thermal energy storage and heat transfer fluid for the next generation of concentrating solar power (CSP), offering major advantages such as high-temperature storage capacity, adjustable temperature ranges, and renewable usability. However, a critical challenge is the lack of an accurate model to analyze the structural and property aspects of this complex melt. In response, a machine learning-based model with high accuracy and applicability has been developed for this purpose. Based on DPMD techniques, the trained model is used to investigate the structure information, transport and thermal properties of the melt. Temperature and composition effects are introduced. Radial distribution functions, coordination numbers, and angular distribution functions are applied to investigate the microscopic ionic environment within the melt. Transport properties studied include density, self-diffusion coefficient, ionic conductivity, and viscosity, while thermal properties include heat capacity at constant pressure and volume and thermal conductivity. Our model accurately reproduces the results obtained from FPMD simulations and accurately predicts the macroscopic properties of the melt through validation. The high stability of the Mg2+-Cl- interactions is introduced, which remains relatively unaffected by temperature and concentration changes. The presence of MgCl53− species in the melt was observed. We are devoted to the development of AI for the CSP science field. Due to the application of machine learning strategies, our models have proven to have exceptional accuracy and ease of study for MgCl2–NaCl–KCl ternary melts, and they can provide valuable insights into their properties and ionic behaviors.