For concentrated solar power systems, the chloride nanocomposite phase change material (PCM) is a promising candidate for heat storage. However, the complex components in composite PCMs have posed a great challenge to accurately predicting the thermal properties by molecular dynamics simulation. A deep potential molecular dynamics (DPMD) method was adopted to describe the interactions in chloride composite PCMs based on the machine learning method. The deep potential (DP) was constructed by training the data sets from ab initio molecular dynamics (AIMD) calculations and was then used to predict the microstructure and thermophysical properties of NaCl and NaCl−SiO 2 PCMs. The structural characteristics at different temperatures were determined by the radial distribution function, mean squared displacement, and diffusion coefficient. The thermophysical properties predicted by DP were well accordant with the experimental results. The largest relative errors of the density and the thermal conductivity were 10.7% and 8.9% for NaCl, respectively. The values were more accurate than those predicted by Lennard-Jones potential and Born−Mayer−Huggins−Tosi−Fumi (BMHTF) potential. Additionally, DPMD results showed that the thermal conductivity of NaCl−SiO 2 increased by up to 8.5% in comparison with that of pure NaCl. The work presents a novel molecular potential based on the machine learning method and narrows data gaps for the prediction of the thermal properties of composite chloride salts.