Graphene, as a high thermal conductivity enhancing filler, can significantly improve the thermal conductivity of phase change materials. The microscopic nature of its enhancement mechanism and cross scale prediction models can reveal the heat transfer laws of phase change. This article uses molecular dynamics simulation to regulate the doping content of graphene in alkane based composite phase change materials, and proposes and constructs a cross scale thermal conductivity prediction model with box constraint factor. The experiment shows that the prediction error of the model is within 5%. The simulation results show that there is a strong C-H···π specific interaction between the graphene interface and alkane molecules. The characteristic peak of the radial distribution function at 1.77 Å and the nonlinear change in mean square displacement indicate that the interface structure not only does not suppress the diffusion of liquid-phase molecules, but also fundamentally reduces the interface thermal resistance by constructing efficient phonon transmission channels. Based on the above simulation results, an innovative SCA-TCN deep learning prediction model was developed, with a determination coefficient R 2 of 0.99762, and the main error indicators were reduced by about 76% to 97% compared to the benchmark model.