This paper focuses on the wide bandgap nitride semiconductors AlN, GaN, and their AlN/GaN superlattice structures. A machine learning force field with the accuracy of density functional theory is constructed and used to replace the time-consuming first-principles calculations. It efficiently and accurately simulates their thermal transport properties. The influence laws and microscopic mechanisms of strain control and layer structure on thermal conductivity are systematically studied. In the study, a high-quality dataset consisting of 2500 configurations was constructed based on multi-temperature ab initio molecular dynamics simulations. The force field was trained using a neural network architecture, and its reliability was verified through comparisons of root mean square error, phonon spectra, and thermal conductivity. The results showed that the developed force field achieved DFT accuracy in energy, atomic forces, and phonon spectrum predictions, and the computational efficiency was increased by 3-4 orders of magnitude. In terms of the strain effect, through simulations of the thermal conductivity under ±5% biaxial strain, it was found that compressive strain significantly increased the thermal conductivity (AlN increased by 11.3%, GaN increased by 20.5%, and superlattice increased by 15.3%), while tensile strain led to a decrease in thermal conductivity (decreasing by 21.1%, 50%, and 36.7% respectively). Microscopic mechanism analysis indicates that the strain alters the phonon transport behavior by regulating the phonon group velocity, phonon lifetime, and scattering rate: compressive strain enhances the group velocity, prolongs the lifetime, and suppresses scattering; tensile strain, on the contrary, enhances non-harmonicity and intensifies phonon scattering. It is particularly important that the thermal conductivity of the AlN/GaN superlattice is much lower than that of its constituent materials (71.93 W/mK at room temperature, approximately 26% of AlN and 34% of GaN). This is mainly attributed to the enhanced intrinsic scattering caused by the hetero-interface. Further studies have shown that in superlattices with a fixed total number of layers, the thermal conductivity significantly increases with the thickening of the AlN barrier layer. Based on this, the "thick barrier, narrow quantum well" structural design principle is proposed, providing a theoretical basis for regulating the thermal transport properties through superlattice engineering. This study has demonstrated the powerful capabilities of machine learning force fields in the research of thermal transport in complex heterogeneous interface systems, revealing the synergistic regulation mechanism of strain and structure on thermal conductivity, and providing key theoretical guidance and design paths for the thermal management optimization of high-power nitride devices.