In order to improve the prediction accuracy of atmospheric diffusion of radionuclides, the application of traditional simulation model and artificial intelligence technology in nuclear safety and environmental protection is discussed. This paper examines the effectiveness of traditional models such as Gaussian, Eulerian and Lagrangian models in different scenarios, and the role of artificial intelligence (especially neural networks) in the rapid and accurate retrieval of nuclear accident source term information. The results show that the study provides key support for nuclear accident emergency management and environmental monitoring, and is of great significance for improving nuclear safety and environmental protection strategies.