Abstract: In order to obtain real-time information of the environment around tailings and make up for the inadequacy of existing air pollution dispersion simulation,Unreal Engine 5 was proposed to realize simulate air pollution diffusion. Long ShortTerm Memory-Improved Empirical Mode Decomposition-Long Short-Term Memory (LSTM-IEMD-LSTM) model was introduced to predict PM_2.5. Firstly,the scenario and corresponding functional logic were set in Unreal Engine 5. Secondly,the simulation of air pollution diffusion in the scene was completed by combining Gauss model and fluid simulation theory. Thirdly, the real-time interaction between the scene and simulation in the front end of the system was realized by pixel flow technology. The LSTM-IEMD-LSTM model was applied to further monitor the gas pollution of tailings impoundment. The research results show that the system can provide a more simple,better visual effect,more interactive and professional dynamic diffusion of gas pollution and a higher precision PM_2.5 prediction model,which provides certain technical support for monitoring the pollution around the tailings impoundment.