Abstract: The particle flow discrete element (PFC^3D) is used to study the landslide movement process in the simulation experiment.Firstly, relying on UAV investigation to establish a high-precision DEM of the landslide and reproduce the original topography, to build a 3D discrete numerical model.Then, using a three-axis compression experiment with the BP neural network, the calibration and inversion of the macro and mesoscopic parameters of the rock and soil are realized, and the parameter assignment of the rock and soil is completed.Finally, the entire landslide movement and accumulation process are simulated, and the inversion and analysis of the landslide movement are completed.The results show that the establishment of discrete numerical models through UAV investigation and the combination of three-axis tests with machine learning can provide a complete modeling process for PFC^3D landslide simulation.PFC^3D can reflect the real the landslide movement process is suitable for the simulation of the movement characteristics of the landslide.The research results are basically consistent with the actual situation.This research method can provide effective solutions for the quantitative analysis of landslides.