Based on random weight particle swarm optimization and least square support vector machine, the displacement inverse analysis model of mechanical parameters of underground cavern was established. The learning samples were determined by orthogonal test and 3DEC model. Using this model, the inverse analysis was carried out. The results show that the inverse analysis of mechanical parameters can be realized by using the nonlinear mapping of support vector machine method. The two lateral pressure coefficients are 0.453 and 0.644, and the elastic modulus of three kinds of rock masses are 10.15, 8.20 and 2.62 GPa. The deviation between the displacement obtained from inverse analysis and on-site measurement is small, which verifies the rationality of the inversion parameters.