The difference between soil parameter assignment and actual engineering is one of the main reasons that numerical calculation results are inconsistent with actual project. Aiming at the shortcomings of BP neural network in parameter inversion, on the basis of the measured surface settlement data of Taiyuan metro, this paper constructed PSO-BP neural network by using MATLAB neural network toolbox combined with orthogonal method and numerical simulation technology, and inverted soil parameters and used the inversion-corrected soil parameters to predict subsequent working conditions. The analysis results show that PSO-BP has higher calculation accuracy. The neural network inversion structure well reflects the non-linear relationship between surface subsidence and soil parameters, and can be used as a reference for the selection of soil parameters of similar projects in the same area.