Aiming at the defect of premature convergence of traditional experiential learning algorithm (EBL), Yin-Yang thought is introduced to improve the idea, and an experiential learning algorithm (IEBL) based on Yin-Yang thought is proposed. Based on the settlement monitoring data of a certain section of a subway tunnel before the station project in Guangzhou, the geotechnical parameter inversion method based on improved empirical learning algorithm is adopted to obtain the dynamic geotechnical parameters of the project, and the inversion parameters are applied to the numerical calculation and settlement analysis of the three-dimensional model of the tunnel before the station. The results show that there is spatial variability in the geotechnical parameters obtained by inversion, and the geotechnical parameters change correspondingly with the continuous excavation. The three-dimensional numerical simulation results show that the maximum surface settlement value is -5.52 mm in the middle of the subway station under the influence of multiple factors such as cross passage and construction sequence when the subway station is constructed according to existing design and construction methods. There is a certain area of uplift near the 3# transverse channel, and the maximum uplift is 2.26 mm. The vertical deformation of the ground surface caused by the complete construction process is relatively stable, and the construction of the tunnel before the station has no adverse effect on the road above.