Field monitoring of foundation pits alone cannot predict the future deformation of retaining structures. Numerical simulations can predict the deformation of foundation pits and the working state of retaining structures to avoid the risk of foundation pit damage in advance. Accurate inversion of the soil parameters used for simulation and prediction is a key step. The associated multivariable problem is transformed into a single-variable problem by using the interval influence coefficient. Soil layer weightings and excavation step weightings are introduced and exploited to optimize the calculation process, and the soil parameters are calculated through inversion based on the least squares method. Based on actual engineering, the excavation sequence is regarded as a progressive sequence for back analysis, and the parameters of each soil layer are calculated through dynamic calculations with the excavation process in a cycle comprising inversion, prediction, reinversion and reprediction. The soil parameters after inversion are used to predict the maximum value and the depth of the deep horizontal displacement of the retaining structure, which verified the feasibility of the back-analysis method. Compared with the results before inversion, after the final inversion, t the overall error of section 2 is reduced by 67.24%, the overall error of section 3 is reduced by 40.5%, and the overall error of section 4 is reduced by 35%. The prediction curves are all close to the monitoring displacement curves, which plays a good guiding role and ensures the safe construction of the foundation pit. A new effective idea is proposed for the inverse analysis of the composite formation parameters of the deep foundation pit engineering.