Abstract Based on the Xiaolangdi North Bank Irrigation Area Project, this study combines numerical simulation and BP neural network methods to investigate the sensitivity of tunnel soil and its parameter inversion under continuous heavy rainfall. The research results indicate that changes in water-level and soil strength parameters have a significant impact on the deformation of tunnel surrounding rock. By comparing the sensitivity factors of different parameters, the main parameter sensitivities affecting the displacement of tunnel surrounding rock were determined to be water level, internal friction angle, and cohesion. The mechanical characteristics of the tunnel construction process were analyzed using finite difference method numerical analysis software FLAC3D, and the results were used as a sample dataset for inversion analysis. Through neural network inverse analysis based on orthogonal design method, the cohesion and internal friction angle of loess layer ④, loess layer ④-1, and loess layer ⑤ were determined, and the data of groundwater level elevation were obtained. Field applications proved the effectiveness and rationality of this method. Keywords: tunnel; water level; soil mechanics parameters; sensitivity analysis; BP neural network