In order to improve the rationality of mechanical parameters of tunnel surrounding rock, a new inversion model of surrounding rock parameters is proposed based on a tunnel project with a super large-section in Zhuhai. After the initial samples are generated based on Latin hypercube sampling (LHS), the parameter sensitivity analysis is carried out to determine the key parameters of the surrounding rock and improve the sample structure. Then, the whale optimization algorithm (WOA) is used to optimize the number of hidden layer nodes, the initial weights and the thresholds of the extreme learning machine (ELM) to form the LHS-WOA-ELM inversion model. The inversion parameters are substituted into FLAC3D to calculate the deformation and compare with the field measured data. The results show that the parameter sensitivity analysis based on LHS can investigate the co-variation of multi-parameters with fewer samples and and determine the main parameters affecting the displacement of surrounding rock as elastic modulus E, cohesion c and internal friction angle φ. Compared with WOA, ELM and BP algorithm models, the difference between the calculated deformation values obtained by LHS-WOA-ELM inversion model and the measured deformation values is smaller, indicating that the inversion analysis method can well reflect the nonlinear and uncertain characteristics between the surrounding rock parameters and deformation, and further improve the accuracy and efficiency of the surrounding rock inversion in super-large section tunnels, which can provide a reference for determining the design parameters of underground caverns and mining projects.