Reliable rock mechanics parameters are crucial for scientifically revealing the mechanical behavior of complex geological tunnels, enabling rational design, and ensuring construction safety. Through 25 groups of orthogonal finite element calculations on tunnel models and network sample construction, a novel swarm intelligence algorithm—sparrow search algorithm (SSA)—was introduced to optimize the weights and thresholds of a classical back-propagation (BP) neural network. This led to the development of an SSA-BP intelligent inversion model for rock mechanics parameters. Results demonstrate that the SSA-BP model significantly outperforms the traditional BP neural network in prediction accuracy and reliability. It exhibits stronger generalization and robustness, particularly for complex datasets, effectively addressing the BP network’s tendency to converge to local minima. The model was applied to invert mechanical parameters of surrounding rock in a complex mudstone-shale interbedded tunnel on Hubei’s Anlai Expressway. Using deformation monitoring data, an optimal exponential fitting model was established to analyze the nonlinear deformation characteristics of tunnel surrounding rock, accompanied by recommended engineering measures. Forward calculations via a three-dimensional mechanical model, using parameters inverted by SSA-BP, showed that the vault settlement error of typical cross-sections relative to field measurements fell within 1.41–5.29 %. Ground settlement and inverted arch uplift also matched well with field data, validating the model’s rationality and reliability. This research enriches the application of intelligent inversion methods for complex rock mechanics parameters, offering both theoretical insights and practical engineering value.