In a bi-arch tunnel with eight lanes using double-side-drift method, it is the deformation of the drifts rather than the whole tunnel section that is measured due to the restriction from the temporary linings. Only by the time when all the temporary linings are removed can the back analysis be implemented if we choose to use the whole section’s deformation data to get the mechanical parameters of surrounding rock, which is time-consuming. This paper discusses a method which involves both FEM and BP neural network to establish a nonlinear mapping relationship between surrounding rock’s mechanical parameters and drifts’ final deformation rather than the whole section’s deformation with construction sequences taking into consideration, followed by a back analysis on the mechanical parameters using the drifts’ in-situ monitoring deformation data on the basis of the above relationship. There’s a good agreement between the in-situ deformation data and the numerical deformation data achieved from a normal analysis using the back analyzed parameters.