Abstract: The ground load acting on the structure is crucial for the design of ultra-large diameter shield tunnels. Taking the Zhuhai Mangzhou Tunnel project as the basis, a load inversion model is established using the GA-BP algorithm based on accurate measured values of lining axial force and bending moment for the shore crossing section.The model uses measured segment internal forces as input and vertical soil pressure on the segment as output. For the underwater crossing section, the GA-BP inversion model is optimized using transfer learning to realize the inversion analysis of vertical soil pressure and lateral pressure coefficient. The results show that for the uniform stratum of the shore crossing section, the relative error between the inverted vertical soil pressure from the GA-BP neural network model and the field measurement is small, indicating that the GA-BP neural network is feasible for the inversion of vertical load on ultra-large diameter shield tunnels. For the complex strata in the underwater crossing section, the GA-BP inversion model optimized by transfer learning can reasonably invert the values of vertical soil pressure and lateral pressure coefficient, demonstrating that the improved TL-GA-BP algorithm is reliable and practical.