Surface settlement is an important problem in shield tunnel construction. The traditional machine learning model neglects the dynamic characteristics and internal physical laws of land subsidence when predicting land subsidence. In order to predict the surface settlement caused by shield tunnel construction more accurately and effectively, a Physical Information Extreme Learning Machine model (PIELM) was proposed to predict the surface settlement. Based on the analytical solution of Verruijt-Booker’s semi-elastic plane hypothesis, the physical governing equations of tunnel settlement deformation in semi-infinite space are solved. By coupling the governing equation with the extreme learning machine, an improved data-physics extreme learning machine model is constructed. The verification analysis shows that the prediction accuracy of PIELM is obviously better than that of the traditional single machine learning model ELM model. The results show that the accuracy of PIELM model is 82.26% higher than that of traditional ELM model. PIELM also greatly increases the speed of operations compared to the recently proposed PINN. Physical data dual drive prediction model can greatly improve the accuracy of land surface settlement prediction. It is proved that the model has good generalization ability and provides a new method for shield tunnel construction safety monitoring.