Abstract: To solve the problems of too many parameters, difficult to obtain and longer time-consuming in numerical simulations of ultimate displacement in the process of tunnelling, the Gaussian process algorithm (GP) combined with the differential evolutionary algorithm (DE) was adopted. The selection of GP hyperparameters in the training process was realized by the DE instead of the conjugated gradient method, which solved the problem that the conjugated gradient method was too dependent on the initial values and easy to fall into local optimal, and improved generalization performances of the GP. A DE-GP method for predicting ultimate displacement during tunnelling was proposed, which simplified the calculation process of ultimate displacement. The ultimate displacement of a highway tunnel during tunnelling was predicted and analyzed. The results showed that the DE-GP coupled model had a better generalization performances than the GP and the LS-SVM algorithm; the forecasting results of ultimate displacement agreed well with the numerical calculations, and the computational efficiency had obviously improved; the method provided a new way to calculate the ultimate displacement.