For the prediction of bearing liquid film dynamic lubrication, the elastic basic boundary conditions of the ship's shafting is considered, the finite element method is used to obtain the displacement between the journal and the sleeve, and the problem of liquid film lubrication characteristics is solved when the journal and the shaft sleeve are relatively deflected. Based on the two-dimensional Reynolds equation, the distribution law of the bearing liquid film is obtained, and the difference method and the over-relaxation iterative method are combined to obtain the equivalent support characteristics of the bearing under steady state. Finally, the least square method and convolutional neural network are used to for deep learning and establishment of bearing support characteristics response surface. The predicted response surface can be used as a model basis for shaft rotor dynamics and dynamic calibration.