Abstract: Objectives A generalized data-driven inverse prediction method for the wake flow field at the paddle disk is developed in order to reduce the dependence on traditional design experience. Methods By applying coordinate affine transformation, the unstructured flow field is first converted into polar velocity data with identical dimensions. Thereafter, a parametric approach is employed to derive a low-dimensional representation of the high-dimensional wake flow field. Control points are established to facilitate the intuitive adjustment, and a Kriging surrogate model is selected for inverse learning from the inflow angle and stern rudder geometric parameters to the compressed wake flow field distribution. Results The MAPE of the Inverse Kriging method for the axial position of the stern rudder and the inlet angle are 0.051% and 0.045%, and the R² are 0.9993 and 0.9983, respectively. Conclusions The proposed method plays as a fast and smart tool for the inverse design of underwater vehicle.