Abstract: Stratospheric airship can stay at an altitude of 20 km for a long time and carry various loads to achieve long-term stable applications. Conventional stratospheric airship configuration mainly includes a low-resistance streamline hull and inflatable “X”-layout fins that realize the self-stabilization. A fast aerodynamic predictive method is needed in the optimization design of airship configuration and the flight performance analysis. In this paper, a predictive surrogate model of aerodynamic parameters is constructed for the stratospheric airship with “X” fins based on the neural network. First, a geometric shape parameterized model, and a flow field parameterized model were established, and the aerodynamic coefficients of airships with different shapes used as the training and test samples were calculated based on computational fluid dynamics (SA turbulence model). The improved Bayesian regularized neural network was used as the surrogate model, and 20 types of airships with different shapes were used to test the effectiveness of network. It showed that the correlation coefficients of Cx, Cy, Cz, CM,x, CM,y, CM,z were 0.928 7, 0.991 7, 0.991 9, 0.958 2, 0.986 1, 0.984 2, respectively. The aerodynamic coefficient distribution contour at different angles of attack and sideslip angles is used to verify the reliability of the method. The method can provide an effective way for a rapid estimation of aerodynamic coefficients in the airship design.