Abstract: The ensemble neural network approximation model method with variable weight coefficient is used to replace the structural finite element calculation in order to analyze the structural performance of the composite duct supporting structure of an air cushion vehicle for efficient structural design. Several neural network models are establishedaccording to the sensitivity of the ply design parameters to the structural performance，and the predicted response values are obtained from the weight coefficients and the output of each model. Based on the designability of composite materials，the influence of material parameters on the performance of the supporting structure is analyzed microcosmically and macroscosmically level by using the optimized Latin hypercube design method. The results show that: the approximate model for the composite material has a high accuracy；and the fiber elastic modulus and matrix shear modulus have significant impact on the mechanical properties of the composite material of the supporting structure. The increase in fiber volume fraction can improve the stiffness and stability of the structure when the properties of component materials are fixed. The influences of fiber volume fraction and ply angles on the structural failure are different under different stress states，so special consideration should be taken in the design.The proposed method provides a theoretical basis for the structural performance analysis and the integrated design of composite material duct supporting structure.

Key words: aircushionvehicle, compositeduct, supportingstructure, neuralnetworkapproximatemodel, mechanicalperformance