Application of an improved RBF neural network on aircraft winglet optimization design

A self-adaptive radial basis function (RBF) neural network is proposed in order to improve the prediction accuracy of the original RBF. A self-adaptive vector with the same dimension as the sample vector is introduced into the traditional RBF network. In contrast to other RBF neural network models, the current approach achieves the self-adaptive construction of the network by altering the form of the basis function directly, which reduces the number of variables to be optimized. This adaptive approach substantially changes the impact of the center and width of the RBF neural network on its prediction as well as the influence of each variable of the independent vector on the dependent vector. Thus the introduced vector enables the adaptability of the RBF neural network with respect to variant problems. Moreover, the accuracy and the universality of the prediction model are also improved due to the optimization of the self-adaptive vector. The proposed self-adaptive RBF neural network is applied to a winglet optimization design of a wing-body-winglet configuration. The optimization objective is to minimize the cruise drag with wing-root bending moment restriction. The optimization result confirms the effectiveness and the capability for engineering application of the self-adaptive RBF neural network.