关键词: 均匀设计, 蒙特卡罗模拟, RBF神经网络, 挡土墙, 稳定性

Abstract: In the method of response surface Monte-Carlo based on artificial neural network, the uniform design, a kind of design of computer experiments, is introduced to arrange training samplings set of RBF neural network that combines the advantages of uniform design and RBF neural network organically. On the one hand, it may decrease the number of training samples as fully as possible, and on the other, it fills training samples into the parameter space uniformly, and enhances the prediction accuracy of RBF neural network which has strongly local approximate capacity. In the probability analysis of retaining wall stability against sliding, for example, the results of the proposed method are very consistent with those of the direct Monte-Carlo simulations. And in addition this probability analysis shows that friction coefficient of the wall base f, the inner friction angle f ,unit weight γ of filled soils, and the friction angle of the soils against wall back δ have influences on the retaining wall stability against sliding from high to low in turn.

Key words: uniform design, Monte-Carlo simulation, RBF neural network, retaining wall, stability