关键词: 径向基神经网络 , 最大动剪切模量,  ,  ,  , Hardin 公式 , 模式搜索法

Abstract:  To avoid the complicated work of searching for quantitative experiential formula, a nonlinear relationship between maximum dynamic shear modulus(G_max) and the influence factors including void ratio(e), cell pressure(σ₃), and consolidation ratio(k_c) was built directly by using Radial Basis Function(RBF) neural network. In addition, the optimal value of spread speed(SPREAD) of RBF was calculated by pattern search method to minimize the prediction error. Taking standard sand in Fujian province as an example, the optimal value of SPREAD calculated by pattern search method equals to 2.287 , and the average relative error of G_max predicted by RBF neural network is 0.931 6 % , which is quite small. It shows that RBF neural network can determine G_max under different conditions conveniently and effectively. Besides, the relationship curve of G-γcan also be simulated by this network. Therefore, the method of using RBF neural network to calculate the maximum dynamic shear modulus is recommended to be used widely.

Key words: radial basis function neural network  ,   maximum dynamic shear modulus ,    Hardin formula ,    pattern search method