AFSs-RBF神经网络模型在轻亚黏土地震液化判别中应用研究

Application of AFSs-based RBF neural network model to evaluation of earthquake-induced liquefaction potential in light loam sites

以自适应模糊系统AFSs为基础,运用径向基高斯函数RBF所建立的AFSs-RBF神经网络模型能够同时容纳模糊系统的推理功能和自适应性,动态调节隐节点数即模糊规则数,具有广泛的适用性.将这种模型应用于轻亚黏土地震液化评价中,选择震中距、上覆有效应力、黏粒含量、标贯击数、地下水位、循环应力比等6个与地震和场地条件有关的影响因子作为网络输入参数,对于轻亚黏土场地的液化势判别具体地建立了模糊神经网络模型AFSs-RBF.以唐山7.8级地震中天津某地区的轻亚黏土液化数据为训练样本,经验证和应用表明,这种AFSs-RBF网络具备更高的自适应性和非线性映射能力.

The adaptive fuzzy systems （AFSs） are incorporated with the radial basic function （RBF） to develop an integrated neural network model AFSs-RBF. In this model, the number of the hidden layers or units, i.e. fuzzy rule number, can be dynamically adjusted and widely used in engineering practice. This model is applied to evaluation of earthquake-induced liquefaction potential in light loam sites. The six parameters related to earthquake and site condition which are composed of epicentral distance of earthquake, effective overburden stress, clay percentage, SPT-blow counts, water table, cyclic stress ratio are chosen to develop the AFSs-RBF neural network model with six-index input. The proposed model is applied to the classification of liquefaction potential of sites located in Tianjin area occurring in the Tangshan earthquake with the magnitude of 7.8. It is shown that such an AFSs-RBF network model is capable to offer a more rational prediction of liquefaction potentials of light loam site compared with those by conventional artificial neural network methods.