基于径向基函数神经网络模型的砂土液化概率判别方法

以国内外25次大地震中的344组场地液化实测资料为基础,通过径向基函数神经网络模型的训练和检验,分析了修正标准贯入击数N1与饱和砂土抗液化强度之间的非线性关系,建立了饱和砂土液化极限状态曲线或抗液化强度临界曲线经验公式。经统计分析,给出了液化和非液化的概率密度函数以及抗液化安全系数与液化概率之间的经验公式,最后导出了具有概率意义的饱和砂土抗液化强度经验公式。当液化概率水平为50%时,即等价于传统的确定性砂土液化判别,该方法预测液化和非液化的可靠性分别为90.4%和81.2%,具有较高的可靠性。本文提出的砂土液化概率判别方法,使工程场地的砂土液化概率判别如同确定性砂土液化判别一样简单、方便,从而使砂土液化概率判别方法用于工程实践和纳入有关规范成为可能。

Probabilistic estimation of sand liquefaction based on neural network model of radial basis function

Based on the 344 liquefaction data of the twenty-five strong earthquakes in the world,through training and testing the neural network model of Radial Basis Function(RBF),the nonlinear relation between corrected blow count N1 of standard penetration test and cyclic resistance ratio CRR of saturated sand was analyzed,and empirical equation CRRcri of liquefaction limit state curve or critical cyclic resistance ratio curve of saturated sand was also constructed.By statistic analysis,probability density functions of liquefaction and non-liquefaction cases as well as empirical equation between safety factor and liquefaction probability of saturated sands were given,then the empirical equation of cyclic resistance ratio CRR of saturated sands with different probability level was educed.When liquefaction probability level was equal to 50%,the present method was consistent to traditional deterministic method of sand liquefaction estimation,and its reliability for liquefaction and non-liquefaction estimation of saturated sands was 90.4% and 81.2%,respectively.The method made the sand liquefaction probabilistic estimation of engineering site as easy and convenient as traditional deterministic method of sand liquefaction estimation.So it was possible that the method of sand liquefaction probability estimation would be applied in the engineering practice and adopted in codes for seismic design.