【摘要】 砂土地震液化的影响因素具有高度的非线性关系,而神经网络在处理非线性问题上具有其独特的优越性。本文在探讨输入层模式的选择以及砂土液化影响因素的基础上,采用改进的贝叶斯正则化方法和"提前停止"算法建立了砂土地震液化预测模型,通过实例计算和模型评价,表明本模型的计算结果与规范法、改进的Seed简化法以及基于传统BP网络算法的计算结果相比具有更高的预测精度和较小的训练步长,并采用该模型对厦门集美大桥新环岛互通桥区进行砂土液化评价,证明了该模型具有较高的精度和泛化能力。更多还原

【Abstract】 The affecting factors of sand seismic liquefaction are highly nonlinear and the neural network has originality in processing non-linear problems.Based on the study of the selection of input layer model and the affecting factors of sand seismic liquefaction,the model for predicting sand liquefaction is built utilizing Bayesian regularization method and early stopping method.Through practical computation examples and the assessment of the model,the model is manifested to have much more accurate results and less learning factors than the norm method,Seed’s simplified method and the calculation based on the traditional BP.The model is applied to evaluating sand liquefaction potential of new Island Ring mutual access area of Amoy Jimei Bridge,and the results show that the model has high precision and high generalization ability.更多还原