基于ANFIS神经网络的红黏土蠕变模型

Creep Model of Red Clay Based on ANFIS Neural Network

为更好地评价红黏土边坡的蠕变特性和长期稳定性,必须建立合理的红黏土蠕变模型.首先利用自行设计改装的红黏土三轴蠕变试验装置,采用分级加载,对在不同围压下固结完成的红黏土试样进行室内排水三轴蠕变试验,获得了不同围压下的红黏土蠕变全过程曲线.然后采用“陈氏加载法”将分级加载曲线转化为不同偏应力水平下的分别加载曲线,利用等时曲线法获得红黏土的长期抗剪强度.选用不同围压、不同偏应力水平下的部分蠕变试验结果进行样本训练,建立了基于ANFIS神经网络且考虑围压及偏应力影响的红黏土蠕变模型.最后,利用训练完成的本文模型对其他蠕变试验结果进行预测,结果表明本文模型的拟合及预测精度均较高.

Long-term stability assessment of slope characterized by the presence of red clays depends essentially on the creep model of red clay adopted.A specially designed device was used to conduct the tri-axial creep tests on red clay specimens.Deviatoric stresses were imposed on the specimens by stepwise loading,which were consolidated under varying confining stresses.The full-process creep curve of axial strains with increasing deviatoric stress was transformed equivalently to a cluster of creep curves under each stress levels by using"Chen's method".Furthermore,the ultimate deviatoric stress of red clay specimens before yielding in creep tests under varying confining stresses was determined using isochronal curve method,and used to establish the long-term shear strength criterion.A creep model of red clay in terms of axial strain accounting for the deviatoric stress and confining stress was established in the framework of ANFIS neural network.In this framework,a part of test data were used in training the creep model,while the remaining test data were used to examine its capability of predicting the creep response of red clay.A good agreement between the measurements and predictions validates the effectiveness and accuracy of this presented creep model.