论文部分内容阅读
为解决BP(back propagation)神经网络收敛速度慢,网络结构需事先定义等缺点,采用了级连相关神经网络模型来建立人工冻土应力和应变之间的关系.基于该模型推导了冻土的一致刚度矩阵形式,利用人工冻土三轴试验数据对神经网络模型进行训练,并用其替换有限元计算中的传统本构模型,将计算结果与性质及含水率相同的冻土的试验结果进行了对比,发现该神经网络本构模型很好地反应了材料的非线性,能够改善数值计算结果,与实测结果吻合地很好,比具有相同隐含层神经元个数的BP模型更接近实测结果.
In order to solve the problems that the back propagation (BP) neural network converges slowly and the network structure needs to be defined in advance, the cascade correlation neural network model is adopted to establish the relationship between stress and strain of artificial permafrost. Based on the model, In the form of uniform stiffness matrix, the artificial neural network model was trained by using the frozen soil triaxial test data, and the traditional constitutive model was replaced by the traditional constitutive model. The experimental results of frozen soil with the same nature and moisture content It is found that the neural network constitutive model reflects the nonlinearity of the material, improves the numerical calculation results, and is in good agreement with the measured results. It is closer to the measured results than the BP model with the same number of hidden layer neurons .