关键词: BP神经网络, 物理力学指标, 因子分析, 黄土湿陷性系数, 预测

Abstract: In order to apply the back propagation (BP) neural network to the evaluation of collapsibility of loess,on the basis of eliminating the correlation between physical and mechanical indices by using the principle of factor analysis,this paper proposes and establishesa BP neural network prediction model of loess collapsibility coefficient based on four physical and mechanical indices including water content, porosity ratio, plasticity index and compression coefficient. The geotechnical test results on loess soil samples from project site of Dingxi to Lintao Expressway were taken as the training and test samples. The predicted value and measured value are compared and analyzed. The results show that the determination coefficient of the training results is 0.95, and the relative error between the predicted value and the measured value is generally less than 11.1% in the prediction analysis.It shows that the proposed BP neural network model can be used to predict the collapsibility of loess.

Key words: BP Neural Network, Physical and Mechanical Index, Factor Analysis, Collapsibility Coefficient of Loess, Prediction