In this work,a large number of sand triaxial test data were collected and sorted,and a machine learning al-gorithm(support vector machine)was used to establish a prediction model of sand constitutive parameters with the av-erage particle size,non-uniformity coefficient,curvature coefficient,relative compactness,dry density and other easily measured basic physical parameters as input parameters and Duncan-Chang constitutive model parameters as output pa-rameters.From the correlation between input parameters and output parameters,the dry density of input parameters has the greatest influence on output parameters.According to the influence of different kernel functions on SVM predic-tion,RBF kernel function has the best prediction effect.On this basis,the parameters of the Duncan-Chang constitu-tive model are predicted,and the average particle size d50 is taken as the input control parameter to further improve the prediction results of the model.Using the established parameter prediction model,researchers only need to carry out a simple indoor physical property test to obtain basic physical property parameters,which can be used to infer the param-eters of Duncan-Chang model for engineering numerical calculation,so as to improve the efficiency and accuracy of en-gineering analysis,and can also be used to judge the correctness of indoor triaxial test results.
维普
万方数据