基于堆栈泛化的膨胀土抗拉强度预测模型

Tensile Strength Prediction Model for Expansive Soils Based on Stacked Generalization

如何准确评估土体的抗拉强度是岩土工程实践中的一个重要问题.该文首先开展一系列试验,获得125组包含干密度、含水量、无侧限抗压强度、单轴抗拉强度、基质吸力、破坏压应变、破坏拉应变、拉剪黏聚力、拉剪内摩擦角等信息在内的试验数据,并在此基础上,利用堆栈泛化算法,建立压实膨胀土单轴抗拉强度预测模型,详尽比较堆栈泛化模型与其他机器学习模型以及基于吸应力的抗拉强度理论模型的预测性能;最后进行特征重要性分析,研究堆栈泛化模型对抗拉强度与其他变量之间的内在联系.结果表明该预测模型的预测性能明显优于参与比较的其他机器学习模型,远优于基于吸应力的抗拉强度理论模型.对抗拉强度影响最大的5个特征最终排序为质吸力(44.6%)>含水量(19.5%)>拉剪内摩擦角(11.9%)>无侧限抗压强度(10.6%)>破坏拉应变(8.0%),与既有研究结果吻合.

Accurate assessment of the tensile strength of soils is an important issue in geotechnical engineering practice.In this paper,a series of tests were carried out to obtain 125 sets of test data containing information on dry density,moisture content,unconfined compressive strength,uniaxial tensile strength,matrix suction,destructive compressive strain,destructive tensile strain,tensile shear cohesion,and tensile shear angle of internal friction,etc..On this basis,the stacked generalization algorithm is used to build the uniaxial tensile strength prediction model for compacted expansive soils,and the prediction performance of the stacked generalization model is thoroughly compared with other machine learning models and theoretical models of tensile strength based on suction stress.Finally,feature importance analysis is performed to investigate how well the stack generalized model captures the relationship between tensile strength and other variables.The results show that the proposed model outperforms all the other models involved.The major influential variables recognized by the proposed model are matrix suction(44.6%)>moisture content(19.5%)>tensile-shear internal friction angle(11.9%)>unconfined compressive strength(10.6%)>failure tensile strength(8.0%),which is consistent with the results of established studies.