基于应变能的砂土液化势BP神经网络模型评估  

Artificial Neural Network Models for Evaluating Sand Liquefaction Resistance Based on Strain Energy Concept

作　　者：胡记磊 王璟 沈文翔 NIMA Pirhadi 万旭升[3] 路建国 HU Jilei;WANG Jing;SHEN Wenxiang;NIMA Pirhadi;WAN Xusheng;LU Jianguo(Key Laboratory of Geological Hazards on Three Gorges Reservoir Area,China Three Gorges Univ.,Yichang 443002,China;College of Civil Engineering&Architecture,China Three Gorges Univ.,Yichang 443002,China;School of Civil Engineering&Geomatics,Southwest Petroleum Univ.,Cheng-du,610500,China)

机构地区：[1]三峡库区地质灾害教育部重点实验室(三峡大学),湖北宜昌443002 [2]三峡大学土木与建筑学院,湖北宜昌443002 [3]西南石油大学土木工程与测绘学院,成都610500

出　　处：《三峡大学学报（自然科学版）》2023年第3期56-61,67,共7页Journal of China Three Gorges University：Natural Sciences

基　　金：四川省科技厅青年科技创新基金项目(2019JDTD0017);土木工程防灾减灾湖北省引智创新示范基地基金项目(2021EJD026)。

摘　　要：液化是地震引起的最具破坏性的现象之一.能量法是液化势评估的常用方法,但已有的模型未考虑细粒含量临界值以及参数不确定性对液化势的影响.因此,本研究基于应变能(W),将数据库以细粒含量28%为临界值分成两个数据集,分别构建了两个BP神经网络(ANN)模型来对砂土液化势进行评估,并通过蒙特卡罗模拟结合响应面分析法进行参数敏感性分析,分别评估了细粒含量临界值以及参数不确定性对土壤液化势的影响.结果显示,提出的两个ANN模型比其它4个现有的模型具有更高的预测性能,其中当仅考虑细粒含量小于28%样本构建模型时,模型预测精度最高,拟合优度达到0.81;敏感性分析表明参数不确定性对砂土抗液化性能存在影响且程度不同,其中曲率系数对W的影响较大,建议在构建模型时加以考虑.Liquefaction is one of the most destructive phenomena caused by earthquakes.The energy method is a common method for liquefaction potential assessment,but the existing models do not consider the influence of the critical value of fines content and parameter uncertainty on the liquefaction potential.Therefore,in this study,two BP neural network(ANN) models were constructed to evaluate the liquefaction potential of sandy soils based on the strain energy(W),and the database was divided into two datasets with 28% fines content as the threshold value,and the effects of the threshold value of fines content and parameter uncertainty on the liquefaction potential were evaluated by Monte Carlo simulation combined with response surface analysis.The results show that the proposed two ANN models have higher prediction performance than the other four existing models,among which the model prediction accuracy is the highest when only samples with a fines content of less than 28% are considered to construct the model,and the goodness-of-fit reaches 0.81.The sensitivity analysis shows that parameter uncertainties have effects on the liquefaction resistance of sandy soils and to different degrees,among which the curvature coefficient has a greater effect on W,which is recommended to be considered in the construction of the model.

关 键 词：液化势 应变能 人工神经网络 细粒含量 敏感性分析 

分 类 号：TU43[建筑科学—岩土工程]