Evaluation of stress history of clays based on intelligent CPTU machine learning algorithm
-
摘要: 土体应力历史是衡量土体稳定性、变形特性的重要指标,常采用超固结比（OCR）表示。基于江苏黏性土孔压静力触探（CPTU）原位测试数据集,以室内固结试验结果为参考值,采用多元自适应回归样条（MARS）和自适应模糊神经网络（ANFIS）智能算法对黏性土应力历史进行评价,并将预测结果与室内试验结果和CPTU经验关系式估计值进行对比,最后进行了参数敏感性分析。结果表明：MARS模型和ANFIS模型均能够准确地预测黏性土的OCR值,且准确度均明显高于传统CPTU经验关系式；相比而言,MARS模型效果更佳。工程实践中,建议采用CPTU原始测试参数（qt、fs和u2）作为机器学习输入变量。MARS模型敏感性分析结果与理论研究结果一致,进一步验证了MARS模型的可靠性。提出的智能CPTU模型可以准确地预测黏性土OCR,指导工程实践。Abstract: The stress history is an important index to measure the stability and deformation characteristics of soils, which is often expressed by the overconsolidation ratio (OCR). Based on the CPTU dataset of Jiangsu Province, and taking the laboratory oedometer test data as the reference values, the stress history is evaluated using the multiple adaptive regression splines (MARS) and adaptive fuzzy neural network (ANFIS) algorithms. Then, the results are compared with the reference values and the estimated results of the traditional CPTU method. Finally, the sensitivity analysis is carried out to study the effect of input parameters. The results show that both the MARS model and the ANFIS model can accurately predict the OCR, and the performance is significantly better than that of the traditional CPTU model. Moreover, the MARS model performs best among all the models. In engineering practice, the original CPTU test parameters (qt, fs and u2) are recommended as the input variables. The results of sensitivity analysis of the MARS model are consistent with those of theoretical analysis, which further proves the reliability of the MARS model. The proposed intelligent models can more accurately predict the OCR of clays and guide engineering practice.
-
-
![]()
图 1 Robertson土分类图
Figure 1. Robertson soil behavior type classification chart based on normalized CPTU data
![]()
图 2 MARS模型和CPTU传统模型预测结果
Figure 2. Predicted results by MARS model and traditional CPTU model
表 1 机器学习模型输入参数
Table 1 Input parameters of machine learning models
组合 输入参数 组合A qt,fs,u2,σ′v0,σv0 组合B Qt,Fr,Bq,σ′v0,σv0 
下载: 导出CSV
表 2 MARS模型和ANFIS模型预测结果评价
Table 2 Performances of MARS model and ANFIS model
模型 输入参数 评价指标 R2 RMSE MAE MARS模型 组合A 0.949 0.710 0.505 组合B 0.933 0.742 0.535 ANFIS模型 组合A 0.923 0.799 0.591 组合B 0.916 0.832 0.588
下载: 导出CSV
表 3 OCR预测模型结果对比
Table 3 Comparison of predicted models by OCR
OCR预测模型 评价指标 R2 RMSE MAE MARS模型 0.949 0.710 0.505 ANFIS模型 0.923 0.799 0.591 MLR模型 0.867 1.028 0.699 Chen和Mayne模型 0.849 1.117 0.751
下载: 导出CSV
-
[1] MAYNE P W. Determination of OCR in clays by piezocone tests using cavity expansion and critical state concepts[J]. Soils and Foundations, 1991, 31(2): 65-76. doi: 10.3208/sandf1972.31.2_65
[2] 刘晓燕, 蔡国军, 邹海峰, 等. 基于CPTU数据融合技术的黏性土应力历史与强度特性评价研究[J]. 岩土工程学报, 2019, 41(7): 1270-1278. https://www.cnki.com.cn/Article/CJFDTOTAL-YTGC201907013.htm LIU Xiao-yan, CAI Guo-jun, ZOU Hai-feng, et al. Prediction of stress history and strength of cohesive soils based on CPTU and data fusion techniques[J]. Chinese Journal of Geotechnical Engineering, 2019, 41(7): 1270-1278. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YTGC201907013.htm
[3] ZHAO Z N, DUAN W, CAI G J. A novel PSO-KELM based soil liquefaction potential evaluation system using CPT and Vs measurement[J]. Soil Dynamics and Earthquake Engineering, 2021(150): 106930.
[4] LUNNE T, POWELL J J M, ROBERTSON P K. Cone Penetration Testing in Geotechnical Practice[M]. CRC Press, 2002.
[5] ROBERTSON P K. Interpretation of cone penetration tests—a unified approach[J]. Canadian Geotechnical Journal, 2009, 46(11): 1337-1355. doi: 10.1139/T09-065
[6] GOH A T C, ZHANG W G. An improvement to MLR model for predicting liquefaction-induced lateral spread using multivariate adaptive regression splines[J]. Engineering Geology, 2014, 170(3): 1-10.
[7] KAYA Z. Predicting liquefaction-induced lateral spreading by using neural network and neuro-fuzzy techniques[J]. International Journal of Geomechanics, 2016, 16(4): 04015095. doi: 10.1061/(ASCE)GM.1943-5622.0000607
[8] ZHANG W G, GOH A T C, ZHANG Y M, et al. Assessment of soil liquefaction based on capacity energy concept and multivariate adaptive regression splines[J]. Engineering Geology, 2015, 188: 29-37. doi: 10.1016/j.enggeo.2015.01.009
[9] ZHANG W G, GOH A T C. Multivariate adaptive regression splines and neural network models for prediction of pile drivability[J]. Geoscience Frontiers, 2016, 7(1): 45-52. doi: 10.1016/j.gsf.2014.10.003
[10] MOAYEDI H, RAFTARI M, SHARIFI A, et al. Optimization of ANFIS with GA and PSO estimating α ratio in driven piles[J]. Engineering With Computers, 2020, 36(1): 227-238. doi: 10.1007/s00366-018-00694-w
[11] ZOU H F, LIU S Y, CAI G J, et al. Multivariate correlation analysis of seismic piezocone penetration (SCPTU) parameters and design properties of Jiangsu quaternary cohesive soils[J]. Engineering Geology, 2017, 228(1/2): 11-38.
[12] SY CHEN B, MAYNE P W. Statistical relationships between piezocone measurements and stress history of clays[J]. Canadian Geotechnical Journal, 1996, 33(3): 488-498. doi: 10.1139/t96-070
计量
- 文章访问数: 216
- HTML全文浏览量: 27
- PDF下载量: 127
