中外公路2025，Vol.45Issue(1)：67-72,6.DOI:10.14048/j.issn.1671-2579.2025.01.008
基于LSTM+Transformer的冻融循环作用下路基土永久变形预估模型
Permanent Deformation Prediction Model of Subgrade Soil under Freeze-Thaw Cycles Based on LSTM and Transformer
摘要
Abstract
To accurately predict the permanent deformation of subgrade soil under the coupling effect of freeze-thaw cycles and traffic loads,a hybrid neural network model based on long short-term memory(LSTM)and Transformer was proposed,building upon the results of dynamic triaxial tests.This model effectively captured the temporal dependencies and complex interactive effects among input variables,thereby significantly enhancing the accuracy and generalization ability of permanent deformation prediction of subgrade soil.The results indicate that higher confining pressure improves the resistance of subgrade soil to deformation,but its permanent deformation behavior is still significantly influenced by the magnitude of cyclic loading and the number of freeze-thaw cycles.A comparative analysis with traditional empirical regression models validates the superiority of the proposed hybrid model in addressing nonlinear deformation issues.Furthermore,sensitivity analysis results demonstrate that confining pressure and fluid limit are the primary factors affecting the permanent deformation of subgrade soil.This study can provide reference and guidance for the construction of durable subgrade in seasonally frozen areas.In the design stage,it is essential to select fillers carefully and implement anti-freezing structures,while during the operation and maintenance stage,attention should be paid to overloading phenomena.关键词
路基工程/永久变形/冻融循环/动三轴试验/混合神经网络模型Key words
subgrade engineering/permanent deformation/freeze-thaw cycle/dynamic triaxial test/hybrid neural network model分类
交通工程引用本文复制引用
张安顺..基于LSTM+Transformer的冻融循环作用下路基土永久变形预估模型[J].中外公路,2025,45(1):67-72,6.基金项目
国家重点研发计划项目(编号:2023YFB2603900) （编号:2023YFB2603900）
