袁捷,李杰,焦华成,等. 基于机器学习的刚性道面临界荷位应力预测及参数重要性分析[J].河南理工大学学报(自然科学版)，doi:10.16186/j.cnki.1673-9787.2025020025.

YUAN J, LI J, JIAO H C, et al． Stress Prediction and Parameter Importance Analysis of Critical Load Position of Rigid Pavement Based on Machine Learning [J]. Journal of Henan Polytechnic University( Natural Science) ， doi: 10.16186/j.cnki.1673-9787.2025020025.

基于机器学习的刚性道面临界荷位应力预测及参数重要性分析(网络首发)

袁捷，李杰，焦华成，贾向阳

同济大学 民航飞行区设施耐久与运行安全重点实验室 上海 201804

摘要: 目的 为了快速预测多种飞机荷载作用下不同刚性道面结构的临界荷位应力，基于机器学习的方法对临界荷位应力的预测及参数重要性进行了研究, 方法 本研究通过建立三维刚性道面应力响应有限元模型，计算了不同工况下的临界荷位应力，并收集所有工况的临界荷位应力，构建了临界荷位应力数据库；选用支持向量机（SVM）、卷积神经网络（CNN）、BP神经网络和随机森林（RF）四种机器学习算法构建应力预测模型，选取常见单轴双轮（B737-800）、双轴双轮（B747-400）和三轴双轮（B777-300ER）的临界荷位应力数据（2916组），经过数据筛选和优化，建立了用于模型训练的数据集（2806组）；采用网格搜索与5折交叉验证优化超参数；基于RMSE、MAE、MBE和R²四种评估指标，对模型性能进行量化；通过SHAP方法对输入参数的重要性进行分析，最后，分析临界荷位应力随各道面结构参数的变化规律。结果 BP神经网络模型的预测效果最佳，其R²、MAE、RMSE和MBE分别为0.94246、0.12092、0.16424和0.00034；面层厚度对应力影响最显著，平均绝对SHAP值为49.24%；临界荷位应力随着面层厚度、道基顶面反应模量、基层模量与厚度的增加而减小，最终减小趋势趋于稳定；基于应力在不同结构参数范围内的应力分布和变化，给出了刚性道面设计中各道面结构参数适宜的取值范围。结论 机器学习模型可以很好地预测临界荷位应力，可为刚性道面应力计算和厚度设计提供技术支持。

关键词: 道路工程；刚性道面；临界荷位应力；机器学习；参数重要性

doi: 10.16186/j.cnki.1673-9787.2025020025

基金项目: 国家重点研发计划课题（2019YFB1310603）

收稿日期：2025-02-23

修回日期：2025-04-21

网络首发日期：2025-04-22

Stress prediction and parameter importance analysis of critical load position of rigid pavement based on machine learning (Online)

YUAN Jie, LI Jie, JIAO Hua-cheng, JIA Xiang-yang

The Key Laboratory of Infrastructure Durability and Operation Safety in Airfield of CAAC, Tongji University, Shanghai 201804, China

Abstract: Objectives In order to quickly predict the critical load stress of different rigid pavement structures under various aircraft loads, the prediction of critical load stress and the importance of its parameters were studied based on machine learning method., Methods In this study, the critical load stress under different working conditions was calculated by establishing a three-dimensional finite element model of rigid pavement stress response, and the critical load stress database was constructed by collecting the critical load stress under all working conditions. Four machine learning algorithms, support vector machine (SVM), convolutional neural network (CNN), BP neural network and random forest (RF), were used to construct the stress prediction model. The critical load position stress data (2916 sets) of common single-axle dual-wheel (B737-800), dual-axle dual-wheel (B747-400), and triple-axle dual-wheel (B777-300ER) configurations were selected. After data screening and optimization, a dataset (2806 sets) for model training was established. The hyperparameters were optimized by grid search and 5-fold cross validation. The performance of the model was quantified based on RMSE, MAE, MBE and R². The importance of input parameters is analyzed by SHAP method. Finally, the variation law of critical load stress with structural parameters of each surface is analyzed. Results BP neural network model has the best prediction effect, and its R², MAE, RMSE and MBE are 0.94246, 0.12092, 0.16424 and 0.00034, respectively. The surface thickness has the most significant effect on the stress, and the average absolute SHAP value is 49.24%. The critical load stress decreases with the increase of the thickness of the surface layer, the reaction modulus of the top surface of the foundation, the modulus of the base layer and the thickness, and finally the decreasing trend tends to be stable. Based on the stress distribution and variation in different structural parameters, the suitable value range of structural parameters in rigid pavement design is given. Conclusions The machine learning model can predict the critical load stress well, and can provide technical support for the calculation of rigid pavement stress and thickness design .

Key words: Pavement Engineering; Rigid Pavement; Critical Load Position Stress; Machine Learning; Parameter Importance