混合正向组合预测和逆向目标优化的沥青混合料配合比设计  

Asphalt mixture proportion design combining forward combination prediction and reverse target optimization

作　　者：艾长发[1,2] 张家康 焦方辉[1,3] 甘国安[1,2] 张傲南 AI Changfa;ZHANG Jiakang;JIAO Fanghui;GAN Guoan;ZHANG Aonan(School of Civil Engineering,Southwest Jiaotong University,Chengdu 610031,China;Highway Engineering Key Laboratory of Sichuan Province,Southwest Jiaotong University,Chengdu 610031,China;Department of Transportation of Sichuan Province,Chengdu 610041,China)

机构地区：[1]西南交通大学土木工程学院,成都610031 [2]西南交通大学道路工程四川省重点实验室,成都610031 [3]四川省交通运输厅,成都610041

出　　处：《东南大学学报(自然科学版)》2025年第2期496-503,共8页Journal of Southeast University:Natural Science Edition

基　　金：国家重点研发计划资助项目(2022YFB2602602);国家自然科学基金面上资助项目(52278462).

摘　　要：为实现沥青混合料配合比的高效合理设计,构建了包含沥青混合料设计参数和设计指标的数据集,以优化初试级配拟定、设计级配和最佳沥青含量确定等流程。建立了4种集成学习正向预测模型,并提出了基于表征模型解释性的Shapley值的组合策略,得到正向预测组合模型。混合正向预测组合模型和灰狼优化算法,得到符合拟定设计指标目标值的配合比,并基于室内试验进行验证。结果表明,该组合策略显著提升了模型性能,决定系数至少提升5%,平均绝对百分数误差至少降低5%。组合模型表现出较好的预测精度和泛化能力,决定系数大于0.8,平均绝对百分数误差小于10%,模型训练集和测试集性能偏差小于10%。设计指标试验值与目标值的相对误差小于10%,说明所提方法具备较好的可靠性。To achieve efficient and reasonable asphalt mixture proportion design,a dataset containing asphalt mixture design parameters and design indicators is constructed to optimize the process of initial gradation proposal and the determination of design gradation and optimal asphalt content.Four ensemble learning-based forward prediction models are established.A combination strategy based on Shapley values,which represent model interpretability,is proposed to construct a forward prediction combination model.By combining the forward prediction combination models with the grey wolf optimization algorithm,the mixture proportions meeting the proposed target values of design indicators are obtained designed and validated through indoor experiments.The results show that the combination strategy can significantly improve model performance with the coefficient of determination increasing by at least 5%and the mean absolute percentage error decreasing by at least 5%.The combination model exhibits good prediction accuracy and generalization ability.The coefficient of determination is greater than 0.8,and the mean absolute percentage error is less than 10%.The performance deviation between the training set and the test set is less than 10%.The relative error between the experimental values and target ones of design indicators is less than 10%,showing this method has good reliability.

关 键 词：道路工程 配合比设计 集成学习 灰狼优化 SHAPLEY值 

分 类 号：U414[交通运输工程—道路与铁道工程]