理论数据协同分析的路面性能预测  

Pavement condition prediction based on theory-data collabrative analysis

作　　者：刘文博 杨旭 汪海年[1] 徐坤[1] 范泰博[1] LIU Wen-bo;YANG Xu;WANG Hai-nian;XU Kun;FAN Tai-bo(School of Highway,Chang'an University,Xi'an 710064,Shaanxi,China;School of Future Transportation,Chang'an University,Xi'an 710064,Shaanxi,China)

机构地区：[1]长安大学公路学院,陕西西安710064 [2]长安大学未来交通学院,陕西西安710064

出　　处：《长安大学学报(自然科学版)》2025年第1期38-49,共12页Journal of Chang’an University(Natural Science Edition)

基　　金：国家重点研发计划项目(2021YFB2601000)。

摘　　要：道路变形及结构强度是路面性能评价的重要内容,实现上述两者的精确预测对于道路科学养护决策具有重要意义,但当前研究对于其发展预测精度仍有待提升。为提升路面性能预测效果,提出一种理论-数据双驱动的路面性能预测方法。首先基于影响机理理论分析确定路面性能发展的关键影响因子,随后基于分析结果及北京环道足尺试验场数据确定模型输入、输出变量,最终提出一种时间卷积网络-门控制循环融合预测模型(TCN-GRU)实现车辙、承载力的发展预测,并与基础单模型及时间卷积网络-长短时记忆网络融合模型(TCN-STM)进行对比。研究结果表明:对于车辙深度预测,TCN-GRU取得了最好的预测性能,其M_(SE)、R_(MSE)、M_(AE)及R^(2)分别为22.635、4.758、3.319及0.940,其中R^(2)相比单模型(TCN、GRU)分别提升0.53%与0.86%;对于弯沉值发展预测,TCN-GRU同样取得了最好的预测性能,其M_(SE)、R_(MSE)、M_(AE)及R^(2)分别为8.009、2.830、1.819及0.850,R^(2)相比单模型(TCN、GRU)分别提升5.85%与2.04%;提出的TCN-GRU对道路车辙及承载力发展预测效果最好,其充分结合了TCN的长时依赖建模能力和GRU的高效状态更新优势,提升了序列数据的预测准确性与效率,可以基于历史数据实现对特定道路的车辙及承载力发展的精确预测,为公路养护管理部门科学养护决策提供数据支撑。The deformation of the road and the structural strength are important contents of pavement performance evaluation.The precise prediction of the above two is of significant importance for road scientific maintenance decision-making.However,the development of the current research in prediction accuracy still needs to be improved.A theory-data dual-driven pavement performance prediction method is proposed to improve the pavement performance prediction effect.Firstly,the key influencing factors of pavement performance development are determined based on the influence mechanism theory.Then,based on the analysis results and the Beijing ring road full-scale test site data,the model input and output variables are determined.Finally,a temporal convolutional network-gated recurrent unit(TCN-GRU)fusion prediction model is proposed to predict the development of rutting and bearing capacity.This model is compared with the basic single model and the temporal convolutional network-long short term memory(TCN-LSTM)fusion model.The research results show that for rut depth prediction,the TCN-GRU achieves the best prediction performance,with M_(SE),R_(MSE),M_(AE),and R^(2)values of 22.635,4.758,3.319,and 0.940,respectively.Compared to the single models(TCN,GRU),R^(2)is improved by 0.53%and 0.86%,respectively.For deflection value development prediction,the TCN-GRU also achieves the best prediction performance,with M_(SE),R_(MSE),MAE,and R^(2)values of 8.009,2.830,1.819,and 0.850,respectively.R^(2)is improved by 5.85%and 2.04%compared to the single models(TCN,GRU).The proposed TCN-GRU demonstrates the best prediction performance for road rutting and bearing capacity development.It fully combines the long-term dependency modeling capability of TCN and the efficient state update advantage of GRU,improving the prediction accuracy and efficiency of sequence data.It can accurately predict the development of rutting and bearing capacity for specific roads based on historical data,providing data support for scientific maintenance decision-ma

关 键 词：道路工程 路面性能预测 双驱动 理论-数据 融合模型 TCN-GRU 

分 类 号：U418.6[交通运输工程—道路与铁道工程]