Abstract: The pavement skid resistance performance index is one of the important indexes to ensure driving safety. Based on the test data of RIOHTrack full-scale test track, this paper selects 357 sets of test data of four structures as input variables, including STR2, STR4, STR9 and STR16. The three influence factors of temperature at pavement surface, equivalent single axle load times(ESALs)and pavement abrasion frequency as input variables, while the british pendulum number(BPN)was taken as output variable. The visualized and implicit prediction model of BPN value are constructed by using group data as samples. The 300 groups of data is taken as training samples, while the 57 groups of data is taken as the verifying samples. Based on the assumption that temperature has a certain influence range on the skid resistance performance of asphalt pavement, the visualized prediction model of BPN value is constructed. The correlation coefficient(R²)of this model is 0.625, while the average relative error is 10.227%. At the- DOI:10.3969/j.issn.1000-1565.2024.05.001足尺试验环道路面摆值变化趋势预测吴将丰,王旭东,关伟,肖倩(交通运输部公路科学研究院 基础研究创新中心,北京 100088)摘 要:路面抗滑性能是路面使用性能的重要指标之一,是行车安全性的重要保障.依托RIOHTrack足尺试验环道的检测数据,选用足尺试验环道STR2、STR4、STR9和STR16 4种结构的357组路面抗滑性能检测数据,采用路面温度、累计标准轴载作用次数和路面磨耗次数3个主要影响因素作为自变量,路面摆值(BPN)指标作为因变量,选用300组检测数据作为训练样本,其余57组数据作为验证样本,构建了BPN的显示化和隐式化预测模型.基于温度对沥青路面抗滑性能具备一定影响范围的假定,构建了路面摆值的显示化预测模型,其模型的相关系数(R²)为0.625,模型预测平均相对误差为10.227%.同时,采用不同的隐含层神经元和训练函数,构建了BP神经网络预测模型(隐式化预测模型),模型的BPN预测值和真实值基本吻合,变化趋势一致,平均相对误差为4.484%.研究提高了路面抗滑性能指标预测的有效性和准确性,不同预测模型具备不同的应用前景,为路面抗滑性能的预测分析提供了参考和依据.关键词:路面抗滑性能;BP神经网络;足尺试验环道;预测模型中图分类号:U416.2 文献标志码:A 文章编号:1000-1565(2024)05-0449-10Prediction of pavement BPN value based on RIOHTrack full-scale test trackWU Jiangfeng, WANG Xudong, GUAN Wei, XIAO Qian(Fundamental Research Innovation Center, Research Institute of Highway, Ministry of Transport, Beijing 100088, China)Abstract: The pavement skid resistance performance index is one of the important indexes to ensure driving safety. Based on the test data of RIOHTrack full-scale test track, this paper selects 357 sets of test data of four structures as input variables, including STR2, STR4, STR9 and STR16. The three influence factors of temperature at pavement surface, equivalent single axle load times(ESALs)and pavement abrasion frequency as input variables, while the british pendulum number(BPN)was taken as output variable. The visualized and implicit prediction model of BPN value are constructed by using group data as samples. The 300 groups of data is taken as training samples, while the 57 groups of data is taken as the verifying samples. Based on the assumption that temperature has a certain influence range on the skid resistance performance of asphalt pavement, the visualized prediction model of BPN value is constructed. The correlation coefficient(R²)of this model is 0.625, while the average relative error is 10.227%. At the- 收稿日期:2023-10-20;修回日期:2023-11-21 基金项目:科技基础资源调查专项资助项目(2022FY101400);交通运输部公路科学研究院交通强国试点攻关项目(QG2021-1-1) 第一作者:吴将丰(1988—),男,交通运输部公路科学研究所助理研究员,主要从事沥青路面长期服役性能研究.E-mail:jiangfeng.wu@rioh.cn 通信作者:王旭东(1968—),男,交通运输部公路科学研究所研究员,主要从事长寿命沥青路面结构与材料设计与建造技术的研究.E-mail:xd.wang@rioh.cn 第5期吴将丰等:足尺试验环道路面摆值变化趋势预测河北大学学报(自然科学版) 第44卷same time, different hidden layer neurons and training functions are adopted to construct BP neural network prediction model of BPN value. The predicted BPN value is basically consistent with the test value while the average relative error is 4.484%. Different prediction models have different application prospects. It improves the effectiveness and accuracy of pavement skid resistance performance index prediction, which provides reference for pavement skid resistance performance detection and analysis.

Key words: pavement skid resistance performance, BP neural network, RIOHTrack full-scale test track, prediction model