考虑多种损坏构成特征的沥青路面预防性养护决策方法

由于沥青路面损坏构成的多样性, 相同的路面状况指数(pavement condition index, PCI)可能代表不同的损坏组合. 当多种损坏并存且损坏程度接近时, 用PCI和主导损坏(最严重、扣分最多的路面损坏)难以得到具有针对性的养护对策. 因此, 通过对PCI的深入分析, 明确了主导损坏代表性不足的路段, 以现行预防性养护决策方法为基础, 补充了一种考虑损坏构成特征、更具针对性的决策方法. 以上海城市道路近5年的检测、养护数据为分析基础, 首先利用有序聚类算法将路段按PCI水平分组, 分析了不同阶段路面损坏构成和差异水平; 然后, 针对多种损坏并存且损坏差异不显著的路段, 根据预防性养护的实施效果筛选了能够反映正确预防性养护经验的有效养护路段; 最后, 基于有效养护路段建立并对比分析了2个基于BP(back propagation)神经网络的养护决策模型. 结果表明: 当PCI水平介于优良(84.4~93.0分)时, 不同损坏程度接近, 主导损坏代表性不足; 考虑多种损坏构成特征的BP神经网络模型表现出更高的决策精度, 测试集决策正确率达86.20%, 优于仅考虑主导损坏的模型(58.50%). BP神经网络与传统决策树法结合能够优化沥青路面决策过程, 提高养护对策选取的针对性.

Decision method for preventive maintenance of asphalt pavements considering multiple damage characteristics

Because of the diverse compositions of asphalt pavement damage, multiple cases of damage that are inspected using the same pavement condition index (PCI) may yield different damage combinations. When multiple types of damage coexist but the degree of damage is similar, obtaining targeted maintenance measures with PCI and determining the predominant damage (i.e., most severe road damage with the maximum deduction value) are challenging. Therefore, this study considers PCI analysis and an existing preventive maintenance decision-making method to clarify those sections in which the predominant damage was not well-targeted during inspection and proposes a supplementary approach to make more appropriate conservation decisions. Based on detection and maintenance data of urban roads in Shanghai from over the past five years, a sequential clustering method is used to classify road sections according to their PCI levels. The composition of and difference in pavement damage at different levels are analyzed. Then, for those sections with multiple cases of damage and no significant damage differences, road sections that historically reflect proper preventive maintenance are then selected based on whether effective preventive maintenance can be implemented. Finally, two back propagation (BP) neural network models for preventive maintenance decisions are established and compared based on the effective maintenance road sections. The main differences between the two models are the compositions of pavement damage. The results showed that when the PCI levels were high (84.4~93.0 points), the degrees of damage were very similar and the predominant damage was not represented. Of the two BP neural network models, Model 2,which considered multiple damage components, showed a higher decision accuracy. Specifically, its decision accuracy with the test set reached 86.20%. This was significantly better than that of Model 1 (58.50%), which considered only the predominant damage. Combining the BP neural network and traditional decision tree method can help to optimize decision-making processes related to asphalt pavement and improve the selection of maintenance measures.