【作者】 张慧；

【导师】 杨顺新； 王捷；

【作者基本信息】 东南大学 ， 交通运输（专业学位）， 2025， 硕士

【摘要】 随着高速公路路网规模的持续扩大和交通流量的快速增长,高速公路路面状况恶化与交通安全风险提高的问题日益突出,养护工程需求量不断增长,对养护方案决策提出了新的挑战。此外,交通事故数量的增长对养护工程提出了新的要求。例如,在事故高发路段,需要及时修复坑洞、裂缝,防止因路面损坏引发二次事故;在弯道、坡道等处,往往需要采用高摩擦系数的路面材料或进行防滑处理。然而,传统的路面养护决策多基于性能衰变规律与养护成本目标,对交通安全因素的量化分析存在明显缺失,并且缺少将交通安全综合评价结果应用于路面养护决策的研究。此外,实际工程数据中有养护措施与无养护措施的数据分布失衡问题限制了预测模型的精度,从而影响路面养护方案的选择。针对上述问题,本研究以山西省高速公路为研究对象,提出了基于数据驱动的考虑安全改善的路面养护决策框架,主要展开了以下四部分研究内容:首先,收集整理研究所需数据集,主要包括路段基础信息数据、2018-2024年交通事故数据、2021-2024年路面性能数据、2023-2024年养护工程数据、2023年交通量数据和气候数据共六项基础数据,并对数据格式、缺失和异常数据进行了数据预处理。统计分析了高速公路交通事故数据、路面性能数据和养护工程数据,表明当前高速公路交通安全风险较高,路面破损病害较为严重且实际养护工程数据中养护措施数量分布失衡,并结合实例分析发现养护工程对交通安全有一定的改善作用。其次,构建了基于历史事故数据的高速公路交通安全综合评价三层指标体系。该体系涵盖了事故率与事故严重程度两个一级指标以及其下10项二级指标。采用熵权法与层次分析法的组合赋权,确定了各项指标的权重,并结合(Technique for Order Preference by Similarity to Ideal Solution,TOPSIS)法计算交通安全综合评价结果。基于上述体系,量化得到高速公路各路段各桩号处的交通安全风险指标(Traffic Safety Risk Index,TSRI),为后续预测与决策提供量化依据。再次,确定了基于CTGAN-RF的交通安全风险指标(TSRI)预测模型。为开展考虑安全改善的路面养护决策研究,将养护措施作为变量之一输入交通安全预测模型中,并针对当前数据中养护措施不平衡的问题,利用条件表格生成对抗网络(Conditional Tabular Generative Adversarial Network,CTGAN)和Copula方法对交通安全特征数据集进行数据增强。比较分析生成数据与原始数据的数据结构和相关性,确定增强效果更优的CTGAN模型,形成增强数据集。使用增强数据集建立随机森林(Random Forest,RF)、梯度提升树(Gradient Boosting Decision Tree,GBDT)和极度梯度提升树(e Xtreme Gradient Boosting,XGBoost)预测模型,对比分析三个模型的预测效果,确定预测精度更高的RF模型作为交通安全风险指标(TSRI)的预测模型。最后,建立了兼顾路面性能改善和安全提升的多目标路面养护决策模型。将最大化养护效益、最小化养护成本和最大化安全效益作为养护目标,并添加路面性能、养护费用和安全性能三项惩罚函数。采用基于精英保留策略的非支配排序遗传算法(Nondominated Sorting Genetic Algorithm II,NSGA-Ⅱ)构建路面养护决策模型。将该模型与仅考虑养护效益和养护成本的双目标决策模型以及实际养护措施进行对比,验证了该模型在满足路面性能提升和养护成本的同时,交通安全性能也得到提升,为养护决策模型的完善提供了参考。综上所述,本研究提出了基于数据驱动的考虑安全改善的路面养护决策方法。该方法弥补了当前研究的一系列不足之处,实现交通安全风险降低、路面性能提升与养护成本控制的协同优化,为路面养护决策中充分考虑交通安全风险提出了一个统一框架。因此,本研究为高速公路养护管理部门制定养护决策方案提供了新思路和参考依据。更多还原

【Abstract】 With the continuous expansion of the highway network and the rapid growth of traffic volume,the deterioration of pavement conditions and the increasing traffic safety risks have become more prominent,presenting new challenges for pavement maintenance decision-making.Furthermore,the growing number of traffic accidents has introduced new requirements for pavement maintenance.For instance,on road segments with high accident incidence rates,it is imperative to promptly repair potholes and cracks to prevent secondary accidents caused by pavement damage;at curved sections and slopes,it is often necessary to employ pavement materials with high friction coefficients or implement anti-skid treatments.However,traditional pavement maintenance decision-making has mainly relied on performance deterioration patterns and maintenance costs,with a noticeable lack of quantitative analysis of traffic safety factors.Additionally,there is a lack of research applying comprehensive traffic safety evaluation results to pavement maintenance decision-making.Moreover,the imbalance in the distribution of data with and without maintenance measures in actual engineering data limits the accuracy of the pavement maintenance decision-making model,thereby affecting the selection of maintenance action.To address these issues,this study focuses on highways in Shanxi Province and proposes a data-driven pavement maintenance decision-making framework that incorporates safety improvements.The following four parts of the study are mainly developed:Firstly,the datasets needed for this study were collected and organized.These datasets include basic road section information,traffic accident data from 2018-2024,pavement performance data from 2021-2024,maintenance action data from 2023,traffic volume data from 2023,and climate data.Data preprocessing was performed on the dataset to address format issues,data omissions,and anomalies.A statistical analysis of highway traffic accident data,pavement performance data,and maintenance action data was conducted,revealing that current highway traffic safety risks are relatively high,the pavement is severely damaged,and the distribution of maintenance measures in the engineering data is imbalanced.Case studies also indicated that maintenance actions have a certain degree of improvement in traffic safety.Secondly,a three-level indicator system for the comprehensive traffic safety evaluation based on historical accident data was constructed.This system includes two primary indicators—accident rate and accident severity—and ten secondary indicators.A combination of the Entropy Weight Method(EWM)and the Analytic Hierarchy Process(AHP)was used to assign weights to each indicator,while the Technique for Order Preference by Similarity to Ideal Solution(TOPSIS)method was applied to calculate the comprehensive traffic safety evaluation result.Based on this system,the Traffic Safety Risk Index(TSRI)for each road section and pile number along the highway was quantified,providing a solid foundation for subsequent predictions and pavement maintenance decision-making.Then,a Traffic Safety Risk Index(TSRI)prediction model based on CTGAN-RF was then developed.To address pavement maintenance decision-making that incorporates safety improvements,maintenance measures were included as one of the variables in the traffic safety risk index prediction model.To address the issue of imbalanced maintenance measures in the current data,Conditional Tabular Generative Adversarial Network(CTGAN)and Copula methods were employed to augment the traffic safety feature dataset.A comparative analysis of the structure and correlations between the generated data and the original data was conducted,and the CTGAN model with the most effective enhancement was used to generate the augmented dataset.Using the augmented dataset,Random Forest(RF),Gradient Boosting Decision Tree(GBDT),and e Xtreme Gradient Boosting(XGBoost)prediction models were constructed.The predictive performance of the three models was compared,and the RF model,which showed the highest accuracy,was selected as the prediction model for TSRI in the next section.Finally,a multi-objective pavement maintenance decision model was developed,considering both pavement and safety performance improvement.The objectives of maximizing maintenance benefits,minimizing maintenance costs,and maximizing safety benefits were included,along with three penalty functions for pavement performance,maintenance costs,and safety performance.A multi-objective genetic algorithm(Nondominated Sorting Genetic Algorithm II,NSGA-Ⅱ)was used to construct the decision model.This model was compared with a dual-objective decision model.which only considered maintenance benefits and costs,as well as with actual maintenance measures,to validate that it not only improves pavement performance and controls maintenance costs but also enhances traffic safety performance,providing a reference for refining maintenance decision models.In summary,this study proposes a data-driven pavement maintenance decision-making method that incorporates safety improvements.The method addresses several shortcomings in current research and achieves a collaborative optimization of traffic safety risk reduction,pavement performance enhancement,and maintenance cost control.This study provides a unified framework for considering traffic safety risks in pavement maintenance decision-making and offers new insights and a valuable reference for highway maintenance management departments in formulating decision-making plans.更多还原