基于完全贝叶斯方法的道路交叉口交通安全综合评价建模与分析

In road safety analysis, the development of quantitative models for crash occurrence and crash severity prediction that consider the influences of the appropriate explanatory factors is essential for crash analysis and prevention, which underpins the implementation of road safety strategies. Road junctions play an important role in roadway networks. The safety of road junction is highly associated with the safety and operational performance of entire road network. A number of studies have addressed safety analysis for road junctions, most of which, however, focus on signal junctions, especially for the studies in China. As the disaggregated turning movements and traffic speed are not available in most situations, the safety analysis for road junctions is usually conducted in an aggregated framework. This research attempts to develop a joint probability model to integrate the predictions of both crash occurrence and crash severity into a single framework. In concern of the different characteristics for different kinds of road junctions, such as signal junction, roundabout, yield junctions and ramps, individual prediction models are developed to evaluate the effects of various explanatory variables (including disaggregated turning movements, approaching speed, road design, traffic control and management, weather conditions and other possible factors) on safety performance for each kind of junctions. The Markov chain Monte Carlo approach (MCMC) full Bayesian method, which provides a flexible framework to handle the proposed model with sophisticated structure, is applied for model analysis and inference in this research. The safety performance function is accordingly obtained and used for safety prediction and evaluation. The proposed research emphasizes on the issues of disaggregated traffic data collection and processing, integrated safety prediction model development, and safety analysis and inference. This research should be useful in the formulation and design of effective road safety management measures, and should be of great social and economic benefit to society.

交通安全评价旨在通过分析事故影响因素及伤害机理进行事故预防，对改变我国严峻交通安全现状必要而迫切。交叉口的安全水平直接关系到整个路网的安全性能及运行效率。国内外特别是我国对交叉口安全的研究主要集中于信控交叉口，多侧重事故风险研究，仍多采用传统宏观评价方法，而且缺乏对实时转向交通流特性影响的考虑。本申请将针对信控、环形、让行交叉口与匝道出入口等各类交叉口的特点，研究提出事故风险与伤亡程度综合预测模型，以实时转向流量为暴露量，将交通特性及其他影响因素（道路设计、管理控制、环境因素等）集成到统一评价体系中进行评估，并选用更为灵活的基于MCMC的完全贝叶斯求解方法，全面探究不同类型交叉口的事故特征及显著影响因素，为交叉口设计与管理提供依据。研究将重点解决实时交通流数据的采集与信息融合、综合安全评价模型构建求解、评价分析与对策研究等关键科学问题，完善交通安全评价研究，为改善交通安全提供理论支持与实践

交通安全关系到人类的健康与发展，近年来得到了更加广泛深入的关注。道路交叉口是路网中重要的结点和枢纽，交叉口的安全性直接关系到路网整体的安全性能与通行效率，本项目以道路交叉口交通安全分析与评价为核心，根据既定研究内容，开展了交叉口安全信息集成与综合数据库的建立、交通安全综合预测模型建立与求解、道路交叉口交通安全综合评价与分析、实时交通安全评价系统构建与基础平台等四大部分研究。.本项目研究成果在数据挖掘、方法论以及实证研究等方面都取得了进展，是对现有研究成果的重要补充，具有较强的学术价值与实践指导意义。数据层面，创新性地研究了实时转向流量对交叉口各类交通事故安全水平的影响，为交叉口精细化设计与管理提供了理论依据。方法层面，提出了配比重采样预测模型，提高了零堆积数据集的预测结果；提出了一种基于连通性的空间权重矩阵，建立了考虑空间相关性的安全评价模型；探索了各类改进神经网络模型在交通安全评价中的应用。实证研究层面，从信号灯交叉口到环形交叉口，从机动车事故到电动自行车事故，从宏观安全到个性化驾驶安全，从事故因素分析到事故对交叉口的通行能力影响分析，从局部交叉口到路网安全水平，都进行了深入研究，揭示了交叉口安全研究的内在机理与外延影响。.基于以上研究成果，本项目借鉴交通拥堵指数的概念，创新性地提出了交通安全指数及其计算方法，并运用GIS技术搭建了可视化交通安全指数发布平台，将实时交通状态、道路线型、交通管理参数、环境因素等信息与相应道路及交叉口关联存储，用以实时评估与发布路网安全水平，为交通安全管理与服务提供了重要的支撑平台。.综上，本项目完成了既定研究任务，达到了预期目标，共计发表学术论文15篇，其中SCI/SSCI收录10篇，EI收录4篇；申请并公开发明专利8项；协助培养研究生8名；研究成果交通安全指数平台已获地方政府立项支持，成果具有广泛的学术价值与应用前景，项目组将在后续工作中对研究成果进一步深化与应用。.