多因素异质性作用下高速公路交通事故风险预测及防控研究

Traffic crashes lead to enormous deaths, injuries and property damages every year. The safety of freeway is particularly important as freeway is the backbone of highway transportation system. Predicting crash risk precisely so as to grasp the corresponding influence mechanism accurately is the premise to improve the safety of freeway, while the heterogeneity of effects of factors on traffic crash affects the precision and reliability of the prediction model significantly. Past studies on crash risk prediction methods are insufficient in accounting for heterogeneity. In this study, prediction methods for freeway crash risk and the corresponding influence mechanisms are studied from the perspective of fully tracking heterogeneity. Firstly, sources of heterogeneity and the effect of heterogeneity on reliability of prediction model are systematically analyzed. Secondly, Markov switching and spatial autoregression random parameters model are developed to address the heterogeneity from temporal and spatial points. Then, in order to account for two kinds of correlations between heterogeneity, correlated random parameters model and random parameters model with heterogeneity in means and variances are established, and the model system that can simultaneously account for various kinds of heterogeneity is built. Finally, the influence mechanism of freeway crash risk is revealed from aspects of road, traffic, environment, etc., and strategies integrating static prevention and dynamic control are proposed to prevent and control the risk of crash. Expected results of this study would have important scientific significance and practical value in improving the predictive theory for crash risk and developing freeway safety countermeasures.

交通事故每年造成大量人员伤亡和财产损失，高速公路作为公路运输系统骨架，其安全性尤为重要。对事故风险进行预测，准确把握其影响机理是提升高速公路安全性的前提，而各因素对交通事故作用的异质性对预测模型精度和可靠性影响巨大，以往事故风险预测研究在刻画异质性方面存在不足。本项目以系统解析异质性为切入点，对高速公路事故风险预测方法和影响机理进行研究。首先，系统分析异质性根源及其对预测可靠性的影响；其次，构建马尔科夫转换和空间自回归随机参数模型，从时间和空间角度刻画异质性；再次，针对异质性间两类相关性，构建相关随机参数模型和异质均值/方差随机参数模型，建立同时解析多种异质性的事故风险预测模型体系；最后，从道路交通及环境等多因素异质性作用角度揭示高速公路交通事故风险影响机理，提出静态预防和动态控制于一体的风险防控策略。预期成果对完善事故风险预测理论和制定高速公路安全对策具有重要科学意义和实用价值。

道路交通事故预防领域，各因素对事故风险影响的异质性对模型精度和可靠性影响巨大。本项目以我国高速公路为研究对象，提出能充分刻画各类异质性的事故风险建模方法，据此深入分析事故影响机理。首先，收集了国内多个省份的高速公路道路、交通及事故等数据，构建事故次数和事故严重程度数据库；然后，从事故因素多样性和事故产生的复杂性等方面定性分析异质性产生根源，并基于随机参数建模方法从理论角度整体把握异质性产生原因；其次，针对事故次数和事故严重程度分别在负二项及多项Logit模型框架下构建了基础随机参数模型、异均值随机参数模型、异均值/方差随机参数模型、相关随机参数模型及异均值相关随机参数模型，形成了能刻画各类异质性的事故分析模型体系；再次，针对上述各类复杂随机参数模型，提出了基于蒙特卡罗模拟的事故风险预测和影响因素分析方法；最后，从定性和定量角度分析了各因素对高速公路交通安全的影响。结果表明：①道路设计、交通运行及环境、驾驶员及车辆特性等方面指标均对事故的影响表现出显著异质性，且各因素的异质性影响间还存在很强的交互作用和时变特性；②相比传统的固定参数建模方法，构建的随机参数系列模型可将事故次数和严重程度预测精度分别提高约5%和20%；③提出的基于蒙特卡罗模拟的分析方法可将绝大多数事故因素的边际效应偏差控制在10%以内；④提出的随机参数系列模型不仅能提高事故风险分析和预测的准确性，还能从中得到各因素对交通事故的复杂交互影响及其他更多有价值的推论。研究结果为改进和优化现有事故风险分析方法提供了必备的理论依据，同时也为保障安全改善措施的合理性和有效性提供了必要的理论支撑。

内容获取失败，请点击重试