College of Traffic Engineering and Key Laboratory of Road & Traffic Engineering of the Ministry of Education, Tongji University, 4800 Cao'an Road, Shanghai 201804, China.
Department of Civil & Environmental Engineering, Colorado State University, Fort Collins, CO 80523, United States.
J Safety Res. 2018 Jun;65:153-159. doi: 10.1016/j.jsr.2018.02.010. Epub 2018 Apr 25.
Driving environment, including road surface conditions and traffic states, often changes over time and influences crash probability considerably. It becomes stretched for traditional crash frequency models developed in large temporal scales to capture the time-varying characteristics of these factors, which may cause substantial loss of critical driving environmental information on crash prediction.
Crash prediction models with refined temporal data (hourly records) are developed to characterize the time-varying nature of these contributing factors. Unbalanced panel data mixed logit models are developed to analyze hourly crash likelihood of highway segments. The refined temporal driving environmental data, including road surface and traffic condition, obtained from the Road Weather Information System (RWIS), are incorporated into the models.
Model estimation results indicate that the traffic speed, traffic volume, curvature and chemically wet road surface indicator are better modeled as random parameters. The estimation results of the mixed logit models based on unbalanced panel data show that there are a number of factors related to crash likelihood on I-25. Specifically, weekend indicator, November indicator, low speed limit and long remaining service life of rutting indicator are found to increase crash likelihood, while 5-am indicator and number of merging ramps per lane per mile are found to decrease crash likelihood.
The study underscores and confirms the unique and significant impacts on crash imposed by the real-time weather, road surface, and traffic conditions. With the unbalanced panel data structure, the rich information from real-time driving environmental big data can be well incorporated.
驾驶环境（包括路面状况和交通状况）随时间推移而变化，对事故概率有很大影响。对于在大时间尺度上开发的传统事故频率模型来说，要捕捉这些因素的时变特征是很困难的，这可能导致在事故预测中丢失关键的驾驶环境信息。
开发了具有精细时间数据（每小时记录）的事故预测模型，以描述这些因素的时变特性。开发了不平衡面板数据混合 Logit 模型来分析高速公路路段每小时的事故可能性。从道路天气信息系统（RWIS）获得的精细时间驾驶环境数据，包括路面和交通状况，被纳入模型中。
模型估计结果表明，交通速度、交通量、曲率和化学湿路面指标较好地建模为随机参数。基于不平衡面板数据的混合 Logit 模型的估计结果表明，I-25 上存在许多与事故可能性相关的因素。具体来说，周末指标、11 月指标、低限速和车辙剩余使用寿命长的指标被发现会增加事故可能性，而 5 点指标和每英里每车道合并匝道数量的指标被发现会降低事故可能性。
该研究强调并证实了实时天气、路面和交通条件对事故的独特而重大的影响。通过不平衡面板数据结构，可以很好地整合实时驾驶环境大数据中的丰富信息。
