NSF/USDOT: Modeling Matched Traffic and Accident Datasets to Significantly Improve Safety

运输系统信息技术的进步已经积累了大量关于运输系统状况的原始数据。为了充分利用这些数据，必须开发新的工具来组合和有效地分析大型数据库。本研究拟应用非线性多元分析方法，分析组合货车交通、详细的交通流、事故和环境数据，以识别不同交通条件下不同类型货车交通混合对事故发生可能性的影响，以及对不同类型网络链路的影响。了解卡车事故周围的复杂因素，可以为干预提供机会，以提高安全性。主要的分析方法是多组混合分类变量、顺序变量和数值变量的非线性典型相关分析。这种特征值方法通过交替最小二乘算法实现，其特点是非线性变量的最优缩放和结果的图形化解释。该项目分为四个不同的阶段：(1)建立一个适用于识别城市高速公路上卡车安全问题的交通流和碰撞信息的综合数据库；(2)通过特定类型的多元非线性模型，识别特定交通流条件下卡车交通混合对安全影响最大的高速公路位置和时间段；(3)确定在有问题的时空情况下提高安全性的方法。(4)确定将我们的研究应用于其他州可用数据的方法。总之，这项工作将开发工具来帮助识别不安全的交通状况，从而避免事故的发生。我们关注与卡车相关的事故，因为这些事故往往比只涉及乘客的事故更严重，而且卡车越来越多地配备了通信设备，司机可以很容易地得到警告，他们正在进入不安全的状况。拟议研究的更广泛影响包括改善交通安全，研究生研究人员的技术培训以及向人口严重不足的当地高中伸出援助之手。

Advances in information technologies for transportation systems have led to the accumulation of large quantities of raw data on transportation system status. To fully leverage such data, new tools must be developed to combine and effectively analyze large databases. The proposed research involves the application of nonlinear multivariate analysis methods to analyze combined truck traffic, detailed traffic flow, accident, and environmental data in order to identify the influences of mixes of truck traffic on the likelihood of accidents by type under different traffic conditions and on different types of network links. Understanding the complex factors surrounding truck accidents, can provide opportunities for intervention to enhance safety. The main analysis method is nonlinear canonical correlation analysis with multiple sets of mixed categorical, ordinal, and numerical variables. This eigenvalue method, implemented through alternating least squares algorithms, is characterized by the optimal scaling of the nonlinear variables and graphical interpretation of results. The project has four distinct phases: (1) establishing a comprehensive database of traffic flow and crash information that is appropriate for identifying truck safety issues on urban freeways (2) identifying, through a specific type of multivariate nonlinear model, freeway locations and time periods where the mix of truck traffic within particular traffic flow conditions has the most adverse safety effects, (3) identifying ways to improve safety in problematic time-space situations. (4) identifying ways to apply our research to data available in other states.In summary, the work will develop tools to help identify unsafe traffic conditions so that accidents can be avoided. We focus on truck involved accidents because these tend to be more severe than those involving only passengers and because trucks are increasingly equipped with communication devices and their drivers can be easily warned that they are entering unsafe conditions. The broader impacts of the proposed research include improvements in traffic safety, the technical training of graduate student researchers and outreach to local high schools with significant under-represented populations.