Date of Award

August 2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Engineering

First Advisor

Xiao Qin

Committee Members

Yue Liu, Yin Wang, Vytaras Brazauskas, Phoenix Do, Jun Zhang

Abstract

Crashes can be accurately predicted through reliable data sources and rigorous statistical models; and prevented through data-driven, evidence-based traffic control strategies. Both predictive analysis and analysis to estimate the causal effect of traffic variables of real-time crashes are instrumental to crash prediction and a better understanding of the mechanism of crash occurrence. However, the research on the second analysis type is very limited for real-time crash prediction; and the conventional predictive analysis using inductive loop detector data has accuracy issues related to inconsistently and distantly spaced loop detectors. The effectiveness of traffic control strategies for improving safety performance cannot be measured and compared without an appropriate traffic simulation application. This dissertation is an attempt to address these research gaps.

First, it conducts the propensity score based analysis to assess the causal effect of speed variation on crash occurrence using the crash data and ILD data. As a casual analysis method, the propensity score based model is applied to generate samples with similar covariate distributions in both high- and low-speed variation groups of all cases. Under this setting, the confounding effects are removed and the causal effect of speed variation can be obtained.

Second, it conducts a predictive analysis on lane-change related crashes using lane-specific traffic data collected from three ILD stations near a crash location. The real-time traffic data for the two lanes – the vehicle’s lane (subject lane) and the lane to which that a vehicle intends to change (target lane) – are more closely related with lane-change related crashes, as opposed to congregated traffic data for all lanes. It is found that lane-specific variables are appropriate to study the lane-change frequency and the resulting lane-change related crashes.

Third, it conducts a predictive analysis on real-time crashes using simulated traffic data. The purpose of using simulated traffic data rather than real data is to mitigate the temporal and spatial issues of detector data. The cell transmission model (CTM), a macroscopic simulation model, is employed to instrument the corridor with a uniform and close layout of virtual detector stations that measure traffic data when physical stations are not available. Traffic flow characteristics at the crash site are simulated by CTM 0-5 minutes prior to a crash. It shows that the simulated traffic data can improve the prediction performance by accounting for the spatial-tempo issue of ILD data.

Fourth, it presents a novel approach to modeling freeway crashes using lane-specific simulated traffic data. The new model can not only account for the spatial-tempo issues of detector data but also account for heterogeneous traffic conditions across lanes using a lane-specific cell transmission model (LSCTM). The LSCTM illustrates both discretionary lane-changing (DLC) and mandatory lane-changing (MLC) activities. This new approach presents a viable alternative for utilizing traffic simulation models for safety analysis and evaluation.

Last, it develops a crash prediction and prevention application (CPPA) based on simulated traffic data to detect crash-prone conditions and to help select the desirable traffic control strategies for crash prevention. The proposed application is tested in a case study with VSL strategies, and results show that the proposed crash prediction and prevention method could effectively detect crash-prone conditions and evaluate the safety and mobility impacts of various VSL alternatives before their deployment. In the future, the application will be more user-friendly and can provide both online traffic operations support as well as offline evaluation of various traffic control operations and methods.

Download

Share

COinS