Research Of Safety Analysis Models Based On Extreme Statistics Of Traffic Conflicts

Road safety problem is always a global problem which influences the sustainable development, and road crashes cause lots of casualties and vast property loss every year. Up to now, the majority of research and engineering efforts that been made are based on the crash statistics. However, because of the randomness and rarity of crash occurrence, the incompleteness of crash records, as well as the aggregation and the passivity of crash statistic analysis, researchers and engineers have also made great efforts to explore surrogate measures of safety which are not based on crash data. Traffic conflict technique is one of the surrogate measures which receives many research interests but is still not so complete in theoretical system and application methods, thus it has large space for development and broad prospects for application.By combining the traffic conflict technique and the extreme value theory, the safety analysis models based on extreme statistics of traffic conflicts were established, investigated, and further applied.Literatures related to traffic conflict techniques and extreme value theory at home and abroad were reviewed at first, and the focused analysis was on traffic conflict techniques. Based on the literature review, the main achievements in traffic conflict definition, traffic conflict data collection, and predictive validity of traffic conflicts were summarized, the advantages and disadvantages of current traffic conflict techniques were concluded, and future research directions were put forward.Based on video and crash data collected from Yuebei segment of Jinzhu freeway, Kaiyang freeway, and Yuegan freeway in Guangdong Province, the traffic volume characteristics and lane change maneuvers were analyzed, the way to measure the post encroachment time during each lane change maneuver was proposed, and the relationship between lane change conflicts and crashes was also briefly analyzed.Considering that both the traffic conflict technique and extreme value theory have a common ground, which is to extrapolate the occurrence of extreme events from the distribution of more observable events, a safety analysis method which combined them was developed, and the traffic conflict based block maxima model and peak over threshold model were established. Two safety analysis indicators, estimated number of crashes and crash return level, were developed based on the model parameter calibration results. Moreover, based on the estimated model parameters and the calculated safety analysis indicators, comparisons were conducted among block maxima model, peak over threshold model, and crash-based negative binomial regression model.To eliminate the subjectivity and uncertainty of threshold selection in peak over threshold model, the shifted Gamma-GPD (Generalized Pareto Distribution) model was established, and the method to calibrate the model parameters was introduced. Furthermore, the underlying safety information in the model was discussed.Both the block maxima method and peak over threshold method were used to modeling extreme traffic conflicts, and it is the first time that a comprehensive comparison between the two methods applied to road safety analysis was conducted. The results showed that, when the observation time period was short or the number of samples was limited, the peak over threshold method was better than the block maxima method from the aspects of data utilization, estimate accuracy and estimate reliability. Based on the concept of “return level” in extreme value theory, a safety analysis indicator called “crash return level” was firstly developed, and the superiority of this indicator over the estimated crashes was demonstrated by comparing the peak over threshold model and the crash-based negative binomial regression model. The usefulness of crash return level also showed the validity of traffic conflict techniques. To eliminate the subjectivity and to reduce the uncertainty in peak over threshold model, the shifted Gamma-GPD model was established. This model is the first parametric model representing the severity hierarchy model of safety continuum, and it can promote the utilization of safety continuum towards to the quantitative direction.This study showed the feasibility and superiority of using extreme statistics for road safety analysis, and it also further demonstrated the validity of traffic conflict techniques. This would provide new ideas and methods for research on road safety analysis and evaluation. On the other hand, the safety analysis indictors can also be directly used in practical safety analysis projects, and the analysis results based on these indicators will provide engineers with more references.