基于贝叶斯时空建模的高速公路事故黑点判别

Identifcation of Freeway Crash Hotspots Based on Bayesian Space-time Modeling

为了识别高速公路事故黑点,基于历史交通事故数据,建立贝叶斯时空交互模型,估计高速公路路段事故率和超常事故率。根据其后验期望序号对路段安全性进行排序,将排序靠前的一定比例路段判定为事故黑点。利用该方法对广东开阳高速公路进行事故黑点判别,并与基于贝叶斯层级泊松模型的黑点判别结果进行对比。结果表明,时空交互模型和层级泊松模型的事故路段排序结果存在显著差异。以事故率为安全评价指标时,2个方法判别的事故黑点中有73%相同;以超常事故率为安全评价指标时,2个方法判别的事故黑点中仅有20%相同。这与类似研究的结论一致,体现了解析时空关联和交互对事故黑点判别的重要性。另外,还对比了基于评价指标后验期望序号和后验均值的事故路段排序序号。结果显示二者的一致性较高。

To identify freeway crash hotspots,based on historical traffic crash data,a Bayesian space-time interaction model is developed to estimate crash rates and excess crash rates,of which the posterior expected ranks are used to rank roadway segments.A certain proportion of the top ranked roadway segments are identified as crash hotspots.The approach is applied to identifying crash hotspots at Kaiyang Freeway in Guangdong.The results are compared with those identified by an approach based on Bayesian hierarchical Poisson model.The comparison results show that there are significant differences in the site ranking results generated by space-time interaction model and hierarchical Poisson model.When crash rate is selected as a decision parameter,73%of the crash hotspots identified by the two approaches are the same;while excess crash rate is selected as the decision parameter,only 20%of the crash hotspots identified by the two approaches are the same.The results conform to the findings of similar studies,which imply the importance of accounting for space-time correlation and interaction in identification of crash hotspots.Besides,the site rank orders under the posterior expected rank and posterior mean of the decision parameters are also compared.The results show that they are generally consistent.