Liu Xiang, Liu Chang, Hong Yili
Department of Civil and Environmental Engineering, Rutgers University, United States.
Department of Statistics, Rutgers University, United States.
Accid Anal Prev. 2017 Oct;107:164-172. doi: 10.1016/j.aap.2017.07.007.
There are annually over two million carloads of hazardous materials transported by rail in the United States. The American railroads use large blocks of tank cars to transport petroleum crude oil and other flammable liquids from production to consumption sites. Being different from roadway transport of hazardous materials, a train accident can potentially result in the derailment and release of multiple tank cars, which may result in significant consequences. The prior literature predominantly assumes that the occurrence of multiple tank car releases in a train accident is a series of independent Bernoulli processes, and thus uses the binomial distribution to estimate the total number of tank car releases given the number of tank cars derailing or damaged. This paper shows that the traditional binomial model can incorrectly estimate multiple tank car release probability by magnitudes in certain circumstances, thereby significantly affecting railroad safety and risk analysis. To bridge this knowledge gap, this paper proposes a novel, alternative Correlated Binomial (CB) model that accounts for the possible correlations of multiple tank car releases in the same train. We test three distinct correlation structures in the CB model, and find that they all outperform the conventional binomial model based on empirical tank car accident data. The analysis shows that considering tank car release correlations would result in a significantly improved fit of the empirical data than otherwise. Consequently, it is prudent to consider alternative modeling techniques when analyzing the probability of multiple tank car releases in railroad accidents.
在美国，每年通过铁路运输的危险材料超过200万车皮。美国铁路公司使用大量的罐车将原油和其他易燃液体从生产地运输到消费地。与危险材料的公路运输不同，火车事故可能会导致多辆罐车脱轨和泄漏，这可能会造成严重后果。先前的文献主要假设火车事故中多辆罐车泄漏的发生是一系列独立的伯努利过程，因此在已知脱轨或受损罐车数量的情况下，使用二项分布来估计罐车泄漏的总数。本文表明，在某些情况下，传统的二项式模型可能会错误地高估多辆罐车泄漏的概率，从而显著影响铁路安全和风险分析。为了弥补这一知识空白，本文提出了一种新颖的、替代性的相关二项式（CB）模型，该模型考虑了同一列车中多辆罐车泄漏的可能相关性。我们在CB模型中测试了三种不同的相关结构，并发现基于罐车事故实证数据，它们都优于传统的二项式模型。分析表明，考虑罐车泄漏的相关性将比不考虑时显著改善对实证数据的拟合。因此，在分析铁路事故中多辆罐车泄漏的概率时，谨慎考虑替代性建模技术是明智的。
