摘要
采用基于风险分析的参数估计方法,利用嘉兴市辖区高速公路上1 062个交通事件持续时间数据,建立交通事件持续时间的威布尔加速失效时间预测模型.并用265个交通事件数据对模型的预测精度进行检验.检验结果表明,交通事故持续时间预测误差小于30min的准确率为78%,而只考虑持续时间大于10min的抛锚事件时,其持续时间预测误差小于10min的准确率为70%.基于风险分析的方法比仅基于贝叶斯或仅基于决策树算法的预测精度更高,数据利用范围更大.
This paper presents a model for incident duration prediction based on accurately recorded and large incident dataset from the Jiaxing city＇ s freeway incident response center system. To develop the duration prediction model, this study utilizes the Weibull accelerated failure time （AFT） model and a 3-year incident duration dataset from 2006 to 2008. Specifically, the 1062 incident dataset is utilized to develop the prediction model and then, the 265 dataset are employed to test the accuracy of prediction results of the 1062 model. Although the duration prediction model has limitations such as large prediction error due to the individual differences of the incident treatment teams in terms of clearing similar incidents, the results show that the prediction accuracy of anchor duration whose duration are more than 10 min is 70% with an error of 10 min while that of accident duration is 78% with an error of 30 min. So the hazard-based method is more extensive for dataset utilization and accurate than the methodbased on Bayesian method or decision tree respectively.
出处
《同济大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第2期241-245,共5页
Journal of Tongji University:Natural Science
