基于随机森林的城市快速路交通事件持续时间估计∗

摘要

Taking the complexity of influence factors of traffic incident duration on urban expressways into consider-ation,a duration estimation model based on random forest is proposed and developed in this paper.It attempts to provide a base for a reasonable estimation of the delays resulting from traffic incidents on urban expressways.The sample data is obtained from Shanghai Urban Expressway Monitoring Center.After data is preprocessed,the distribution of traffic inci-dent duration is analyzed.The minimum error of decision trees has been calculated through OOB data error estimation.An estimation model of incident duration is then developed.This model is then verified and its overall accuracy is found to meet the requirements as a reasonable accuracy with an average error of 6 min is achieved.It is also found that its accuracy is higher for duration within the range of 7~30 min,.The model is found to be more accurate and robust than decision-tree method.%针对城市快速路交通事件持续时间影响因素的复杂性,为合理估计交通事件造成的影响提供依据,提出基于随机森林的快速路持续时间估计模型。利用上海市快速路监控中心获得的样本数据,经过数据预处理后,分析交通事件持续时间的分布特征。通过袋外数据估计得到模型最小误差；同时自动判定各属性的重要程度,即事件持续时间的显著影响因素。建立交通事件持续时间估计模型,最后对模型进行验证,模型估计精度总体达到要求,估计平均误差为6 min,在持续时间7~30 min内精度较高,优于仅基于决策树的方法,且鲁棒性更强。