Abstract

Real-time traffic movements at intersections is vital for transportation and traffic engineering. It helps in providing intersection traffic data and optimizing signal control plans. This study attempts to extend the data coverage by developing algorithms to estimate through and left-turn movements in real-time at signalized intersections. This study is the first attempt to estimate short-term traffic movement counts at signalized intersections at the cycle-level for signal control. Real-time data were collected from 19 intersections along two main corridors in Orange county, Florida. A framework was proposed to identify different variables considering the characteristics of the cycle-level traffic movement estimation. To develop generic algorithms, a proposed approach was utilized to generalize the traffic movement estimation at the corridor level. Signal timing and movement counts at the upstream and downstream intersections were utilized to estimate through and left-turn movements at target intersections. An extensive comparison study was carried out based on the processed data. The experimental results show that the proposed Gradient Boosting model outperformed the baseline models with Mean Absolute Percentage Error 9.53% and 4.6% for the through and left-turn movements, respectively. In addition, the transferability of the developed models for the abnormal traffic conditions was validated and the estimated model proved that it could instantaneously capture the traffic change due to an accident. It is expected that the proposed method could reduce the sensor cost and extend the movement data coverage, which could help in developing efficient signal control plans at signalized intersections.

摘要

十字路口的实时交通运动对于交通和交通工程至关重要。它有助于提供交叉口交通数据和优化信号控制计划。本研究试图通过开发算法来实时估计信号交叉口的通过和左转运动，从而扩展数据覆盖范围。本研究首次尝试在信号控制的周期级估计信号交叉口的短期交通流量计数。从佛罗里达州奥兰治县两条主要走廊的 19 个十字路口收集实时数据。考虑到循环级交通运动估计的特征，提出了一个框架来识别不同的变量。开发通用算法，一种提议的方法被用来概括走廊层面的交通运动估计。上游和下游交叉口的信号计时和移动计数用于估计目标交叉口的通过和左转移动。根据处理后的数据进行了广泛的比较研究。实验结果表明，所提出的梯度提升模型在直行和左转运动中的平均绝对百分比误差分别为 9.53% 和 4.6%，优于基线模型。此外，验证了所开发模型对异常交通状况的可迁移性，估计模型证明了它可以瞬时捕捉事故引起的交通变化。