Abstract:

Merging behaviors in weaving sections heavily affect traffic operations and may trigger traffic congestions and breakdowns. Merging location is one of the most important merging behaviors. A new method called Gradient Boosting Decision Tree(GBDT) is presented in this paper to develop a merging location model. Because merging behaviors involve both longitudinal and lateral driving behaviors, initial lateral location is considered in this paper. Data are extracted from NGSIM dataset and used to train the model. Compared with Lognormal model using AIC, BIC and R2, the proposed GBDT model is better than Lognormal model. It is shown that later location is most important variable. The partial effects of exploratory variables indicate that GBDT can reveal the hidden nonlinear relationships between merging location and exploratory variables.

Key words: highway transportation, weaving section, GBDT, merging location, lateral location

摘要：

匝道车辆的汇入行为对高速公路交织区的通行能力有重要的影响，汇入位置是汇入行为中最重要的行为参数之一.本文利用梯度提升决策树(GBDT)建立了一个车辆汇入位置模型并对各变量进行了分析.考虑到汇入行为是一个二维驾驶行为，我们在模型中引入了车辆进入辅助车道时的初始横向位置这一变量.利用NGSIM中的车辆轨迹数据对模型进行训练，并与Lognormal 进行对比.结果表明，GBDT模型在AIC，BIC 和R²这3个指标上均大幅优于Lognormal模型.最后，本文对解释变量的重要性和其偏效应进行了分析，其中初始横向位置的重要性最高；敏感性分析表明，GBDT模型能够深度挖掘汇入位置与变量之间隐藏的非线性关系.

关键词: 公路运输, 交织区, 梯度提升决策树, 汇入位置, 初始横向位置