• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>COTA international conference of transportation professionals >Gradient Boosting Regression Tree for Traffic Flow Prediction Considering Temporal and Spatial Correlations
【24h】

Gradient Boosting Regression Tree for Traffic Flow Prediction Considering Temporal and Spatial Correlations

机译：考虑时间和空间相关性的交通流量预测的梯度提升回归树

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Short-term traffic flow prediction is a vital component of transportation modeling and traffic management. In this paper, we employ the gradient boosting regression tree (GBRT) for short-term speed prediction of urban expressways considering temporal and spatial correlations of related traffic detectors. Relevant features are extracted from historical measurements of the target detector and upstream neighboring detectors for training and testing GBRT. The model combines additional trees by adjusting weights of its previous basic models and is thus capable of improving the speed prediction accuracy. Parameter tuning of GBRT is presented to show the influence of a variety of parameter combinations on the prediction performance. For illustrative purposes, fixed-location microwave detection data are extracted at the 5-min aggregation time interval from the Traffic Management Platform of Hangzhou, China, between June 21 and July 7, 2015. The first two-week data are utilized as the training dataset and the remaining three-day data are used as the test dataset. Several experiments are conducted to test the GBRT performance in short-term speed prediction. The results show that GBRT outperforms other two learning models, i.e., adaptive boosting (Adaboost) and random forests (RF), for 5-min, 15-min, and 30-min predictions. The consideration of temporal and spatial correlations improves the prediction accuracy for all the three models. Overall, the paper demonstrates that GBRT is a favorable, powerful, and flexible tool in short-term traffic speed prediction.
机译：短期交通流量预测是交通建模和交通管理的重要组成部分。在本文中，我们考虑相关交通探测器的时空相关性，将梯度增强回归树（GBRT）用于城市高速公路的短期速度预测。从目标探测器和上游相邻探测器的历史测量值中提取相关特征，以训练和测试GBRT。该模型通过调整其先前基本模型的权重来合并其他树，因此能够提高速度预测的准确性。提出了GBRT的参数调整，以显示各种参数组合对预测性能的影响。出于说明目的，在2015年6月21日至7月7日之间，从中国杭州市交通管理平台以5分钟的聚合时间间隔提取了固定位置的微波检测数据。前两周的数据用作培训数据集和剩余的三天数据用作测试数据集。进行了一些实验以测试GBRT在短期速度预测中的性能。结果表明，对于5分钟，15分钟和30分钟的预测，GBRT优于其他两个学习模型，即自适应增强（Adaboost）和随机森林（RF）。时间和空间相关性的考虑提高了所有三个模型的预测精度。总体而言，本文证明了GBRT是短期交通速度预测中的一种有利，强大且灵活的工具。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号