交通运输系统工程与信息 ›› 2019, Vol. 19 ›› Issue (2): 116-121.

• 系统工程理论与方法 • 上一篇    下一篇

基于LightGBM算法的公交行程时间预测

王芳杰*1,王福建 2,王雨晨 2,边驰 2   

  1. 1. 浙江国际海运职业技术学院 航海工程学院,浙江 舟山 316021;2. 浙江大学 建筑工程学院,杭州 310058
  • 收稿日期:2018-09-17 修回日期:2018-12-07 出版日期:2019-04-25 发布日期:2019-04-25
  • 作者简介:王芳杰(1985-),男,浙江舟山人,讲师.

Bus Travel Time Prediction Based on Light Gradient Boosting Machine Algorithm

WANG Fang-jie1, WANG Fu-jian2, WANG Yu-chen2, BIAN Chi2   

  1. 1. School of Marine Engineering, Zhejiang International Maritime College, Zhoushan 316021, Zhejiang, China; 2. College of Civil Engineering and Architecture, Zhejiang University, Hangzhou 310058, China
  • Received:2018-09-17 Revised:2018-12-07 Online:2019-04-25 Published:2019-04-25

摘要:

在城市公交网络运行中,公交车的站点间行程时间会受到道路和环境条件的影响. 本文对公交车运行过程中的车辆速度特征、道路特征及天气特征等进行了分析.建立了基于特征的 LightGBM (Light Gradient Boosting Machine)公交行程时间预测模型,通过调整 LightGBM算法中的相关参数,以分配各个影响特征和因素的权重大小.然后利用天津市某条公交线路 24天的公交车 GPS数据对模型进行了训练和验证,并与基于历史平均值和卡尔曼滤波的行程时间预测模型进行对比.比较结果表明,LightGBM模型在 MAE (Mean Absolute Error)和 MAPE (Mean Absolute Percentage Error)这两个指标上均大幅度优于其他两个模型,说明 LightGBM模型在公交车行程时间预测上具有很好的稳定性和应用前景.

关键词: 城市交通, 预测, LightGBM算法, 公交车行程时间, GPS数据

Abstract:

In the operation of urban public transport networks, the travel time between stations of a bus is affected by road and environmental conditions. This paper analyzes the bus speed characteristics, road characteristics and weather characteristics during bus operation, and a feature- based LightGBM bus travel time prediction model is established. By adjusting the relevant parameters in the LightGBM algorithm, the weights of the influencing features and factors are assigned. Then the model is trained and verified by using the 24-day bus GPS data of one bus line in Tianjin, and compared with the travel time prediction model based on historical mean and Kalman filter. The comparison results show that the LightGBM model is superior to the other two models in MAE (Mean Absolute Error) and MAPE (Mean Absolute Percentage Error), indicating that the LightGBM model has good stability and application prospects in bus travel time prediction.

Key words: urban traffic, prediction, LightGBM algorithm, bus travel time, GPS data

中图分类号: