利用梯度提升树预测公交车到站时间

A Bus Arrival Time Prediction Method Based on GBDT

当前我国公交公司普遍采用让具有丰富经验的公交调度人员以人工估计车辆到站的方法来调度车辆的发车。这种方式缺少计算辅助,加上工作量大,经常容易出现错误预估导致无法缓解道路上常发生的同路公交车遇到一起(串车)或者相隔太远(大间隔)的情况。公交到站时间受道路交通、乘客人数、时间、天气等诸多因素影响,具有不确定性。本文基于该现实问题从公交公司角度出发,提出了一种基于动态特征选择和梯度提升树的公交到站时间预测算法。其动态主要体现在对于不同线路、同一线路不同方向经过特征选择分别选取对该线路该方向站点停留和站间行驶影响较大的特征。该算法用于辅助公交调度人员参考到站时间,从而使得调度人员可以作出更准确有效的调度策略。

At present, the bus dispatchers with rich experience generally use the method of estimating the arrival of vehicles manually to dispatch the departure of vehicles. However, this method lacks computational assistance, and with a large workload, it is often prone to erroneous estimates, which still cannot alleviate the common occurrence of the same-way bus on the road. The situation of meeting together or being too far apart. However, bus arrival time is influenced by many factors, such as road traffic, passenger number, time, weather and so on. It is uncertain. Based on this practical problem and from the perspective of bus companies, this paper proposes a bus arrival time prediction algorithm based on dynamic feature selection and GBDT. Its dynamics are mainly embodied in the characteristics of different lines and different directions of the same line, which have great influence on the stopover and inter-station driving of the stations in this direction. This algorithm is used to assist bus dispatchers to refer to arrival time, and can make more accurate and effective dispatching strategy.