【作者】 赵君豪； 李志恒； 于海洋； 刘清祥； 方洁； 张锦旺；

【Author】 Zhao Junhao;Li Zhiheng;Yu Haiyang;Liu Qingxiang;Fang Jie;Zhang Jinwang;Tsinghua-Berkeley Shenzhen Institution;Ministry of Logistics and Transportation,Shenzhen Graduate of Tsinghua University;Operation Department,Shenzhen Eastern Public Transport Co.,Ltd.;Transportation Research Department,Shenzhen Yihang Network Traffic Technology Co.,Ltd.;

【机构】 清华大学清华-伯克利深圳学院； 清华大学深圳研究生院物流与交通学部； 深圳市东部公共交通有限公司营运部； 深圳市易行网交通科技有限公司交通研究部；

【摘要】 公交串车是一种不利的交通现象,会导致运力浪费、线网承载能力不足、乘客等待时间增加等状况发生。公交串车的发生与公交调度相关,通过对车辆串车的监测与预测是预防公交串车的有效手段。本文在对影响串车的因素进行了分析的基础上,基于遗传算法和LS-SVM建立公交串车预测模型,模型计算量少、预测精度高同时有一定泛化能力;以深圳市的公交IC卡数据为基础,进行算例分析,提取出模型所需的以公交班次为单位的行驶数据;基于行驶数据获取车头时距、上下车人数、站间行驶时间等特征参数;通过相关性分析确定上游车头时距、上下车人数、站间行驶时间作为下游站点车头时距预测模型的输入变量,并输入模型进行训练和测试;将模型预测结果与KNN、神经网络、随机森林的结果进行比较,结果表明本文模型在算例上预测精度最高,效果最优。更多还原

【Abstract】 Bus-bunching is a harmful traffic phenomenon, which can lead to the waste of transportation capacity, lack of capacity of the road network, and the increase of passenger waiting time. The occurrence of bus-bunching is related to the dispatching of buses. The monitoring and forecasting of bus-bunching is an effective means to prevent it. After analysing the factors influencing bus-bunching, this paper establishes the bus-bunching prediction model based on genetic algorithm and LS-SVM. The model has less calculation, high prediction accuracy and some generalization ability. Based on the data of Shenzhen IC card, an example analysis was conducted to extract the model’s required travel data. Based on the extracted data, the parameters such as headway, number of passengers getting on and off the bus, and travel time between stations were extracted further; Correlation analysis is used to determine the former parameters as the input variables of the prediction model for model’s training and testing; Carry out the model prediction testing results with KNN, neural network, and random forest results. By comparison, the results show that this model has the highest prediction accuracy and the best effect.更多还原