基于多源数据与粒子滤波的城轨断面客流预测

Section Passenger Flow Prediction of Urban Rail Transit Based on Multi-source Data and Particle Filter

为提高运营组织服务水平,合理、准确地预测城市轨道交通断面客流十分重要。针对单一数据源限制客流预测准确度的问题,提出一种基于车载称重数据和自动售检票系统(AFC)数据的粒子滤波(PF)断面客流预测模型。以成都地铁9号线的历史运营数据为基础,分析不同性质客流与断面客流的相关性以筛选模型输入特征,并对所提出方法进行验证。实验结果表明:相较于单源PF模型,多源PF对节假日(工作日)早高峰、平峰以及晚高峰30min粒度客流预测的平均绝对误差(MAE)分别下降了28.541%(60.969%)、10.687%(19.662%)和22.685%(27.941%)。对比多源卡尔曼滤波(KF)和多源长短时记忆(LSTM)模型,多源PF对节假日和工作日客流预测可决系数(R2)的提升至少为21.599%和0.314%,该模型具有较快的计算速度,可为城市轨道交通客流预测提供参考建议。

In order to improve the service level of operating organizations,it is very important to reasonably and accurately predict the section passenger flow of urban rail transit.Aiming at the problem that a single data source limits the accuracy of passenger flow forecasting,a particle filter(PF)section passenger flow prediction model based on vehicle weighing data and automatic fare collection system(AFC)data is proposed.Based on the historical operation data of Chengdu Metro Line 9,the correlation between passenger flow of different natures and section passenger flow is analyzed to filter the input features of the model,and the proposed method is verified.The experimental results show that:compared with the single-source PF,the mean absolute error(MAE)of the multi-source PF for the 30 min granularity passenger flow prediction of the morning peak,flat peak and evening peak on holidays(working days)decreased by 28.541%(60.969%)and 10.687%(19.662%)and 22.685%(27.941%)respectively.Compared with the multi-source Kalman filter(KF)and multi-source long short term memory(LSTM)models,the multi-source PF improves the coefficient of determination(R2)for passenger flow forecasting on holidays and working days by at least 21.599%and 0.314%.The multi-source PF model has a relatively fast calculation speed and can provide reference for passenger flow prediction of urban rail transit.