基于深度集成神经网络的城市轨道交通短时进站客流预测

Short-term Inbound Passenger Flow Forecasting for Urban Rail Transit Based on Deep Ensemble Neural Network

准确、可靠的城市轨道交通短时客流预测是智慧地铁的重要组成部分。现有的短时客流预测模型大多是在常态条件下提出的,在异常条件下难以获得满意的预测精度。为此,提出一种基于深度集成神经网络(deep ensemble neural network,DENN)的短时进站客流预测模型。该模型建模并整合了天气、时间和特殊事件等外部环境因素,最近时段进站客流的时间依赖性,以及出站客流的相关性,具有高度的灵活性和可扩展性。具体地,在DENN中,首先,嵌入一个门控循环单元(gated recurrent unit,GRU)网络,用于提取最近时段进站客流数据的时间依赖性;其次,引入Transformer网络,用于自适应地捕获出站客流数据中对进站客流影响最大的时段,以提取出站客流的相关性;最后,应用全连接网络编码外部环境因素和实现特征融合及预测。在上海地铁徐泾东站和上海体育场站的数值实验表明,提出方法在普遍条件下都能取得较高的预测精度。

Accurate and reliable short-term passenger flow forecasting for urban rail transit is an important component of a smart metro.Most of the existing short-term passenger flow prediction models,proposed under normal conditions,can hardly obtain satisfactory prediction accuracy under abnormal conditions.In this paper,a model was proposed based on Deep Ensemble Neural Network(DENN)for short-term inbound passenger flow forecasting.The model models and integrates external environmental factors such as weather,time of day and special events,the time dependence of inbound passenger flows in the recent period,and the correlation of outbound passenger flows,with a high degree of flexibility and scalability.Specifically,in DENN,firstly,a Gated Recurrent Unit(GRU)network was embedded to extract the time-dependency of inbound passenger data for the most recent time period.Secondly,a Transformer network was introduced to adaptively capture the time period that has the greatest impact on inbound passenger flow to extract the correlation of outbound passenger flows.Finally,fully connected networks were applied to encode external environmental factors as well as to achieve feature fusion and prediction.In addition,numerical experiments at Xujing East Station and Shanghai Stadium Station of Shanghai Metro show that the proposed method can achieve high prediction accuracy under common conditions.