Abstract:

Passenger flow prediction is an important part of urban intelligent transportation system, which has significant guiding meaning for people’s travel and traffic control. In view of the spatial dimension and temporal dimension attributes of subway passenger flow data, this paper proposes a prediction model that can capture the spatial and temporal characteristics of the data simultaneously. The model is based on the encoder-decoder architecture design, in which both decoder and encoder are composed of spatio-temporal prediction module. In this module, graph convolution is used to learn the station’s spatial structure and the gating cycle unit to capture temporal characteristics. In addition, the model uses the dual time series composed of entrance and exit passenger flow in unit time as input to cooperatively predict the entrance and exit passenger flow of each station. The experimental on the data collected from Shanghai metro cards show that the proposed model achieves a better result in the entrance and exit passenger flow prediction, which indicate taking the spatial dependence into consideration can improve the model prediction accuracy.

Key words: passenger flow prediction, spatio-temporal data, encoder-decoder, graph convolutional network, gated cycle unit

摘要：

客流量预测是城市智能交通系统的重要组成部分，对人们出行和交通管控有着重要的指导意义。针对地铁客流量数据具有时间维度和空间维度属性的特点，提出一种可以同时捕获数据时空特征的预测模型。该模型基于编码器解码器架构设计，其中解码器和编码器均由时空预测模块组成，在该模块中利用图卷积学习地铁站的空间拓朴结构、门控循环单元来捕获数据的时间特征。此外，模型将单位时间间隔内进站和出站客流量分别构成的两个时间序列，即进出站双时间序列作为输入，最终协同预测各站点的进站与出站人数。在上海地铁一卡通数据集上进行对比实验，实验结果表明，所提出的模型在进站与出站客流量预测上均取得了更好的效果，这表明考虑空间依赖能够有效地提高模型预测精度。

关键词: 客流量预测, 时空数据, 编码器解码器, 图卷积网络, 门控循环单元