基于时空特征的公交站点短时客流量预测-《计算机技术与发展》

文章信息/Info

Title:: Short-term Passenger Flow Prediction of Bus Stops Based on Spatiotemporal Features

Author(s):: YAO Si-jia; GUI Zhi-ming; GUO Li-min; Faculty of Information,Beijing University of Technology,Beijing 100124,China

Keywords:: bus passenger flow prediction; spatiotemporal features; graph convolution network; long short - term memory network; attention mechanism

摘要:: 针对以往公交客流量预测只考虑时序特征而忽略空间维度特征的缺点,提出一种结合注意力机制的图卷积长短期记忆单元预测模型( AGLSTM) 来预测公交站点的客流量。该模型运用图卷积网络( GCN) 对每个时刻的公交站点客流量进行空间维度的特征提取,使用长短期记忆网络( LSTM) 对公交站点客流量进行时间特征的提取。为了更关注公交站点客流量有重大影响时刻的特征,该模型还引入了注意力机制模块。注意力机制可以通过计算不同时刻长短期记忆单元隐藏状态的权重,来评估各隐藏状态对输出结果的影响。通过对北京 4 条公交线路的真实刷卡数据进行实验分析,并与部分经典预测算法进行对比,证明了提出的考虑时空特征的组合模型能够有效地提高模型的预测精度。

Abstract:: Aiming at the shortcomings of the previous bus passenger flow prediction which only considers the time - series features and ignores the spatial dimension features,a graph convolutional long short-term memory unit prediction model combined with an attention mechanism ( AGLSTM) is proposed to predict the bus stops passenger flow. The model uses graph convolution network ( GCN) to extract the spatial features of the passenger flow at each time,and uses long short-term memory network ( LSTM) to extract the temporal features of the passenger flow. In order? to pay more attention to the features of the time when the passenger flow has a significant impact,the attention mechanism module is also introduced into the model. Attention mechanism can evaluate the influence of each hidden state on the output result by calculating the weight? ? ? of hidden state of long short-term memory unit at different times. Through the experimental analysis of the real card data of four bus lines in Beijing,and compared with some classic prediction algorithms,it is proved that the combination model considering the temporal and spatial features can effectively improve the prediction accuracy of the model.