基于多维度组合深度学习的轨道交通短时进站客流预测研究
Research on Short-term Inbound Passenger Flow Forecasting of Rail Transit Based on Multi-dimensional Combined Depth Learning
朱永霞 1张佩云2
作者信息
- 1. 安徽交通职业技术学院城市轨道交通与信息工程系,合肥 230051
- 2. 安徽中澳科技职业学院,合肥 230041
- 折叠
摘要
高效精确的短时进站客流预测是城市轨道交通运营管理的重要前提,为提高短时进站客流预测精度,结合CNN、Transformer、BiLSTM 等 3 种预测模型的优点,构建了 ConvTB(CNN-Transformer-BiLSTM)组合模型.其中,卷积神经网络模块能够对多维度轨道交通客流量数据矩阵进行特征提取,注意力机制能够赋予提取到的特征中更为重要的部分更大的权值,双向长短时记忆神经网络能够将客流时间序列中前向与后向的双向相关性提取出来.以合肥地铁不同类型车站多个工作日的进站客流量为数据集,训练并测试了该模型性能,并与ARIMA、SVR、CNN、LSTM、BiL-STM、CNN-LSTM、CNN-BiLSTM等预测模型比较.结果表明:ConvTB组合模型在短时进站客流时序下的预测误差均小于其他模型,具有更高的精度,能更为有效反映短时进站客流的变化趋势.
Abstract
Efficient and accurate short-term inbound passenger flow forecast is an important prerequisite for urban rail transit operation and management.In order to improve the forecasting precision of short-term inbound passenger flow,a combined model of CNN-Transformer-BiLSTM(ConvTB)was established,which combined the advantages of three fore-casting models:CNN,Transformer and BILSTM.In this model,the convolutional neural network module can extract fea-tures from a multidimensional mass transit passenger traffic data matrix,and the attention mechanism can give greater weight to the more important parts of the extracted features,bidirectional long short-term memory neural network can ex-tract the forward and backward bidirectional correlation in the time series of passenger flow.The performance of this mod-el is trained and tested by taking the incoming passenger flow of different types of stations in Hefei metro as data set,and compared with Arima,SVR,CNN,LSTM,BILSTM,CNN-LSTM,CNN-BiLSTM and so on.Taking the incoming passen-ger flow of different types of stations in Hefei metro as data set,the performance of the model is trained and tested,and compared with the existing models in the literature,the results show that the forecasting errors of ConvTB combination model are smaller than other models in short-term inbound passenger flow time series,and it has higher accuracy and can reflect the changing trend of short-term inbound passenger flow forecasting more effectively.
关键词
城市轨道交通/短时进站客流预测/多维度组合/深度学习Key words
urban rail transit/short-term inbound passenger flow forecasting/multi-dimensional combinations/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据