期刊文献+

基于多维可预知特征的TCN-LSTM城轨短期客流预测

Short⁃term urban rail passenger flow prediction using temporal convolutional network-long short⁃term memory(TCN-LSTM)based on multidimensional predictable features
下载PDF
导出
摘要 地铁客流量波动受众多因素影响,准确的客流预测数据有利于制定更高效的行车控制方案和客流管控方案。为提高客流预测精度,提出一种基于多维可预知特征的时序卷积神经网络-长短期记忆神经网络模型(TCNLSTM)地铁短期客流预测方法。考虑外部因素的影响,引入Spearman相关系数分析并提取日期、天气等可预知特征及其状态集,以提升预测精度,缩小特征空间,克服了冗余特征数据导致的模型过于复杂问题;通过融合时序卷积神经网络(TCN)提取的客流时间序列特征和可预知特征状态集构建了长短期记忆神经网络(LSTM)层输入,组合模型学习客流与外部影响因素的长短期依赖,从而实现常规日、节假日、不同天气等多场景下的短期客流预测。基于某西南城市地铁刷卡交易数据,对比差分整合移动平均自回归模型(ARIMA)、TCN、LSTM及TCN-LSTM模型的短期客流预测结果,得出组合模型的总体平均绝对误差(MAE)值比其他方法低27%~48%,均方误差(MSE)值低13%~35%,平均绝对百分比误差(MAPE)值低2.8%~6.7%,上述3项指标均表明TCN-LSTM模型的客流预测效果更好。此外,对比实验表明通过融入提取的可预知特征数据,TCN-LSTM模型在测试集上的预测误差评价指标明显降低,所提方法能有效提高地铁短期客流预测精度。 Subway passenger flow is affected by many factors,and accurate passenger flow prediction data facili-tates to the formulation of more efficient traffic control schemes and passenger flow control schemes.In order to improve the accuracy of passenger flow prediction,a short-term subway passenger flow prediction method based on multidimensional predictable features and temporal convolutional network-long short-term memory(TCN-LSTM)has been proposed.Considering the influence of external factors,the prediction accuracy was improved and the fea-ture space was reduced to overcome the initial overly-complex model caused by redundant feature data.The long short-term memory(LSTM)network layer input was constructed by integrating the time series features of passenger flow extracted from the temporal convolutional network(TCN)and the set of predictable feature states.The LSTM network layer input was used to learn the long-term and short-term dependence of passenger flow and external influ-encing factors,so as to achieve short-term passenger flow prediction under multiple scenarios such as working days,holidays and different weather conditions.Based on the Automatic Fare Collection System(AFC)data of a subway station in a southwest city,the short-term passenger flow prediction results of ARIMA,TCN,LSTM and TCN-LSTM models were compared.The overall mean absolute error(MAE)value of the TCN-LSTM method was 27%-48%lower than the other methods,and the mean squared error(MSE)value was 13%-35%lower,and the mean absolute percentage error(MAPE)value decrease by 2.8%-6.7%,which indicates that the TCN-LSTM model gives a better prediction of passenger flow.In addition,comparative experiments show that incorporating the extracted predictable feature data significantly reduces the prediction error evaluation metrics of the TCN-LSTM model on the test set.Thus the proposed method can effectively improve the prediction accuracy of short-term subway passenger flow.
作者 赵利强 李瑞森 唐水雄 唐金金 张涛 ZHAO LiQiang;LI RuiSen;TANG ShuiXiong;TANG JinJin;ZHANG Tao(College of Information Science and Technology,Beijing University of Chemical Technology,Beijing 100029;Beijing Yilu Rail Transit Engineering Co,Ltd,Beijing 101200;School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China)
出处 《北京化工大学学报(自然科学版)》 CAS CSCD 北大核心 2024年第5期86-96,共11页 Journal of Beijing University of Chemical Technology(Natural Science Edition)
基金 国家重点研发计划(2019YFB1600200)。
关键词 城市轨道交通 客流预测 长短期记忆神经网络(LSTM) 时序卷积神经网络(TCN) Spearman相关系数 urban rail transit passenger flow prediction long short-term memory(LSTM) temporal convolu-tional network(TCN) Spearman correlation coefficient
  • 相关文献

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部