基于时序聚类CEEMDAN-LSTM的轨道短时客流预测  

Prediction of short-term passenger flow in rail transit based on time series clustering with CEEMDAN-LSTM model

作　　者：朱永霞[1] 刘洋[2] 肖赟 ZHU Yongxia;LIU Yang;XIAO Yun(Department of Urban Rail Transit and Information Engineering,Anhui Communications Vocational and Technical College,Hefei 230051,China;College of Transportation Engineering,Chang'an University,Xi'an 710064,China;School of Urban Construction and Transportation,Hefei University,Hefei 230601,China)

机构地区：[1]安徽交通职业技术学院城市轨道交通与信息工程系,安徽合肥230051 [2]长安大学运输工程学院,陕西西安710064 [3]合肥大学城市建设与交通学院,安徽合肥230601

出　　处：《安徽科技学院学报》2024年第5期73-83,共11页Journal of Anhui Science and Technology University

基　　金：安徽省高校自然科学研究项目(KJ2021A1427,2022AH052456,2023AH052975);安徽省高校学科(专业)拔尖人才学术资助项目(gxbjZD2022148);安徽省技能大师工作室(2022jnds018);安徽省教学成果奖一等奖(2023jxcgj530);安徽省质量工程项目(2023jyxm1520)。

摘　　要：目的:高效精确的短时客流预测是城市轨道交通运营管理的重要前提,为提高短时客流预测精度,提出一种基于时序聚类的CEEMDAN-LSTM组合模型。方法:以DTW距离为度量标准,采用Kmeans算法对客流时序进行分类,在此基础上通过CEEMDAN算法进行时序分解以弱化样本噪声干扰,再将分量输入到LSTM模型中进行预测。结果:CEEMDAN-LSTM模型在3类客流时序下的预测误差均小于其他4个基线模型,并能有效反映短时客流的变化趋势;考虑时序聚类的预测模型的预测精度与时效性均优于不分类下的预测模型。结论:以合肥南站地铁的短时进站客流数据为例进行实证分析,证实客流时序聚类对预测精度提升的贡献,并与SARIMA、RF、XGBoost、LSTM等4个预测模型比较,CEEMDAN-LSTM模型具有较高的预测精度,且能有效反映实际客流曲线的变化趋势。Objective:Efficient and accurate short-term passenger flow forecasting is a critical prerequisite for urban rail transit operation and management.To enhance the precision of short-term passenger flow forecasts,a combined CEEMDAN-LSTM model was proposed based on time series clustering.Methods:Using the DTW(Dynamic Time Warping)distance as a metric,the Kmeans algorithm was employed to categorize the passenger flow time series.On this basis,the CEEMDAN algorithm was applied to decompose the time series,mitigating sample noise interference.Subsequently,the decomposed components were fed into the LSTM(Long Short-term Memory)model for prediction.Results:The prediction errors of the CEEMDAN-LSTM model under three types of passenger flow time series were smaller than those of the other four baseline models and could effectively reflect the trend of short-term passenger flow.The prediction accuracy and timeliness of the prediction model that considerd time series clustering is better than that of the prediction model under no classification.Conclusion:The model was empirically analyzed using short-term inbound passenger flow data at Hefei South Railway Station subway and compared with other four forecasting models,CEEMDAN-LSTM model had higher prediction accuracy and could effectively reflect the changing trend of actual passenger flow curve.

关 键 词：城市轨道交通 短时客流预测 时序聚类 CEEMDAN算法 长短期记忆神经网络 

分 类 号：U491[交通运输工程—交通运输规划与管理]