摘要
城市轨道交通的不同运营状态,通常对应着客流时间序列中不同的本征模态分量(IMF)及时间尺度特征。基于自适应噪声的完全总体经验模态分解(CEEMDAN)算法和双向长短期记忆(BiLSTM)网络,该文构建了地铁短时客流时间序列的组合深度学习预测模型。具体包括:基于CEEMDAN算法实现了客流时间序列的模态分解。分别使用样本熵和层次聚类对IMF分量进行复杂性和相似度分析,并在此基础上完成IMF分量的分类合并与重构;使用Optuna框架中的树形Parzen优化器(TPE)对模型的超参数进行优化,构建CEEMDAN-TPE-BiLSTM组合预测模型。采用实际数据对该文模型进行验证,结果表明,对于特定特征的客流时间序列数据,该文模型的精确性、有效性指标均达到最优。
Different operational states of urban rail transit usually correspond to different Intrinsic Mode Functions(IMFs)and time-scale characteristics in passenger flow time series.A combined deep learning prediction model for short-term passenger flow time series of subway is proposed based on the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and Bidirectional Long Short Term Memory network(BiLSTM),including:mode decomposition of passenger flow time series based on the CEEMDAN algorithm.The sample entropy and hierarchical clustering are used respectively to analyze the complexity and similarity of IMFs.The IMFs are then classified,merged and reconstructed on this basis.The hyper-parameters of the model are optimized using the Tree-structured Parzen Estimator(TPE)in the Optuna framework,and the combined prediction model CEEMDAN-TPE-BiLSTM is established.Actual data are used to validate the model.The results show that the accuracy and validity indicators of the model all reach the optimum for passenger flow time series data with specific characteristics.
作者
朱广宇
孙歆霓
杨荣正
刘康琳
魏运
吴波
ZHU Guangyu;SUN Xinni;YANG Rongzheng;LIU Kanglin;WEI Yun;WU Bo(Beijing Research Center of Urban Traffic Information Sensing and Service Technologies,Beijing Jiaotong University,Beijing 100044,China;Beijing Mass Transit Railway Operation Corporation Limited,Beijing 100014,China;Taiyuan China Railway Rail Transit Construction and Operation Co.,Ltd,Taiyuan 030006,China)
出处
《电子与信息学报》
EI
CSCD
北大核心
2023年第12期4421-4430,共10页
Journal of Electronics & Information Technology
基金
基本科研业务费(2022JBZX024)
国家自然科学基金(62272036,62173167,62132003)。
关键词
城市轨道交通
短时客流时间序列
自适应噪声的完全总体经验模态分解
双向长短期记忆
组合预测
Urban railway transit
Short-term passenger flow time series
Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)
Bidirectional Long Short Term Memory(BiLSTM)
Combined prediction
