摘要
准确预测城市轨道交通短时客流量的变化,有助于运营部门做出决策,并帮助轨道交通集团提高服务水平和实现智慧化运营。然而,客流数据的动态性和随机性使短时客流预测变得困难,因此,文章提出了一种组合预测模型,将Transformer模型中的位置编码(Positional Encoding)层与长短期记忆(Long Short-Term Memory,LSTM)神经网络相结合,构建了LSTM-Transformer预测模型。随后以青岛市的106个站点的进站客流数据为研究对象,并使用聚类算法对站点进行聚类分析。在10分钟的时间粒度下,利用前四周的客流数据作为训练数据,对未来一天的客流数据进行预测研究。同时,将差分自回归移动平均模型(Auto-Regressive Integrated Moving Average,ARIMA)、LSTM、GA-SLSTM和Transformer作为对照模型进行验证。通过多组实验证明了文章提出的LSTM-Transformer模型相较于对照模型组具有更好的预测精度和实用性。
Accurately predicting changes in short-term passenger flow for urban rail transit is crucial for operational decision-making and improving service levels and intelligent operations within rail transit groups.However,the dynamic and stochastic nature of passenger flow data presents challenges in short-term prediction.To address this,the study proposes a combined prediction model,the LSTM-Transformer,which integrates the Positional Encoding layer from the Transformer model with the Long Short-Term Memory(LSTM)neural network.The paper focuses on the inbound passenger flow data from 106 stations in Qingdao and conducts clustering analysis using clustering algorithms to group the stations.Subsequently,based on a 10-minute time granularity,the paper utilizes passenger flow data from the preceding four weeks as training data to predict and analyze the passenger flow for the following day.Additionally,the paper compares LSTM-Transformer model with several control models,including the Differential Autoregressive Integrated Moving Average(ARIMA),LSTM,GA-SLSTM,and Transformer.Through multiple experiments,the study demonstrates that the proposed LSTM-Transformer model outperforms the control models in terms of prediction accuracy and practicality.
作者
张思楠
李树彬
曹永军
ZHANG Sinan;LI Shubin;CAO Yongjun(School of Traffi c Engineering,Shandong Jianzhu University,Jinan 250101,China;Institute of Road Traffi c Safety,Shandong Police College,Jinan 250014,China;Jinan Zhiye Electronics Co.,Ltd.,Jinan 250013,China)
出处
《物流科技》
2024年第14期103-106,114,共5页
Logistics Sci-Tech
基金
国家自然科学基金项目(71871130,71971125)
山东省公安厅科技服务项目(SDGP370000000202202004905,SDGP370000000202202006498)。
作者简介
张思楠(1999-),男,陕西咸阳人,山东建筑大学交通工程学院硕士研究生,研究方向:智能交通;通信作者:李树彬(1977-),男,山东聊城人,山东建筑大学交通工程学院,山东警察学院道路交通安全研究所,教授,博士,硕士生导师,研究方向:系统分析与集成、智能交通系统。
