基于CNN-LSTM的城市轨道交通短时进站客流预测研究
Short-term Passenger Inbound Flow Prediction for Urban Rail Transit Based on CNN-LSTM
曹阳 1孙亚 1林立 2郭佳峰2
作者信息
- 1. 上海市城乡建设和交通发展研究院,上海 200040;上海城市综合交通规划科技咨询有限公司,上海 200040
- 2. 卡斯柯信号有限公司,上海 200072
- 折叠
摘要
准确有效预测城市轨道交通站点短时进站客流,对支撑城市轨道交通更安全和高效地运行具有重要意义.结合卷积神经网络(CNN)和长短时记忆网络(LSTM)的特点,提出基于CNN-LSTM组合模型的短时进站客流预测方法.通过对典型站点的数据分析,揭示进站客流在日常通勤和会展活动时的变化规律及其影响因素,并引入日期类型、会展活动等外部特征.最后,在目标站点进行实例分析,并将结果与2个基准模型进行对比,CNN-LSTM组合模型在MAE、RMSE和WMAPE指标中均取得较高精度,验证了其在预测该类站点进站客流方面的准确性和应用价值.
Abstract
Accurately and effectively short-term passenger inbound flow prediction at urban rail transit stations is crucial for supporting safer and more efficient operation of urban rail transit.Based on the characteristics of Convolutional Neural Networks(CNN)and Long Short-Term Memory networks(LSTM),a short-term inbound passenger flow prediction method is proposed using a combined CNN-LSTM model.By analyzing data from a typical station,the patterns and influencing factors of inbound passenger flow during regular commuting and exhibition events are revealed,while also incorporating external features such as date types and exhibition activities.Finally,a case study is conducted at the target station,and the results are compared with two benchmark models.The CNN-LSTM combination model achieves higher accuracy across MAE,RMSE,and WMAPE metrics,confirming its precision and practical value in predicting inbound passenger flow for such stations.
关键词
城市轨道交通/短时客流预测/特征工程/机器学习/组合模型Key words
Urban rail transit/Short-term passenger flow prediction/Feature engineering/Machine learning/Combined model引用本文复制引用
基金项目
上海市城市数字化转型专项(202201034)
出版年
2024
NETL
NSTL
万方数据