摘要
准确预测地铁站短时客流量,有助于提前开展安全预警工作,快速做出人员疏导方案。根据上海轨道交通2016年3月2. 4亿条刷卡数据,以及该时间段的天气数据,利用Pearson相关分析法提取了客流量的7个外部天气影响因子,以及3个基于历史数据的内部影响因子。通过对数据的分析,综合考虑工作日、非工作日和高峰时段对客流的影响,提取2个内部显著影响因子。以上海轨道交通莘庄站为例,提出了一种基于深度学习长短期记忆(LSTM)网络结构的地铁站短时客流预测方法。最后,将预测结果与典型时间序列预测算法MLR(多元线性回归)和BP(反向传播)神经网络进行对比,验证了LSTM网络在地铁站短时客流量预测中具有更高的准确性和很好的适用性。
The accurate forecast of short-time passenger flow at subway station is helpful to conduct security pre-warning and make personnel dredging decision quickly.According to 240 million historical smart card data of Shanghai rail transit in March 2016,and the external weather data of that time period,7 external weather impact factors and 3 internal impact factors based on historical data are extracted by using Pearson correlation analysis.Then,based on the data analysis,2 internal significant impact factors are extracted,which take the influence of working days and peak hours on passenger flows into account.Then,taking Xinzhuang Station of Shanghai rail transit as an example,a forecasting method of station passenger flows based on deep learning LSTM is proposed.Finally,compared with the typical time series forecasting method such as MLR(multivariable linear regression)and BP neural network,the LSTM model is validated to have higher accuracy and good applicability in the forecasting of subway station passenger flow.
作者
李梅
李静
魏子健
王思达
陈赖谨
LI Mei;LI Jing;WEI Zijian;WANG Sida;CHEN Laijin(School of Economics and Management,Beijing Jiaotong University,100044,Beijing,China)
出处
《城市轨道交通研究》
北大核心
2018年第11期42-46,77,共6页
Urban Mass Transit
基金
国家自然科学基金"青年基金"项目(71103014)
国家级大学生创新创业训练计划项目(170140032)
北京市哲社办课题(14JGC095)
北京市交通委员会科技课题(B17M00080)
关键词
地铁站
短时客流量预测
深度学习
长短期记忆网络
metro station
short-time passenger flow forecasting
deep learning
long term and short term memory network(LSTM)
作者简介
李梅,硕士研究生。
