摘要
为提高城市轨道线网站点短时客流预测精度,在对轨道站点分类的基础上分别对各类型站点进行客流预测。以动态时间弯曲作为度量,采用K-means算法对站点进行分类,分析各类型站点客流时序特征;为弱化原始客流数据中噪声的影响,利用经验模态分解(empirical mode decomposition,EMD)方法对各类站点原始客流进行时频分解;提出一种融合图卷积网络(graph convolution network,GCN)和门控循环单元(gated recurrent unit,GRU)的深度学习模型,并以分解得到的分量作为模型输入。以西安地铁为例进行研究,结果表明,根据连续一周站点客流时序特征可将站点分为办公就业型、密集居住型、休闲娱乐型、偏远居住型和职住均衡型5类。采用平均绝对百分比误差及均方根误差作为评价指标,结果表明本研究所提方法对各类站点客流预测的精度优于基准模型。
To improve the accuracy of short-term passenger flow forecasting at urban rail stations,the passenger flow of each type of station on the basis of the classification of rail stations was predicted.Dynamic time warping was used as a measure,and the K-means algorithm was used to classify the stations.The time-series characteristics of passenger flow at various stations were analyzed.To weaken the influence of noise in the original passenger flow data,the empirical mode decomposition(EMD)method was used to perform time-frequency decomposition on the original passenger flow of various stations.A deep learning model that combined graph convolution network(GCN)and gated recurrent unit(GRU)was proposed,and the components decomposed by EMD were used as model input.Taking Xi'an Metro as an example,the results showed that the stations could be divided into five types:office employment type,dense residential type,leisure and entertainment type,remote residential type,and occupation-residential balance type according to the time series characteristics of passenger flow in a continuous week.The average absolute percentage error and the root mean square error were used as evaluation indicators.The results showed that the method proposed in this study outperformed the baseline model in terms of accuracy for predicting passenger flow at various stations.
作者
徐金华
罗义凯
李昱燃
李岩
XU Jinhua;LUO Yikai;LI Yuran;LI Yan(College of Transportation Engineering,Chang'an University,Xi'an 710064,Shaanxi,China)
出处
《山东大学学报（工学版）》
CAS
CSCD
北大核心
2024年第2期60-68,79,共10页
Journal of Shandong University（Engineering Science）
基金
国家自然科学基金资助项目(51408049)
陕西省自然科学基础研究计划项目(2020JM-237)。
关键词
客流预测
图卷积
门控循环单元
站点分类
经验模态分解
passenger flow prediction
graph convolution
gated recurrent unit
station classification
empirical mode decomposition
