摘要
分析高铁日客流量历史数据特征,设计日期和节假日2类标签,并给出其取值范围;结合高铁日客流量的自回归特征,构建适用于高铁日客流量中期(120 d左右)预测的双层平行小波神经网络(Double Layer Parallel Wavelet Neural Network,DLP-WNN)模型。模型中,子网络1以预测所得到的连续若干天日客流量为输入,预测接下来1 d的日客流量;子网络2对子网络1的输出结果进行修正,以每1 d的日期标签和节假日标签等确定型数据为输入,分别预测各天的日客流量。DLP-WNN模型通过对2个子网络每天输出值加权求和得到各天预测结果,其中,子网络1体现近期日客流量的总体趋势,子网络2体现日客流量的逐日波动情况,以此保证中期预测的精度。实例应用表明:利用DLP-WNN模型分别对4种不同距离下的典型O-D对进行120 d的日客流量中期预测,平均绝对百分比误差为7%~12%,明显低于BP神经网络、ELM极限学习机、ELMAN神经网络、GRNN广义回归神经网络和VMD-GA-BP等方法所测结果,验证了DLP-WNN模型适合于开展高铁日客流量中期预测。
The historical data characteristics of daily passenger flow for High-Speed Railway(HSR)are analyzed,and the labels of date and holiday are designed and their value ranges are given.Combined with the autoregressive characteristics of HSR daily passenger flow,a Double Layer Parallel Wavelet Neural Network(DLP-WNN)model is constructed to predict the medium-term(about 120 days)HSR daily passenger flow.In this model,the subnetwork 1 takes the passenger flow of several consecutive days predicted as the input to forecast the daily passenger flow of the next day.The subnetwork 2 is to modify the output of subnetwork 1.It uses the deterministic data such as the date label and holiday label of each day as the input to predict the passenger flow of each day.The DLP-WNN model obtains the forecast results of each day through the weighed sum of the daily output values of two subnetworks,in which the subnetwork 1 shows the overall trend of the recent daily passenger flow,and the subnetwork 2 reflects the daily passenger flow fluctuation,so as to ensure the accuracy of medium-term forecast.Applications show that the mean absolute percent errors of using the DLP-WNN model to forecast the medium-term(about 120 days)daily passenger flow for 4 typical O-D pairs under four different distances are between 7%-12%,which are significantly lower than the measured results of BP neural network,ELM extreme learning machine,ELMAN neural network,GRNN generalized regression neural network,VMD-GA-BP and other methods.It can be seen that the DLP-WNN model is suitable for the medium-term prediction of the daily passenger flow for high-speed railway.
作者
魏堂建
杨星琪
徐光明
史峰
WEI Tangjian;YANG Xingqi;XU Guangming;SHI Feng(School of Transportation and Logistics,East China Jiaotong University,Nanchang Jiangxi 330013,China;Research Centre for High-Speed Railway of Jiangxi Province,East China Jiaotong University,Nanchang Jiangxi 330013,China;School of Traffic and Transportation Engineering,Central South University,Changsha Hunan 410075,China;School of Economics and Management,Beihang University,Beijing 100191,China)
出处
《中国铁道科学》
EI
CAS
CSCD
北大核心
2021年第6期194-204,共11页
China Railway Science
基金
国家自然科学基金资助项目(72171236,71701216)
国家重点研发计划项目(2020YFB1600400)
江西省教育厅科学技术研究重点项目(GJJ200605)
湖南省自然科学基金资助项目(2020JJ5783)。
关键词
高速铁路
客流预测
日客流量
中期预测
预测精度
小波神经网络
High-speed railway
Passenger flow forecast
Daily passenger flow
Medium-term forecast
Prediction accuracy
Wavelet neural network
