摘要
针对传统BP神经网络模型存在的计算效率和泛化能力低的问题,采用双层次特征分析方法对铁路旅客发送量统计数据的时间特征进行分析,提取出日趋势特征、月趋势特征、日周期性特征、月周期性特征、春运-暑运特征和黄金周-小长假特征作为模型的输入变量,建立双层次的BP神经网络模型,然后根据Gram-Schmidt正交化定理对双层次BP神经网络模型进行改进,在隐含层的输出采用Gram-Schmidt变换增加投影层,从而得到双层次正交神经网络模型。该模型包括2个相对独立的网络模型,1个用于处理客运量日数据,另1个用于处理月数据,2个网络模型的输出经过合成,最终得到客运量的预测结果。模型的应用证明,在铁路客运量预测中双层次正交神经网络模型比传统的BP神经网络模型更为有效。
The problems of low computational efficiency and low generalization ability existed in the traditional BP neural network model.Bi-level orthogonalization method was adopted to analyze the time feature of the statistical data for railway passenger traffic volume.The trend characters of the day and month,the periodic characters of the day and month,the characters of the spring festival and summer holiday transportation,the characters of the golden week and the minor vacation were extracted as the input variables to build the bi-level BP neural network model.The bi-level BP neural network model was then improved based on Gram-Schmidt orthogonalization theorem.The virtual structure named projection layers were added between the hidden layers and the output layers by Gram-Schmidt transform.The bi-level orthogonalization neural network model was consequently obtained,which consisted of two relatively independent models,one for processing the day data of the passenger traffic volume,the other for processing the month data.The outputs of the two models were incorporated to obtain the prediction results of the passenger traffic volume.The application of the model has proved that the bi-level orthogonalization neural network model is more effective than the traditional BP neural network model in the prediction of railway passenger traffic volume.
出处
《中国铁道科学》
EI
CAS
CSCD
北大核心
2010年第3期126-132,共7页
China Railway Science
基金
铁道部科技研究开发计划项目(2000X056-A)
中国铁道科学研究院行业服务技术创新项目(2008YJ42)
关键词
铁路客运量
运量预测
双层次正交化
神经网络模型
Railway passenger traffic volume
Traffic volume prediction
Bi-level orthogonalization
Neural network model
