摘要
铁路网的建设进程应与经济社会发展保持适度匹配,其路网规模受人口资源、经济社会、交通政策和运营组织等因素影响,具有动态、时滞、非线性的复杂特征。首先,在不依赖先验信息的情况下,运用互信息法对人均GDP、全社会货运量、旅客周转量等12项影响铁路网规模的指标进行互信息计算。接着,运用Granger因果检验对初选指标进一步筛选,获得7项最具解释力的指标。然后,利用NARX良好的学习记忆与延迟反馈功能构建测算模型,以筛选所得7项指标作为自变量输入、铁路网里程序列作为因变量自回归输入测算铁路网里程。最后,将本模型与传统BP、NAR和单一NARX等神经网络模型的测算结果进行验证、对比。结果表明本模型解释能力更强、泛化能力更好和结果精度更高。
The construction process of the railway network should appropriately match the economic and social development.The scale of its railway network,affected by many factors such as population resources,economy and society,transportation policies as well as operating organization,has complex characteristics such as dynamics,time lag,and nonlinearity.Firstly,without relying on prior information,mutual information method was used to calculate the mutual information of selected 12 indicator variables that affect the scale of the railway network,including GDP per capita,freight volume of the whole society,and passenger turnover.Secondly,the Granger causality test was used to further screen the primary indicators,and seven most explanatory indicators to railway network scale were obtained.Then,taking the advantages of the good learning,memory and feedback functions of the NARX,a measurement model was established,using the time series of the seven indicators selected as the independent variable input and the railway network mileage sequence as the dependent variable autoregressive input to measure the corresponding railway network mileage.Finally,in order to verify the effectiveness and accuracy of the method,the calculation results were compared with those of traditional BP,NAR and the single NARX neural network models.The results show that the model mentioned in this paper has stronger interpretation ability,better generalization ability and higher estimation accuracy.
作者
钱名军
李引珍
何瑞春
曾海军
QIAN Mingjun;LI Yinzhen;HE Ruichun;ZENG Haijun(School of Traffic and Transportation, Lanzhou Jiaotong University, Lanzhou 730070, China;Ningbo Rail Transit Group Operation Branch, Ningbo 315101, China)
出处
《铁道学报》
EI
CAS
CSCD
北大核心
2021年第1期28-38,共11页
Journal of the China Railway Society
基金
甘肃省教育厅高等学校创新基金(2020A-038)
兰州交通大学青年基金(2014029)。
关键词
铁路网规模
互信息
GRANGER因果关系检验
NARX
多元时间序列预测
railway network scale
mutual information
Granger causality test
Nonlinear Auto-Regressive Models with Exogenous Inputs Neural Network
multivariate time series prediction
