基于圈层人口变量的城市轨道交通车站客流预测

考虑城市轨道交通车站客流(指进出站客流)吸引范围内不同距离的人口对车站客流贡献率不同的情况,将人口按照距离车站的远近分为不同圈层,以不同圈层人口作为变量进行车站客流预测.通过偏相关分析验证圈层人口作为变量的合理性,同时获得影响车站客流的其他显著因素.针对线性多元回归预测模型的不合理性,建立了可反映车站客流与自变量高度非线性关系的BP(back propagation)神经网络预测模型.案例研究表明:基于圈层人口变量和BP神经网络的车站客流预测模型在减小误差方面明显优于其他模型,且具有很好的实时性.在上述模型的基础上,构建了已知任意车站背景变量,车站圈层人口对客流的贡献率模型.该模型验证的结果进一步说明基于圈层人口变量和BP神经网络的车站客流预测模型能够很好地反映圈层人口与其他影响车站客流的显著影响因素同车站客流之间的关系.

Forecasting Method of Urban Rail Transit Ridership at Station level Based on Population Variable in Circle Group

Prediction model of urban rail transit station based on multi-variable regression in present has not considered different contribution for station riderships (in this essay, it means exit and enter riderships) of people within different distance to the station, which makes deviation of the riderships. So it is necessary to make people into different circle group according to their distance to the station and people variable in circle group is used to predict the riderships. People variable in circle group has been certified by partial correlation analysis and at the same time, other significant elements influencing station riderships have been obtained. Because of the unreasonable of linear multi-variable regression , Back Propagation Neural Networks prediction model has been built to reflect high non-linear relation between independent variable and dependent variable. The case study indicates that prediction model based on people variable in circle group and BP neural networks has significant more advantage than linear multi-variable regression and BP neural networks models based on total people variable and BP neural networks model based on people variable in circle group and meanwhile, it is real-time. On the base of the above model, the contribution model of people in different circle group to station riderships has been built, where any background variable of the station has already known. The result of the contribution model also indicates the prediction model based on people variable in circle group and BP neural network can reflect the relationship between station riderships and all the elements influencing the riderships better.