基于属性加权回归的组团式城市轨道交通进出站客流预测模型研究

A Prediction Model of Entry and Exit Passenger Flows of Rail Transit Stations for Group-structured City Based on Attribute Weighted Regression

为增强轨道交通进出站客流回归预测模型在组团式城市的适应性,利用多源数据细化和完善各影响因素的统计指标,更加精细地体现不同轨道车站之间的差异。针对组团式城市进出站客流在不同尺度下表现出截然不同的空间分布特征的特点,结合K近邻非参数回归和地理加权回归(Geographically Weighted Regression,GWR)模型,采用样本之间的属性差异表征异质性特征,提出一种属性加权回归(Attribute Weighted Regression,AWR)模型。重庆中心城区的案例分析表明:AWR模型能够兼顾样本集合在不同尺度下的空间分布特征,更适用于样本差异较大的情况,且对样本的空间相关特性没有特定的限制条件,针对组团式城市具有更强的适应性;相比于采用普通最小二乘法(Ordinary Least Squares,OLS)的多元线性回归模型和GWR模型,AWR模型对组团式城市轨道交通进出站客流需求的拟合优度和预测精度均显著提高,且误差的空间负相关性明显减弱,是轨道交通进出站客流预测方法的一种有益补充。

To enhance the adaptability of the regression model of entry and exit passenger flow prediction of rail transit stations for a group-structured city,this paper used multi-source data to refine the statistical indicators of various influencing factors,so as to accurately reflect the differences between different rail transit stations.Since entry and exit passenger flows for a group-structured city have different spatial distribution characteristics at different scales,the attribute differences between samples were used to characterize the heterogeneities of passenger flows,and an Attribute Weighted Regression(AWR)model was proposed by combing the K-nearest neighbors algorithm and Geographically Weighted Regression(GWR)model.The case study in the central area of Chongqing shows that the AWR model can consider the spatial distribution characteristics of sample sets at different scales,and it is more suitable for situations where the samples vary greatly.At the same time,the AWR model has no specific restrictions on spatial correlation characteristics,which makes it more adaptable to group-structured cities.Compared with the Multiple Linear Regression model based on the Ordinary Least Squares(OLS model)and GWR model,the AWR model can significantly improve the goodness of fit and the prediction accuracy of passenger flow demand of rail transit stations for the group-structured city,and the negative spatial correlation of prediction errors is significantly weakened.Therefore,the AWR model proposed is useful for the prediction of entry and exit passenger flow of urban rail transit stations.