共享单车骑行需求预测研究

Bike-Sharing Demand Forecast Based on Multi-Scale Geographically Weighted Regression Model

传统的共享单车骑行需求预测模型难以估计骑行需求影响因素作用尺度和影响程度的空间差异性,导致骑行需求预测失准等问题。本文提出使用多尺度地理加权回归模型,分析了交叉口密度和地理兴趣点等的密度在不同空间尺度及对骑行需求的影响,并与传统的最小二乘回归模型、地理加权回归模型进行了对比。基于北京实际共享单车骑行数据分析表明,多尺度地理加权回归模型优于最小二乘回归模型和地理加权回归模型,可以捕捉骑行需求影响因素作用尺度和影响程度在空间上的变化及差异,能更精准地预测精细化区域的骑行需求;每个需求影响因素都具有不同的作用尺度,具有时间和空间差异性。影响因素对骑行需求的影响程度依赖于该因素在不同区域的密度,呈现非线性变化。研究成果为科学合理的共享单车站点规划和运营管理提供技术支撑。

It is difficult for traditional bike-sharing demand forecasting models to identify the differences in the spatial scale and degree of the factors,which leads to problems such as inaccurate demand forecasting.This paper proposes to use the Multi-Scale Geographically Weighted Regression(MGWR)to analyze the impact on bike-sharing demand in different spatial scales of the density of intersections and points of interest(POI),and compare the results with Least Squares Regression(OLS)and Geographically Weighted Regression(GWR).According to the results of the bike-sharing data of Beijing,MGWR is better than the other two types of models above,and it can distinguish the differences and changes in the spatial scale of the factors.Thus,the bike-sharing demand can be forecast with higher accuracy.Each factor has a different spatial scale.Also,they have different influences during different times and spaces.The degree of influence of factors on bike-sharing demand depends on the density of the factors in different areas,showing non-linear changes.The results could provide technical support for scientific and reasonable bike-sharing planning and operation management.