校园无桩共享单车时空动态需求预测

Time and space dynamic demand forecasting of station-free sharing bikes on campus

为了提出及时有效的校园无桩共享单车运营管控策略,挖掘了新冠肺炎疫情发生前东南大学校园无桩共享单车海量出行数据,包含15 687条起点数据和15 410条讫点数据,分析了校园内无桩共享单车时空分布特征,利用时间序列回归方法分别以15和30 min间隔建立了校园内无桩共享单车短时供需预测模型,进一步综合运用空间密度(DBSCAN)聚类和K-dist图构建了校区内共享单车停放热点区域的辨识方法,在此基础上给出了共享单车运营管控策略。结果表明:校园无桩共享单车出行需求分布存在明显的时空分布不平衡性。在时间上,校园无桩共享单车出行的工作日日均出行量明显高于休息日,日出行量高峰发生在周一,高峰时段与师生在校区上下课时间紧密相关,出行讫点的高峰时段为工作日07:00~08:00与13:00~14:00,出行起点的高峰时段为工作日11:00~12:00与17:00~18:00;在空间上,出行起讫点位置呈现明显“热点”分布,停放的热点区域集中在校门、图书馆、体育馆、重要教学楼等位置。所构建的校园内无桩共享单车短时供需预测模型的平均绝对误差介于0.600~0.989,表明模型精度较高,能够用于校园无桩共享单车供需缺口预测,通过校园无桩共享单车时空动态需求预测,实现建立与校园空间承载能力、停放硬件设施、出行需求分布等相适应的车辆投放或调配机制,为校园管理者和共享单车运营方优化校园无桩共享单车停放管理提供依据。

To put forward effective balance strategy of station-free sharing bikes on campus, massive trip data of station-free sharing bikes before the COVID-19 occurring on Southeast University were mined. The trip data included 15 687 trip productions and 15 410 trip attractions. Firstly, the data’s temporal and spatial characteristics were analyzed, and a short-term travel prediction model for station-free sharing bikes on campus at intervals of 15 and 30 min separately using autoregressive integrated moving average(ARIMA) was established. Then, an identification method of station-free sharing bikes hot spots on campus by combining the density-based spatial clustering of applications with noise(DBSCAN) clustering method and the K-dist graphs was constructed. Finally, the management and control strategy was proposed for station-free sharing bikes on campus. The results show that imbalanced spatial and temporal demand of bike sharing trips on campus. From temporal demand, the average daily campus trip volume of station-free sharing bikes on weekdays is significantly higher than that on weekend, and the peak of daily campus trip volume occurs on Monday. The campus trip peak hours of station-free sharing bikes are closely related to the school time of teachers and students. The peak hours of trip origins are 07:00 to 08:00 and 13:00 to 14:00 on weekdays, and the peak hours of trip destinations are 11:00 to 12:00 and 17:00 to 18:00 on weekdays. From spatial demand, the origin and destination locations of station-free sharing bikes appear obvious distribution of “hot spots”, and the parking hot spots are concentrated in the school gate, library, gymnasiums and important teaching buildings. The time series forecasting model is developed and the mean absolute error value is between 0.600 to 0.989, and it indicating a high prediction accuracy of the model. The time series forecasting model can provide technical support for real-time scheduling of station-free sharing bikes on campus. By predicting the temporal and spatial travel demand of station-free sharing bikes on campus, research results can help establish sharing bikes’ delivery or allocation mechanism which are adapt to campus space capacity, parking hardware facilities, travel demand distribution and so on. Meanwhile, research results can provide the basis for campus administrators and sharing bikes operators to optimize sharing bikes parking management on campus. 7 tabs, 9 figs, 26 refs.