Abstract: To solve the problem of user dissatisfaction and demand distortion caused by the imbalance of docked Bike-Sharing System (BSS),a station demand prediction and allocation method based on adaptive clustering was proposed.Stations in BSS were clustered according to the principle of XGBoost first,and then the demand of regions and stations were estimated in turn considering the two situations that leaded to demand distortion.The system service process was simulated many times according to the prediction results of the withdraw/return demand of each station in each period of time to get the optimal value of initial inventory at each station which met the most customer needs on average.The CitiBike system in New York was taken as an example to make an analysis,the operation suggestions was summarized and a comparative experiment was designed to verify the effectiveness of the proposed method for capturing the potential demand of BSS and improving the service capacity.

Key words: bike-sharing systems, demand prediction, demand distortion, site clustering, XGBoost algorithm, optimal initial inventory

摘要： 为了解决有桩共享单车系统(BSS)发生站点失衡导致用户不满和真实需求失真的问题,提出一种基于自适应聚类的站点借还车需求预测和投放量决策方法。首先基于XGBoost原理将站点聚类,然后依次在类层次和站点层次估计借还车需求量,考虑导致需求失真的两种场景,最后根据各站点各时段需求量的预测结果多次模拟系统服务过程,以被拒需求数量的期望值为指标计算各站点以客户满意为导向的最优投放量。论文以纽约CitiBike系统为例进行分析,总结了运营建议,并设计对比试验验证了所提方法对捕捉BSS潜在需求并提高服务率的有效性。

关键词: 共享单车系统, 需求预测, 需求失真, 站点聚类, XGBoost算法, 最优投放量