Abstract: The prediction of rental demand of Shared bicycles is necessary for bicycle enterprises to improve operation efficiency and is the premise of rescheduling bicycles. In this paper, three data-driven prediction algorithms, namely random forest, XGBoost and GBDT, are used to predict the rental demand of Shared bikes. In order to predict it more accurately, then a combinational prediction model based on vectorial projection method and the weighted logarithm averaging operator is proposed. We define the concepts of the superiority, non-inferiority and inferiority of the combinational model and prove that this method is at least a non-inferiority prediction method. Finally, this method is applied to the reality to solve the actual bicycle rental problem. The experimental results show that the method can be an effective model in the forecast of Shared bicycle rental demand and it can play a positive role in bicycle rescheduling. This method provides a new direction for the research of bicycle rental demand prediction.

Key words: random forest, XGBoost, GBDT, projection method, demand prediction

摘要： 共享单车的租赁需求量预测对于单车企业提升运营效率十分必要,是单车再调度的前提。为了更加准确地预测出共享单车的租赁需求量,本文结合随机森林、XGBoost、GBDT三类数据驱动预测算法的优点,提出了一种基于向量投影法的加权对数平均组合模型。定义了组合模型的优性,非劣性,劣性的概念。并证明了该方法至少是一种非劣性的预测方法。通过将该方法运用于现实问题中,以解决实际单车租赁需求量预测问题。实例研究发现:该方法在单车租赁需求量预测中可以为优性预测模型, 能够对单车再调度起到正向作用。该方法可以为单车租赁需求量预测的相关研究提供一种切实有效的解决方向。

关键词: 随机森林, XGBoost, GBDT, 投影法, 需求量预测