城市公共自行车租赁系统的合理调度对提高公共自行车使用率至关重要，其关键在于对未 来自行车使用情况进行合理预测，确定每个站点调度需求量。通过对站点历史借还车数据与运行 特性进行归纳分析，利用BP 神经网络模型对单站点借（还） 车频次随时间分布的规律进行预 测，预测值和真实值平均相差约3 辆车，曲线拟合良好，证明模型可实践性较高。在此基础上， 按照调度时间窗内站点饱和度动态平衡的原则确定单站点最佳调度需求量。对浙江温州鹿城区街 心公园站点的实例分析表明，实行按需调度能使早晚高峰单站点“无位可还”的时间缩短约0.5h 以上，从而有效提升站点服务质量和满意度。

The scheduling of urban public bicycle rental system plays an essential role in the promotion of public bicycle usage ratio. The key is reasonably predicting the future bicycle usage and ascertaining the scheduling demand of single station. According to the analysis and induction of the historical rent-and- return data and operating characteristic，the frequency of bicycles′ rent-and-return in single sta⁃ tion was predicted using BP Neural Network model. The average difference between predicted and real values is about 3 units, and the curve fitting is good. It proves the high feasibility of the model. On this basis, the best scheduling demand of single station was confirmed according to the principle of dynamic balance of station saturation in scheduling time window. The case analysis of the central park in Lucheng district, Wenzhou City, Zhejiang Province indicates that implementing the predicted schedule on demand basis could shorten the hours of "no parking space to return" of the single station in the morning and evening busy period by more than 0.5h, thereby the service quality and satisfaction degree could be enhanced effectively.