考虑动态多时段借还车需求的城市自行车租赁系统规划模型

目的/意义气温、降雨量和风力等天气因素对共享单车骑行量构成了非线性影响，而传统的多元线性回归模型以及基于时序的预测模型，由于其对非线性关系表达能力不足，预测结果往往存在较大误差。 设计/方法在单车价格等经济因素以及市场竞争态势等相对稳定情况下，结合用户潮汐骑行规律和地理空间兴趣点数据构建了基于BP神经网络的天气因素对共享单车骑行量的预测模型，并利用成都市某区域内摩拜单车骑行与天气数据，分别训练了“逐日”和“逐时”为单位的BP预测模型。 结论/发现对比实验结果表明，基于BP神经网络的预测模型具有更高的预测准确度，实验结果还表明不同兴趣点在不同时刻的骑行量受温度和降雨量的影响程度差异较大。研究结果为共享单车公司实现基于天气因素的单车骑行量预测提供了科学方法，并且为单车数量的精准投放与回收提供了依据。

Public bikesharing in North America: early operator understanding and emerging trends

[Purpose/Significance] Weather factors such as temperature, rainfall and wind constitute the nonlinear influence on the riding volume of shared bikes, but the traditional multiple linear regression model and the prediction model based on time sequence often have large errors due to their poor expression ability of the nonlinear relationship. [Design/Methodology] In this paper, the economic factors such as price and the marketing competition situations are supposed relatively stable. A BP neural network model combined with the user tidal riding rule and geographical spatial Point of Interest data to predict the riding volume of shared bikes affected by weather factors. The two prediction models based on per “day” and per “hour” were respectively trained by the real riding data and weather data in Chengdu city, China. [Findings/Conclusions] The comparative experimental results indicate that the BP neural network model has higher prediction accuracy. The results also show that the riding volume of shared bikes in different POIs at different time is greatly affected by temperature and rainfall. This study provides a scientific approach to estimate the riding volume and also a basis for the precise delivery and collection of bikes for shared bike enterprises.