自行车共享系统中的短时出租量预测方法

Short-term rent forecast method in bicycle sharing system

自行车共享系统目前存在的主要问题是高峰时段用户租存车困难,实时监控不能很好地解决。为提高公共自行车的利用率,提出一种预测模型,可以预测未来期间将从每个站点集群租出的自行车数量,提前执行重新分配。通过双因素聚类算法对站点进行聚类,考虑时间、天气、温度和风速的因素,预测整个城市公共自行车的出租量,通过变异系数函数计算每个集群所占整体的比例,预测每个集群的出租量。利用纽约市花旗自行车系统的出行数据和纽约市的气象数据进行检测,与GBRT、历史平均法及LSTM进行比较,验证了该模型的有效性。

The main problems currently existing in bicycle sharing systems are that it is difficult to rent cars by users during peak hours, and real-time monitoring cannot solve them well. To increase the utilization rate of public bicycles, a prediction model was proposed to predict the number of bicycles that would be leased from each site cluster in the future and to perform redistribution in advance. The two-factor clustering algorithm was used to cluster the stations. The time, weather, temperature and wind speed factors were taken into consideration to predict the rental amount of the public bicycles in the entire city, and the proportion of each cluster was calculated using the coefficient of variation function. The rental amount for each cluster was predicted. The travel data of the Citi bike system of New York City and the weather data of New York City were used to detect, the detection results were compared with that of GBRT, historical average method and LSTM, the validity of the model was then verified.