摘要
自行车共享系统逐渐出现在许多城市中,由于在不同时间和站点的自行车需求量(租/还量)不平衡,系统中各站点的自行车需要人工频繁地调整使其不断达到平衡状态,然而实时监控并不能很好地解决这个问题。因此,提出了一个基于网络图的预测模型,可以预测未来时间段内的某个站点自行车的需求量,提前对站点自行车进行分配。通过分层聚类算法对预测站点进行聚类,得到与其相关的站点簇,并对站点簇构建网络模型。最后,使用纽约(NYC)和华盛顿(D.C.)两个自行车共享系统的数据进行实验,并与基线法、历史平均法及ARIMA模型进行比较。结果发现同一簇的站点具有相似的使用模式,模型预测误差率不高于0.45,网络模型预测性能较好,且能够应用于不同城市的自行车共享系统。
Bicycle-sharing systems are widely deployed in many major cities. As the rents/returns of bicycles at different stations in different periods are unbalanced, the bicycles in a system need to be rebalanced frequently. Real-time monitoring cannot tackle this problem well. Therefore, this paper developed a prediction model based on network diagram to predict the num- ber of bicycles that would be rent from/returned to each station in a future period so that rcallocation could be executed in advance. First, it proposed a hierarchical clustering algorithm to cluster bike stations into groups-relevant station clusters. Then, it constructed network model of a relevant station cluster. Finally, it evaluated the model by two bicycle sharing systems in New York City (NYC) and Washington D.C. (D. C. ) respectively, compared with the baseline method, historical mean method and ARIMA model. The results manifest that there is a similarity in the same cluster, model prediction error rate is not higher than 0.45. Performance of the network model is better, and can be applied to different urban bicycle sharing systems.
作者
林燕平
窦万峰
Lin Yanping Dou Wanfeng(School of Computer Science & Technology, Nanjing Normal University, Nanjing 210023, China Jiangsu Research Center of Information Security & Confidential Engineering, Nanjing 210000, China)
出处
《计算机应用研究》
CSCD
北大核心
2017年第9期2692-2695,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(41171298)
关键词
自行车共享系统
分层聚类算法
需求量
预测
bicycle sharing system
hierarchical clustering algorithm
demand number
prediction