A Hierarchical Demand Prediction Method with Station Clustering for Bike Sharing System

Bike sharing system is widely used in many cities. However, the imbalanced usage pattern of bicycles causes over-demand issue which affects user experience. Bike demand prediction is necessary as it is the basis of bike pre-allocation which can satisfy people's demand in advance. In this paper, we propose a hierarchical traffic prediction model to predict bike check-out/in number of each station cluster. Firstly, we conduct station clustering with iterative spectral clustering algorithm, since the pattern of bike usage of several stations close to each other is more regular compared with that of a single station. Then, we adopt gradient boosting regression tree to predict the total check-out number of the whole bike sharing system. Thirdly, each cluster's check-out number is inferred based on its predicted proportion in the total check-out number. Fourthly, we propose inter-cluster transition proportion model which can describe the bike rent-return relationships between cluster pairs and predict check-in numbers of clusters. Finally, we evaluate our prediction model with data from Citi Bike System in New York City. Experiment results show that our prediction method can achieve more accurate and reasonable result compared with existing work.