Forecasting usage and bike distribution of dockless bike‐sharing using journey data

Abstract

Dockless bike-sharing (DBS) is a novel and prevalent bike-sharing system without stations or docks. DBS has the advantages of convenience and real-time positioning, whereas it brings about some problems such as bike over-supply, disordered parking, and inefficient rebalancing. Forecasting usage and bike distribution are critical in the rebalancing operation for maintaining DBS inventory. By dividing the virtual stations through K -means clustering and processing the four-week Mobike journey data of Nanjing, China, the data of usage and bike count in the 4000 virtual stations are identified. Random forest (RF) is developed to predict the real-time passenger departure, passenger arrival and bike count in the virtual stations. The operation analyses indicate that there is a positive correlation between bike count and usage. RF provides accurate predictions of usage and bike distribution, and almost outperforms five benchmark methods. Forecasting bike distribution is more challenging than forecasting usage because of the volatility of many factors. The results also suggest that bike distribution forecasting based on the usage gap prediction is better than that based on the departure and arrival prediction. This study can help DBS companies in dynamically rebalancing bikes from over-supply regions to over-demand regions in a better way.