BRAVO: Improving the Rebalancing Operation in Bike Sharing with Rebalancing Range Prediction

Bike sharing systems, which provide a convenient commute choice for short trips, have emerged rapidly in many cities. While bike sharing has greatly facilitated people’s commutes, those systems are facing a costly maintenance issue — rebalancing bikes among stations. We observe that existing systems frequently suffer situations such as no-bike-to-borrow (empty) or no-dock-to-return (full) due to existing ad hoc rebalancing practice. To address this issue, we provide systematic analysis on user trip data, station status data, rebalancing data, and meteorology data, and propose BRAVO – the first practical data-driven bike rebalancing app for operators to improve bike sharing service while reducing the maintenance cost. Specifically, leveraging experiences from two-round round-the-clock field studies and comprehensive information from four data sets, a data-driven model is proposed to capture and predict the safe rebalancing range for each station. Based on this safe rebalancing range, BRAVO computes the optimal rebalancing amounts for the full and empty stations to minimize the rebalancing cost. BRAVO is evaluated with 24-month data from Capital, Hangzhou and NiceRide bikeshare systems. The experiment results show that given the same user demand, BRAVO reduces 28% of the station visits and 37% of the rebalancing amounts.