Improving the Rebalancing operation in Bike Sharing Systems with Data-Driven Cyber-Control

This CPS project is driven by largescale online multimodal transit feeds to achieve equilibrium in demand and supply relationship in transportation CPS systems through cyber-control. Our work improves the efficiency of existing transportation systems because (i) in addition to macrolevel historical statistics, the availability of massive micro-level trip information will make it possible to apply fine grained real-time control to handle rapid changes in dynamic urban environments, and (ii) aggregated information from multimodal transit allows unbiased control using the multi-view learning and probabilistic Bayesian model. In addition to basic research objectives mentioned in the proposal, we focus on practical projects with direct collaboration with local government and industry (Nice Ride in Minnesota, Capital Bikeshare in Arlington County, and global transit networks over hundreds of cities) to amplify the broader impact of the CPS program.  

This poster, in particular, addresses inefficiencies in bikesharing such as no-bike- to-borrow (empty) or no-dock-to-return (full) due to existing ad hoc rebalancing practice. We provide the first systematic analysis on user trip data, station status data, rebalancing data, and meteorology data, and propose a practical bike rebalancing system with data-driven cyber-control to improve the bike sharing service while reducing the maintenance cost. Specifically, leveraging comprehensive information from four data sets, this paper proposes a data-driven model to capture and predict the demand on bikes/docks and figure out the rebalancing range for each station. We further close the loop by designing an optimal rebalancing route under practical constraints to minimize the rebalancing cost. Our integrated system design is evaluated with the data sets from (i) capital bikeshare system, which has more than 3000 bikes and 350 stations, and (ii) hangzhou bikeshare system, which has more than 84000 bikes and 3300 stations. The experiment results show that given the same user demand, our de- sign reduces 29% of the station visits and 37% of the rebalancing amounts.