Balancing Bicycle Sharing Systems: An Approach for the Dynamic Case

Abstract

Operators of public bicycle sharing systems (BSSs) have to regularly redistribute bikes across their stations in order to avoid them getting overly full or empty. We consider the dynamic case where this is done while the system is in use. There are two main objectives: On the one hand it is desirable to reach particular target fill levels at the end of the process so that the stations are likely to meet user demands for the upcoming day(s). On the other hand operators also want to prevent stations from running empty or full during the rebalancing process which would lead to unsatisfied customers. We extend our previous work on the static variant of the problem by introducing an efficient way to model the dynamic case as well as adapting our previous greedy and PILOT construction heuristic, variable neighborhood search and GRASP. Computational experiments are performed on instances based on real-world data from Citybike Wien, a BSS operator in Vienna, where the model for user demands is derived from historical data.