Balancing bike sharing systems with constraint programming

Bike sharing systems need to be properly rebalanced to meet the demand of users and to operate successfully. However, the problem of Balancing Bike Sharing Systems (BBSS) is a demanding task: it requires the design of optimal tours and operating instructions for relocating bikes among stations to maximally comply with the expected future bike demands. In this paper, we tackle the BBSS problem by means of Constraint Programming (CP). First, we introduce two different CP models for the BBSS problem including two custom branching strategies that focus on the most promising routes. Second, we incorporate both models in a Large Neighborhood Search (LNS) approach that is adapted to the respective CP model. Third, we perform an experimental evaluation of our approaches on three different benchmark sets of instances derived from real-world bike sharing systems. We show that our CP models can be easily adapted to the different benchmark problem setups, demonstrating the benefit of using Constraint Programming to address the BBSS problem. Furthermore, in our experimental evaluation, we see that the pure CP (branch & bound) approach outperforms the state-of-the-art MILP on large instances and that the LNS approach is competitive with other existing approaches.