Simulation of dynamic pricing for station-based bike-sharing systems using MATSim: A case study in Zurich, Switzerland

Bike-sharing systems (BSS) are becoming more popular across the world but are faced with the challenge of asymmetric demand patterns, which lead to unbalanced systems and low levels of service. Current rebalancing strategies use trucks to manually redistribute the bikes, which bears high operating costs and causes emissions. A user-based rebalancing approach is tested in this thesis by simulating a station-based BSS in Zurich with a dynamic pricing model using the agent-based transport simulation framework MATSim. Agents are offered monetary incentives to participate in the rebalancing process with the aim of improving the level of service (LoS) of the system while reducing the operating cost for rebalancing. Two different schemes were simulated with varying incentives and time horizons for demand prediction. One scheme allowed the agents to find their own incentives, in the other, incentives were restricted to be accessible through offers only. The author finds that the dynamic pricing approach can increase the LoS in all simulated scenarios but experiences a decrease in the marginal utility. The cost for dynamic pricing heavily varies between scenarios: Lower incentives achieve similar results as high incentives at a lower cost and the restricted scheme can reduce the amount of needlessly paid incentives. Increasing time horizons for demand prediction was found to have little effect on the LoS but can have negative impacts on the costs. While some scenarios showed that dynamic pricing can pay for itself by generating more revenue, most rely on a significant reduction of cost for operator-based rebalancing to be profitable. The model shows promising results regarding the LoS but more research is needed to assess the profitability of dynamic pricing. This includes adding operator-based rebalancing to the model, calibrating it further to represent the demand patterns from the real-world BSS better, varying participation rates, and adding a learning framework to find optimal values for the incentives