Modeling and Multi-agent Simulation of Bicycle Sharing

Abstract

Bicycle sharing services are bicycle rental services which are widely introduced in large cities in Europe, America, and Asia. They differ from ordinary bicycle rental systems in that users are able to rent and return from any station which offers bicycles. This study modeled bicycle sharing services using agent-based modeling and calculated the Nash equilibrium with game theoretic approach. The equilibrium states were shown to depend on the condition such as user’s payoff and cost parameters. The study was developed further by multi-agent simulation using Q learning, constructing several scenarios. The results reproduced the problem which can be found in real bicycle sharing services, the problem where bicycles gather at the bottom of a hill. A reward system for users to take bicycles up the hill was introduced into the simulation, and the best level of reward was found.