Bicycle sharing systems represent a new urban mode of transportation in developed countries. The exploitation of such systems implies numerous operational challenges and one of them is to ensure users that they will be able to find a bicycle or to leave it at each station. Therefore, the rebalancing system is necessary for maintaining a prefixed number of bicycles at each station in order to improve bicycle utilisation. The main contribution of this paper is the development of an original approach based on the integration of a stochastic Petri net model and a genetic algorithm for performance optimisation of such complex systems. As an application, the proposed method is applied to Cristolib, a real self-service bicycle system of Creteil city, France.