Bike sharing systems (BSS) consist of a set of bikes strategically scattered in order to cover a collection of demand points of a city. Most of the models assume that exists a set of stations in which bikes can be picked up and delivered by users. In the BSS an excessive accumulation of bikes in few stations can occur while shortage takes place in others, that is why repositioning of bikes is needed. In this paper we deal with the dynamic multi-vehicle bike repositioning problem. We first divide bike stations by zones using clustering algorithms, while in a second stage we propose mathematical models for bikes relocation. For second stage two models are proposed: one ensuring that every station is visited exactly once by a vehicle and another one where the model is free to choose how many times visits every station. Models are tested with a real instance from a BSS in Medellín (Colombia) with 51 stations. The models allow to redistribute bikes for a time horizon of 15 hours and several different days. Then, to contrast both models, we compared the results for a real instance of 14 stations using data from March 2017 with the given solution by the real operation.