A key drive towards 5G networks is the Cloud Radio Access Networks architecture (C-RAN). It consists in decoupling two main elements composing the base station: The Base Band Unit (BBU) and the Remote Radio Head (RRH), allowing for centralized processing of BBUs and efficient resource management. The BBU-RRH associations are optimized in this context, where many RRHs are associated to a single BBU when network load conditions are low. Many studies address the BBU-RRH association problem, also known as RRH clustering, for enhancing the energy efficiency. However, rare contributions seek to optimize the BBU-RRH re-association due to the dynamicity of network load conditions. In fact, frequent re-associations between BBUs and RRHs cause a serious degradation in the Quality of Service (QoS) since connected User Equipments (UEs) experience costly handovers (HOs) to different BBUs. This paper tackles such problem by proposing a hybrid algorithm that is divided into two stages: The first is based on a potential game among RRHs to associate to active BBUs, while the second is a centralized scheme, where a central controller decides of the activation/deactivation of BBUs. We compare the results to a previous centralized proposition and to a classical formulation of BBU-RRH association problem as a bin packing optimization. The results show a good tradeoff between power saving and HO rate reduction. In addition, our proposition outperforms the performances of the bin packing problem.