As one of the key technologies of C-RAN, the Real-Time-Virtualization ask for virtual management of hardware resources to implement the dynamic allocation of physical resources and load balancing, and ensure the real-time performance of virtual base station, processing delay and jitter controllable. Taking the burst and transient load in the C-RAN centralized access network system, we can improve the effectiveness of the migration algorithm to reduce unnecessary migrations. To achieve this goal, we propose a method to detect the potential overload points, which combined with traditional virtual machine migration and fractal theory. The simulation results show that this method has high prediction accuracy, which can effectively reduce the number of migration of virtual base station and ensure the load balance at the same time.