To address the proactivity of shared energy storage undermined by individual user-driven energy storage behavior and the diminished effectiveness and economic efficiency of day-ahead scheduling plans formulated by shared energy storage plants caused by load prediction errors, a bi-level scheduling optimization strategy is proposed for active shared energy storage communities. First, a community structure with prosumers utilizing active shared energy storage is designed. Second, a master-slave game decision-making model is constructed, where the shared energy storage power plant acts as the leader and the prosumer cluster as the follower. A genetic algorithm is used to solve the optimal day-ahead scheduling plan for the communities. Lastly, intra-day power generation and load are predicted by using Latin hypercube sampling, and an intra-day rolling scheduling model based on model predictive control (MPC) is developed. This allows the implementation of bi-level day-ahead and intra-day scheduling optimization for communities with prosumers, which enhances the proactivity of the shared energy storage plant. Experimental results show that the proposed model effectively maximizes the balanced interests of all parties in the game. Compared to that of existing user-driven shared energy storage, the economic return of the energy storage plant increases by 16.3%, which ensures the proactivity of shared energy storage.