The interactive energy sharing with multi-microgrids and shared energy storage facilitates the consumption of renewable energy and improves the efficiency of multi-body operations. Aiming at the source-load uncertainty and data privacy issues, the multi-microgrid cooperative optimal scheduling method is proposed by taking advantage of data-driven and energy sharing. First, a shared energy storage and microgrid economic dispatch model is constructed based on the multi-subject interaction framework. Then, the charging and discharging power pricing strategy of shared energy storage and the economic dispatch decision of multi-microgrids are realized by the data-driven deep reinforcement learning method. Finally, the simulation data analysis shows that the proposed dispatch method can realize the collaborative optimal situation of multi agents.