This paper proposes a novel framework based on the Stackelberg game and deep reinforcement learning for multi-microgrids (MGs) in achieving peer-to-peer (P2P) energy trading. A multi-leaders

multi-followers Stackelberg game is utilized to model the P2P energy trading process. Stackelberg equilibrium (SE) is regarded as a P2P optimal trading strategy. A two-stage privacy protection solution technique combining data-driven and model-driven is developed to obtain the SE. Specifically

energy storage scheduling problem in MGs is formulated as a Markov decision process with discrete periods

and a multi-action single-observation deep deterministic policy gradient (MASO-DDPG) algorithm is proposed to tackle optimal scheduling of energy storage in the first stage. According to optimal scheduling of energy storage

the closed-form expression for SE based on model-driven is derived

and distributed SE solution technique (DSET) is developed to obtain SE in the second stage. Case studies involving a 4-Microgrid demonstrate the P2P electricity price obtained by the two-stage method

as a novel pricing mechanism

can reasonably regulate microgrid operation mode and improve microgrid income participating in the P2P market

which verifies effectiveness and superiority of the proposed P2P energy trading model and two-stage solution method.