A distributed dynamic multi-agent reinforcement learning based cooperative game framework for multi-sectoral transaction in electricity markets considering dynamic revenue allocation optimisation and Renewable Energy Certificate trading

This study proposes an innovative framework for constructing a tripartite cooperative game model in the electricity market, aiming to harmonise the power dispatch and the trading of Renewable Energy Certificates (RECs) among multiple customers within the same region. In this framework, industrial, commercial, and residential customers engage in energy trading and RECs trading to comprehensively optimise carbon emissions and revenues for industrial self-power plants (SPPs). Given the complexity of the proposed model, the tripartite cooperative game model is formulated as a Markov Decision Process (MDP), which can be solved by the proposed distributed dynamic multi-agent deep reinforcement learning algorithm. The proposed solution algorithm adopts the distributed training distributed execution (DTDE) architecture to enhance training efficiency and incorporates benefit-sharing mechanisms grounded in cooperative contributions. Furthermore, a dynamic reward module is introduced and integrated into the deep reinforcement learning training process to encourage trading agents to explore a broader operational state space. The results show that (1) The renewable energy utilisation increases by 21.14% under the tripartite cooperative game compared to the no-cooperation case; (2) The total electricity cost is reduced by about 26.62% under the proposed tripartite cooperative game model; (3) The proposed algorithm significantly improves training efficiency compared to traditional reinforcement learning methods, achieving up to a 37.32% increase in reward performance. These findings demonstrate the effectiveness of the proposed model in improving market efficiency, reducing costs, and promoting renewable energy utilisation.