Personalized Dynamic Operating Envelope-Based P2P Energy Trading: A Novel Multiagent Safe Reinforcement Learning

While the growing proliferation of distributed energy resources (DERs) promotes the local energy market to form a user-friendly peer-to-peer (P2P) trading mode, it also renders a physical network system security violation risk. To safeguard the safety network operation, this paper proposes a personalized dynamic operating envelop (DOE) based P2P energy trading clearing model, solved by a novel multiagent safe reinforcement learning (MASRL) method. During the operation horizon, prosumers configure the expected injection power by flexibly scheduling internal resources. Based on the expected injection power, the distribution system operator (DSO) would incorporate the Shapley value into the DOE computation scheme to ensure network integrity and reflect each prosumer’s marginal contribution to P2P trading. A state-wise cost function is introduced to quantify the deviation between the expected injection power and the personalized DOE assignment, thereby indirectly indicating the degree of network violation or potential safety margin. To meet the basic requirements of security network operation, a multiagent absolute state-wise constraint policy optimization (MAASCPO) algorithm is proposed to minimize state-wise cost while enhancing the economic benefits of P2P trading. Simulation results for distribution networks are illustrated to validate the effectiveness of the proposed model in terms of scalability, security, and financial benefits.