To ensure fair and reasonable benefit distri-bution between prosumers while enhancing decision-making timeliness and energy-sharing benefit distribution efficiency for large-scale prosumers in real-time energy management

this paper firstly constructs a collaborative optimization scheduling framework considering energy sharing for large-scale prosumers. Based on the framework

the energy sharing and optimal operation model is characterized for heterogeneous prosumers with different quantities and types of flexible resources. Secondly

to avoid environmental state instability while improving offline collaborative training efficiency and global policy optimization performance in high-dimensional input spaces

a multi-agent attention proximal policy optimization algorithm adapted to collaborative training of large-scale prosumers is proposed. To further enhance computational efficiency during offline training and benefit settlement

a rapid Shapley value calculation method based on hierarchical random sampling is proposed to distribute benefits of the energy sharing process among prosumers. Case study results demonstrate the proposed energy sharing optimization scheduling and efficient benefit distribution method achieves good scalability and applicability in large-scale prosumer scenarios

effectively improving offline training efficiency while considering prosumers' operational economy and decision-making timeliness.