Priority-Aware Sequential Decision-Making Reinforcement Learning in Energy Communities

Renewable Energy Communities provide an innovative approach for decentralised energy management and cost reduction. However, coordinating diverse resources, particularly electric vehicles and storage systems, presents significant challenges, as simultaneous use often leads to operational conflicts. To address this, this work presents a sequential Multi-Agent Reinforcement Learning strategy in which agents execute actions in a specific order, leveraging the most recent environmental state information. Two existing priority-sorting mechanisms are utilised to verify the proposed approach: a fixed-priority mechanism and an adaptive approach informed by agents’ contributions relative to the community. The models were trained on one month of data and evaluated against a simultaneous baseline using a two-year simulation. Results demonstrate that the sequential approach yields consistent cost savings, achieving up to a 7% monthly decrease and an overall 4% reduction compared to the simultaneous baseline.