Por Eduardo Gomes (Instituto Superior Técnico).

The rise of Energy Communities presents new challenges in managing distributed energy resources such as electric vehicles and storage systems. Effective coordination must account for conflicting resource demands and scalability across varying community sizes. This work proposes a turn-based Multi-Agent Reinforcement Learning approach designed to address coordination challenges by enabling sequential decision-making using the most current environment information, thereby reducing resource allocation conflicts.

The proposed framework was tested using fixed and adaptive agent prioritization mechanisms to enhance decision-making and system responsiveness. Evaluations conducted using two years of data demonstrate that the turn-based approach achieves consistent improvements in cost efficiency, with up to 7% monthly and 4% overall cost reductions compared to simultaneous execution strategies.

Transmissão via Zoom (pw: SES2024IDL).