This work introduces a decentralized multi-agent reinforcement learning (MARL) framework designed to enable sustainable smart power grids to accommodate high levels of renewable energy generation while maintaining resilient operations under severe disruptions and multi-line faults. The architecture integrates autonomous feeder reconfiguration, Q-learning-based service restoration, coordinated battery energy storage dispatch, and communication-efficient decentralized training, validated under full AC power flow on the IEEE 33-bus feeder and the 179-bus Oberrhein benchmark network.
Key Quantitative Findings:
• Fault Service Restoration: Achieves identical performance to exhaustive centralized searches under single faults, and restores 55.1% of isolated load across all 496 double-fault line pairs while requiring 72.7% fewer feasibility evaluations.
• Renewable Hosting Capacity: Autonomous local Volt–VAr droop control increases photovoltaic (PV) hosting capacity by +51% over unity power factor without active power curtailment.
• Microgrid Islanding & Decarbonization: Coordinated storage dispatch raises fully served critical-load hours under localized islanding, avoiding 158–945 kg of CO₂ emissions per storm event.
@article{sajjadi2026sustainable,
title = {Sustainable Smart Grid Resilience and Renewable Integration Using Multi-Agent Reinforcement Learning},
author = {Sajjadi Mohammadabadi, Seyed Mahmoud and Sajjadi, Sayed Mohammad},
journal = {Sustainability},
volume = {18},
number = {20},
pages = {10574},
year = {2026},
publisher = {MDPI},
doi = {10.3390/su182010574},
url = {https://www.mdpi.com/2071-1050/18/20/10574}
}
S. M. Sajjadi Mohammadabadi and S. M. Sajjadi, "Sustainable Smart Grid Resilience and Renewable Integration Using Multi-Agent Reinforcement Learning," Sustainability, vol. 18, no. 20, p. 10574, 2026, doi: 10.3390/su182010574.
Sajjadi Mohammadabadi, S. M., & Sajjadi, S. M. (2026). Sustainable Smart Grid Resilience and Renewable Integration Using Multi-Agent Reinforcement Learning. Sustainability, 18(20), 10574. https://doi.org/10.3390/su182010574