Elamur Research⌕
☰
home / papers / 2026 / LARA: LLM-based Agile Power Distribution Network Restoration from Disastrous Events

LARA: LLM-based Agile Power Distribution Network Restoration from Disastrous Events

Paper · 2026 · Jishnu Warrier, Heqing Huang, Yuzhang Lin et al.

Summary

LARA tackles power distribution network restoration after disasters. Repair crews (RCs), mobile power sources (MPSs) and switch actions must be coordinated over many time steps under strict safety limits. Exact MILP produces strong plans but slows down sharply on bigger feeders. CTDE multi-agent RL needs separate training for each topology and careful reward design. The authors instead treat restoration as language-conditioned planning. A simulator first builds compact tables of pre-checked feasible options (crew-to-fault and MPS-to-island assignments). A greedy routine handles switch closures while keeping the network radial. An LLM then chooses RC/MPS macro-actions only from those tables. Several candidate plans are sampled and scored in short simulated rollouts, and the cycle repeats. Operators can add natural-language field updates mid-episode. On IEEE 13-node and 33-node feeders, LARA nearly matches optimal MILP on the small network. On the larger one, it outperforms a one-hour time-capped MILP, with zero constraint violations and decisions within seconds. LARA therefore offers a fast, interpretable planner that needs no retraining for grid emergencies.

Key points

  • On IEEE 33-node, LARA (k=3) reaches 112.99% of the time-capped MILP score. It finishes in 12.1 timesteps, versus 14.6 for MILP and 22.5 for CTDE-RL. Median latency is 18.1 s, against 58.8 min for MILP, which reached optimality in only 1 of 10 scenarios.
  • On IEEE 13-node, LARA attains 95.58% of optimal MILP, compared with 94.45% for RL and 58.20% for Greedy Proximity. It takes more steps (8.27 vs 4.44) because the objective rewards early restoration of high-priority loads.
  • Across backbones on the 33-node feeder, LLama 3.3 70B leads with 112.99%, ahead of Claude 3.7 Sonnet (103.27%) and Gemini 2.5 Pro (96.80%). Gemini scores best on 13-node (97.85%). LLama is also much faster (18.1 s vs roughly 100–120 s).
  • In ablations, the best setting is k=3 with temperature 0.2. Raising T to 0.8 cuts the 33-node score to 99.93%, and k=1 gives 103.83%.
  • Limitations: experiments use small feeders with only 3–4 agents and run purely in simulation. Voltage and thermal constraints via LinDistFlow are left for future work.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓New York UniversityNew York · United States201 works · 100,180 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Engineering · 133 worksElectrical and Electronic Engineering · 18 works

Cite

Cite as Jishnu Warrier, Heqing Huang, Yuzhang Lin and Sai Qian Zhang. 2026. LARA: LLM-based Agile Power Distribution Network Restoration from Disastrous Events. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.321

@inproceedings{warrier2026lara,
  title = {{LARA: LLM-based Agile Power Distribution Network Restoration from Disastrous Events}},
  author = {Jishnu Warrier, Heqing Huang, Yuzhang Lin and Sai Qian Zhang},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.321},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.321}
}

References

25 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 2017. Proximal policy optimization algorithms. In ICML Workshop / arXiv preprint arXiv:1707.06347. 2018. QMIX: Monotonic value function factorisation for deep multi-agent reinforcement learning. In ICML. 2018. Value-decomposition networks for cooperative multi-agent learning. In AAMAS (Extended Abst… find
  2. B. Chen, C. Chen, J. Wang, and K. L. Butler-Purry. 2018. Multi-time step service restoration for advanced distribution systems and microgrids. IEEE Transactions on Smart Grid, 9(6):6793–6805. find
  3. Tao Ding, Zhaoyu Wang, Wei Jia, Bo Chen, Chen Chen, and Mohammad Shahidehpour. 2020. Multiperiod distribution system restoration with routing repair crews, mobile electric vehicles, and soft-open-point networked microgrids. IEEE Transactions on Smart Grid, 11(6):4795–4808. find
  4. Roshni Anna Jacob, Steve Paul, Souma Chowdhury, Yulia R. Gel, and Jie Zhang. 2024. Real-time outage management in active distribution networks using reinforcement learning over graphs. IEEE Transactions on Smart Grid. find
  5. William H. Kersting. 2001. Radial distribution test feeders. In IEEE Power Engineering Society Winter Meeting. find
  6. Jinsub Kim and Yury Dvorkin. 2019. Enhancing distribution system resilience with mobile energy storage and microgrids. IEEE Transactions on Smart Grid, 10(5):4996–5006. find
  7. S. Konar, A. K. Srivastava, and A. Dubey. 2023. Distributed optimization for autonomous restoration in der-rich distribution system. IEEE Transactions on Power Delivery, 38(5):3205–3217. find
  8. Shunbo Lei, Chen Chen, Hui Zhou, and Yunhe Hou. 2019. Routing and scheduling of mobile power sources for distribution system resilience enhancement. IEEE Transactions on Smart Grid, 10(5). 2021. The surprising effectiveness of MAPPO in cooperative multi-agent games. In NeurIPS Deep RL Workshop / arX… find
  9. Meysam Nazemi, Payman Dehghanian, Xiaonan Lu, and Chen Chen. 2021. Uncertainty-aware deployment of mobile energy storage systems for distribution grid resilience. IEEE Transactions on Smart Grid, 12(4):3200–3214. 2022. Improvement of distribution systems’ resilience via operational resources and dem… find
  10. R. Roofegari Nejad and W. Sun. 2019. Distributed load restoration in unbalanced active distribution systems. IEEE Transactions on Smart Grid, 10(5):5759–5769. 2024. Multiperiod distribution system restoration with routing: Repair crews, mobile electric vehicles, and networked microgrids. IEEE Transa… find
  11. Michael Ahn, Anthony Brohan, Noah Brown, and 1 others. 2022. Do as i can, not as i say: Grounding language in robotic affordances. In Robotics: Science and Systems (RSS).
  12. M. E. Baran and F. F. Wu. 1989a. Network reconfiguration in distribution systems for loss reduction and load balancing. IEEE Transactions on Power Delivery, 4(2):1401–1407. find
  13. M. E. Baran and F. F. Wu. 1989b. Optimal capacitor placement on radial distribution systems. IEEE Transactions on Power Delivery, 4(1):725–734. find
  14. R. Roofegari nejad, W. Sun, and A. Golshani. 2019. Distributed restoration for integrated transmission and distribution systems with ders. IEEE Transactions on Power Systems, 34(6):4964–4973. find
  15. Timo Schick, Jane Dwivedi-Yu, and 1 others. 2023. Toolformer: Language models can teach themselves to use tools. arXiv:2302.04761.
  16. Feifan Shen, Qiuwei Wu, and Yusheng Xue. 2020. Review of service restoration for distribution networks. Journal of Modern Power Systems and Clean Energy, 8(1):1–14. find
  17. X. Sun, J. Chen, H. Zhao, W. Zhang, and Y. Zhang. 2023. Sequential disaster recovery strategy for resilient distribution network based on cyber–physical collaborative optimization. IEEE Transactions on Smart Grid, 14(2):1173–1187. find
  18. B. Taheri, R. K. Gupta, and D. K. Molzahn. 2025. Optimizing parameters of the lindistflow power flow approximation for distribution systems. IEEE Transactions on Smart Grid. find
  19. L. Vu, T. Vu, T. L. Vu, and A. Srivastava. 2024. Multi-agent deep reinforcement learning for distributed load restoration. IEEE Transactions on Smart Grid, 15(2):1749–1760. find
  20. Y. Wang and 1 others. 2019. Coordinating multiple sources for service restoration to enhance resilience of distribution systems. IEEE Transactions on Smart Grid, 10(5):5781–5793. find
  21. Yi Wang, Dawei Qiu, Fei Teng, and Goran Strbac. 2024. Towards microgrid resilience enhancement via mobile power sources and repair crews: A multi-agent reinforcement learning approach. Applied Energy. find
  22. Xiaodong Yang, Chongbo Xu, Jinyu Wen, Youbing Zhang, Qiuwei Wu, Wenping Zuo, and Shijie Cheng. 2023. Cooperative repair scheduling and service restoration for distribution systems with soft open points. IEEE Transactions on Smart Grid, 14(3):1827–1841. find
  23. Shunyu Yao, Jeffrey Zhao, and 1 others. 2023. React: Synergizing reasoning and acting in language models. arXiv:2210.03629.
  24. Shuyu Yao, Peng Wang, Xiaoyu Liu, Hong Zhang, and Tianshu Zhao. 2020. Rolling optimization of mobile energy storage fleets for resilient service restoration. IEEE Transactions on Smart Grid, 11(2):1030–1043. find
  25. Zhigang Ye, Chen Chen, Bo Chen, and Kai Wu. 2021. Resilient service restoration for unbalanced distribution systems with distributed energy resources by leveraging mobile generators. IEEE Transactions on Industrial Informatics, 17(2):1386–1396. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/warrier-2026-lara-llm-based-agile-power/