SR-DTMA: A Digital Twin-Driven LLM Multi-Agent Framework for Systemic Risk Simulation and Coordinated Decision-Making in Supply Chains
DOI:
https://doi.org/10.71222/76wh4d16Keywords:
Systemic Risk, Digital Twin, Multi-Agent Systems, Large Language Models, Supply Chain Resilience, Risk ContagionAbstract
Systemic risk in global supply chains has intensified with increasing interdependencies among firms, logistics infrastructure, and financial institutions, where localized disruptions such as port closures, financial crises, or policy shocks can propagate across networks and trigger large-scale cascading failures. Existing supply chain risk management and simulation approaches are limited in their ability to simultaneously capture decentralized agent behavior, dynamic system interactions, and the semantic complexity of real-world risk information expressed in natural language. To address these challenges, this paper proposes SR-DTMA (Systemic Risk-aware Digital Twin Multi-Agent framework), a digital twin-driven simulation and decision-making framework that integrates Large Language Model (LLM)-powered agents with decentralized multi-agent learning for systemic risk analysis in supply chains. In SR-DTMA, heterogeneous supply chain entities-including manufacturers, ports, and financial institutions-are modeled as autonomous agents that perceive local operational states, interpret unstructured risk events from policy announcements and news, and make adaptive decisions under systemic shocks. By embedding LLM-based risk cognition within a digital twin environment, the framework enables large-scale simulation of risk contagion, strategic interactions, and the conflict between locally rational decisions and global system stability. Extensive experiments under port closure and financial tightening scenarios show that conventional rule-based and RL-driven agents exhibit higher systemic loss and deeper risk propagation. In contrast, SR-DTMA, integrating LLM-driven agents with a lightweight coordination mechanism, reduces the cumulative systemic loss ratio from 0.327 to 0.163, decreases risk propagation depth from 4.8 to 2.6, and shortens recovery time from 21 to 9 steps. These results highlight the effectiveness of coordinated LLM-based decision-making in mitigating systemic supply chain risk and enhancing network resilience.References
1. C. S. Tang, "Perspectives in supply chain risk management," International journal of production economics, vol. 103, no. 2, pp. 451-488, 2006. doi: 10.1016/j.ijpe.2005.12.006
2. L. V. Snyder, Z. Atan, P. Peng, Y. Rong, A. J. Schmitt, and B. Sinsoysal, “OR/MS models for supply chain disruptions: A review,” IIE Transactions, vol. 48, no. 2, pp. 89–109, 2016.
3. D. Acemoglu, V. M. Carvalho, A. Ozdaglar, and A. Tahbaz‐Salehi, "The network origins of aggregate fluctuations," Econometrica, vol. 80, no. 5, pp. 1977-2016, 2012.
4. V. M. Carvalho, M. Nirei, Y. U. Saito, and A. Tahbaz-Salehi, "Supply chain disruptions: Evidence from the great east japan earthquake," The Quarterly Journal of Economics, vol. 136, no. 2, pp. 1255-1321, 2021. doi: 10.1093/qje/qjaa044
5. M. Grieves, and J. Vickers, "Digital twin: Mitigating unpredictable, undesirable emergent behavior in complex systems," In Transdisciplinary perspectives on complex systems: New findings and approaches, 2016, pp. 85-113. doi: 10.1007/978-3-319-38756-7_4
6. F. Tao, H. Zhang, A. Liu, and A. Y. Nee, "Digital twin in industry: State-of-the-art," IEEE Transactions on industrial informatics, vol. 15, no. 4, pp. 2405-2415, 2018.
7. D. Ivanov, "Predicting the impacts of epidemic outbreaks on global supply chains: A simulation-based analysis on the coronavirus outbreak (COVID-19/SARS-CoV-2) case," Transportation Research Part E: Logistics and Transportation Review, vol. 136, p. 101922, 2020. doi: 10.1016/j.tre.2020.101922.
8. A. Nair, and J. M. Vidal, "Supply network topology and robustness against disruptions-an investigation using multi-agent model," International Journal of Production Research, vol. 49, no. 5, pp. 1391-1404, 2011.
9. Z. Zhang, Z. Yang, and T. Başar, “Multi-agent reinforcement learning: A selective overview of theories and algorithms,” in Handbook of Reinforcement Learning and Control, 2021, pp. 321–384.
10. G. Zhang, H. Zeng, and L. Jiang, “Uni-FinLLM: A unified multimodal large language model with modular task heads for micro-level stock prediction and macro-level systemic risk assessment,” arXiv preprint arXiv:2601.02677, 2026.
11. S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. R. Narasimhan, and Y. Cao, "React: Synergizing reasoning and acting in language models," In The eleventh international conference on learning representations., October, 2022.
12. T. Schick, J. Dwivedi-Yu, R. Dessì, R. Raileanu, M. Lomeli, E. Hambro, and T. Scialom, "Toolformer: Language models can teach themselves to use tools," Advances in Neural Information Processing Systems, vol. 36, pp. 68539-68551, 2023.
13. J. S. Park, J. O'Brien, C. J. Cai, M. R. Morris, P. Liang, and M. S. Bernstein, "Generative agents: Interactive simulacra of human behavior," In Proceedings of the 36th annual acm symposium on user interface software and technology, October, 2023, pp. 1-22. doi: 10.1145/3586183.3606763
14. Z. Li, X. Li, and X. Lin, “Design and implementation of a platform for business intelligence knowledge mining and graph construction based on deep learning,” in Proc. 2nd Int. Symp. Integrated Circuit Design and Integrated Systems, 2025, pp. 137–141.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Yunting Ling, Wenxuan Liu (Author)
This work is licensed under a Creative Commons Attribution 4.0 International License.
