Towards LLM-Enhanced Digital Twins of Intelligent Computing Center
Yi Yang, Manxian Liu, Kun Jia, Wenjie Dong, jian Zi Wang · 2024
Digital twin techniques enable the digital modeling of the power and environmental systems of intelligent computing systems. However, the traditional analysis methods of digital twins do not integrated expert experience with data-driven approaches flexibly. They also analyze the information from multiple dimensions in isolation without considering their relations. This paper proposes a large language model-enhanced digital twin system for addressing these issues. The proposed system provides natural language-based interactive analysis, allowing experts to collaborate with algorithms for risk analysis; it employs a large language model-driven multi-agent system to simulate the interactions of the dynamic environmental systems, achieving multi-dimensional joint prediction. Additionally, this paper presents a case study to demonstrate the usefulness and effectiveness of the proposed system.