Multi-Agent Enhanced Complex Decision-Making Support Framework: An Urban Emergency Case Study

With the rapid development of large language model (LLM) technology, its advantages in text generation and experimental design are gradually emerging, such as powerful semantic understanding and generation capabilities, as well as high flexibility and adaptability. Complex decision-making problem experiment design is a field that requires high flexibility and complex design, where traditional methods often struggle to cope with this complexity, resulting in low accuracy, poor consistency, and inefficiency. This paper presents a MA-DMS framework, a multi-agent enhanced design framework for the formulation of experiments on complex decision-making problems based on large language models. Through multi-round collaboration among multiple intelligent roles, this framework transforms broad and complex strategic challenges into experiment-ready problems with clear conditions and well-defined objectives. The experimental results show that the MA-DMS system performs excellently in the transformation process of complex decision-making problem, significantly enhancing the efficiency and accuracy of the problem formulation phase.