Abstract: In wildfire emergency rescue operations, effective aviation resource scheduling is crucial for enhancing rescue efficiency. An innovative approach is proposed that integrates Large Language Models（LLMs）with multi-agent simulation models to enhance emergency decision-making capabilities. Initially, a comprehensive and scalable multi-agent simulation framework is built to model diverse fire scenarios and rescue operations, providing an experimental basis for the intelligent verification and iteration of decision-making processes. Subsequently, LLMs are incorporated into the simulation model, and the task allocation scheme is further optimized by natural language processing techniques.To improve the quality and stability of LLM-assisted decision-making, a Retrieval Augmented Generation（RAG）technical framework is introduced into the study. Simulation experiments show that LLMs can play a critical auxiliary role in emergency rescue decision-making, significantly improving both efficiency and accuracy.