基于决策思维链驱动和多智能体协同的防洪大模型构建与实践OA

To address the the issues of weak knowledge support capabilities and low intelligence level in decision-making processes for flood management of river basins,this paper proposes a decision-making Chain-of-Thought(CoT)framework aligning with the operational workflows of flood management based on Large Language Model(LLM)technology.A multi-agent collaborative flood management model is developed to provide intelligent technical support for decision making on flood scheduling.The method closely follows the information logic of"forecasting-early warning-scenario analysis-plan making"for flood management,to build a flood management knowledge system and the CoT of decision-making for flood scheduling.It integrates a dynamic knowledge-based combination mecha-nism and multi-agent collaborative technology to achieve efficient decomposition and accurate solutions for complex flood management tasks.Experiments and verification were conducted using the 2020 Changjiang River basin flood control operational scenario.The results show that the agents constructed using this method significantly outperforms existing generic LLMs in terms of the logicality and professionalism of responses to flood management business que-ries,the accuracy of analysis and calculation processes,and the interpretability of decisions.This effectively enhances the knowledge accuracy and decision-making reliability of LLMs in flood management,providing a feasible technical path and theoretical reference for the development of intelligent decision support capability on flood manage-ment.