大模型智能体驱动的重大突发事件情报决策研究OACHSSCD

[Purpose/Significance]Major emergencies,characterized by high complexity and intertwined derivative risks,pose severe challenges to traditional emergency management systems.Conventional decision-making modes,relying on human expert experience or rule-based systems,have critical pain points:disconnection between intelligence and decision-making,insufficient risk analysis,and poor scheme implementability,failing to process massive multi-source data and generate targeted outputs within a limited time window.To address the academic gaps and practical pain points in intelligence decision-making for major emergencies,this study systematically explores the fundamental theory,process framework,and practical implementation of intelligence decision-making driven by large model agents,so as to provide theoretical and practical support for the intelligent upgrading of emergency risk governance.[Method/Process]Based on the macro intelligence view,taking major emergency risk governance as the entry point and modern decision-making theory as the foundation,this study analyzed the implementation logic of large model agent-driven intelligence decision-making for major emergencies.It constructed a corresponding decision-making framework and established a three-dimensional evalu-ation system featuring traceable evidence,reviewable process,and restrictible expression.Furthermore,taking the Zheng-zhou"July 20"extraordinary rainstorm disaster as an empirical case,an intelligent decision-making agent is built on the Baidu Ernie Agent platform and compared with general large models in a controlled experiment.[Result/Conclusion]The experimental results show that the specialized agent constructed in this study significantly outperforms general large models in evidence binding,process traceability and expression standardization.The decision schemes it generates have stronger scientificity,operability and implementability,which can effectively support the practical needs of major emergency risk governance.This study clarifies the operating mechanism of human-machine collaborative intelligence decision-making in emergency scenarios,and provides a practical path for matching intelligence research with risk governance needs.Future research will expand case coverage and establish quantifiable evaluation indicators to optimize the full-process intelligence decision-making system.