Abstract: In order to solve the industry pain points of multi-source information heterogeneity，low knowledge utilization efficiency and lagging risk decision-making in coal mine safety management，aiming at the problems of imperfect technical system of intelligent safety management and control in coal mines and insufficient support for the application of large models，a systematic study on the theoretical framework and key technical system of large models of coal mine safety management is carried out.The research ex⁃ plores from the two core dimensions of theory and technology，and builds an intelligent management and control system suitable for coal mine safety scenarios. At the theoretical level，a four-layer closed-loop model of coal mine safety management large model based on cognitive intelligence is constructed，and the two-way interactive operation mechanism of data-driven，knowledge-driven and decision-making synergy in the semantic space of the large model is clarified.At the technical level，a complete technical system in⁃ tegrating multi-modal security perception and semantic alignment，security knowledge graph construction and knowledge injection，disaster risk reasoning and cognitive decision-making，cloud edge collaborative deployment and model lightweight is built.The research results show that the constructed theoretical model and technical system can effectively adapt to the complex safety management and control scenarios of coal mines and solve the shortcomings of traditional safety management intelligence. The large model of coal mine safety management has strong scene adaptation and application value. Deep specialization，multi�modal fusion and intelligent agent coordination are the core development directions of the future iterative evolution of this technology，which can provide theoretical support and technical reference for coal mine in⁃ telligent safety management and control.

Key words: large model, coal mine safety management, theoretical model, key technology