Abstract: To enhance the collaborative early warning capability and safety production level under complex geological conditions of gassy mines, this study develops a multi-hazard integrated early warning technology system and platform integrating monitoring, analysis, and warning based on the “5G + Industrial Internet” architecture, using Baode Coal Mine as the engineering background. Aiming at the multi-source heterogeneous and spatiotemporally complex characteristics of coal mine hazard data, a multi-granularity representation and processing method for big data is proposed, achieving real-time dynamic collection and dedicated storage of hazard information such as gas, water, fire, roof, and dust. By analyzing typical hazard causation mechanisms and historical data, a knowledge graph for mine hazard early warning is constructed. Machine learning algorithms such as association rules and Bayesian networks are employed to explore the spatiotemporal evolution patterns of hazard big data, establishing a data-driven self-evolving dynamic risk analysis indicator system. Combining domain expert experience and a reinforcement learning mechanism, a risk decision model library integrating prior knowledge is built. Adaptive machine learning algorithms suitable for complex underground environments are developed, proposing a progressive flexible early warning model based on multi-scale knowledge granules. Finally, a multi-hazard integrated early warning system with functions including precise data mining, intelligent situation analysis, and graded risk warning is developed and demonstrated in Baode Coal Mine. Application results show that the system achieves classified, graded, and accurate early warning of hazard risks, significantly enhancing the mine’s intelligent perception, risk identification, and emergency response capabilities. It provides a promotable technical solution for coal mines with similar conditions and holds important significance for advancing the intelligent development of coal mine safety production.