Abstract: In the process of coal mine safety production management, there are some problems, such as illegal operation of hazard sources, failure of real-time early warning, excessive discretion of managers, lack of credible electronic evidence chain for rewards and punishments. This paper mainly establishes a real-time tracking and processing system for hazard sources identification based on multi-feature fusion, hazard sources and abnormal conditions of operators. This paper studies and discusses the construction of safety management system based on “self-learning” in underground coal mine enterprises, and perfects the coal mine with positive incentive mechanism-oriented through five ways: condition detection and early warning processing system, formation of credible electronic evidence chain for safety management rewards and punishments, safety management big data analysis platform and safety management “self-learning” platform system.