Abstract: Belt fire in underground coal mine belt conveyors poses a significant external hazard threatening coal mine safety production. This study takes the belt conveyor system in the haulage roadway of the 220101 working face in Xinji No.2 Mine as the research object. Addressing the practical needs of fire early warning and prevention, a systematic investigation is carried out on the causes of belt fires, experimental combustion characteristics, and multi-parameter coupled simulation research. By summarizing typical accident cases and combining them with actual field conditions, the main inducing factors for belt conveyor fires in this haulage roadway are identified, including friction-induced temperature rise, electrical faults, and combustible material accumulation. Subsequently, thermogravimetric analysis is employed to experimentally study the combustion characteristics of coal dust and PVC conveyor belts under different mixing ratios（1∶1, 1∶2, 1∶3）. The results indicate that as the proportion of coal dust increases, the weight-loss stages of the mixture are delayed, and the total heat release increases significantly, suggesting a delayed fire occurrence risk and a more severe fire scale. Based on the pyrolysis and combustion parameters obtained from the experiments, a numerical model of a roadway 100 meters long, 5 meters wide, and 3 meters high is established using Pyrosim software. The dynamic development process of belt fires under different load conditions is simulated through a multi-physics coupling method integrating flow field, temperature field, and gas concentration field. Based on the simulation results, a dynamic graded early warning model is constructed, using temperature range as the primary indicator and temperature rise rate and CO concentration as auxiliary criteria. A three-level response mechanism is proposed: at the Blue Alert （Level 1） stage, personnel are arranged for onsite inspection after the early warning threshold is reached; when the Yellow Alert （Level 2） is activated, the belt conveyor is decelerated, and water spray is initiated for cooling; upon triggering the Red Alert （Level 3）, the power supply to the belt conveyor is immediately cut off, and the sprinkler system is fully activated. This study provides a dynamic early warning method and a graded prevention and control strategy based on multi-parameter fusion for underground coal mine belt conveyor fires, which holds significant practical importance for enhancing mine fire emergency response capabilities and safety assurance levels.