Abstract: Aiming at the problems of the monitoring method of thermodynamic disaster in coal mines is single, which are prone to misjudgment and leakage, a method for monitoring thermodynamic disaster of mines based on BP neural network is proposed. According to the mutual coupling relation of the mine thermodynamic causal factors, the method uses three-level architecture for data-characteristics-decision of multi-source data fusion, first, the feature parameter data is normalized using the Kalman filter algorithm at the data level, and then, BP neural network is used for multi-source fusion recognition of feature parameter data at feature level to obtain the feature recognition result of coal seam spontaneous combustion and open flame combustion. Finally, the characteristic recognition results are fused with gas concentration, coal dust concentration and characteristic signal duration at the decision level to get the final monitoring verdict results. The research shows that this method integrates multi-parameter fusion judgments, improves the accuracy of judgment and identification of mine thermodynamic disaster, can effectively solve the problem of misjudgment and leakage of mine thermodynamic disaster monitoring.