A risk evaluation model for coal spontaneous combustion in goaf based onBP neural network

Spontaneous combustion fires in goaf account for the largest proportion of mine fires. During the prevention stage, it is difficult to effectively assess the spontaneous combustion risk, posing significant challenges for prevention and control. To address this, deep learning neural network technology was introduced into the field of coal spontaneous combustion risk assessment, and a risk evaluation model based on a BP neural network was proposed. A total of 135 sets of field data were collected to establish a multi - level evaluation index system. The optimal neural network topology was determined using the golden section method, and the model was trained, optimized, and evaluated for prediction accuracy using training and test datasets. The proposed model was applied at the Huojitu Mine of the Daliuta Coal Mine. The results show that the BP neural network-based risk evaluation model for coal spontaneous combustion in goaf, leveraging its strong self-learning capability, good generalization ability, fault tolerance, and mapping capability, enables quantitative assessment of spontaneous combustion risk in goaf. It offers high prediction accuracy and strong practicality, providing a scientific and reliable basis for mine fire prevention and control, safe underground mining, and related operations, and is of practical significance for improving the overall level of mine safety production.