Intelligent Prediction Model of Coal Spontaneous Combustion Based on Machine Learning

Spontaneous combustion of coal in goaf poses a significant latent threat to coal mine safety, contributing to 60% of coal mine fire incidents. The application of disparate prediction models in different coal mines is predicated on the recognition of variations in coal type, coal quality, geological structure, ventilation conditions, the morphology of the air-mining zone, and the accuracy of monitoring equipment. These variations have the potential to result in different data characteristics and complexity. In this study, a model was constructed utilizing a data-driven machine learning approach, and experiments were conducted in the 16,402 generalized working face of the Laotian mine in Inner Mongolia. The experiments focused on the temperature and gas observation of the air-mining zone, and a total of 180 sets of data were collected. The data were loaded into the established model following data preprocessing, integration of multiple models and performance evaluation. The findings indicate that the stacked regression model demonstrates the most effective performance in predicting the spontaneous combustion temperature of coal in the air-mining zone for the 16,402 generalized working face in the Laotian mine in Inner Mongolia. The multilayer perceptron model demonstrates optimal performance when applied to data from the 150,802 general mining face in Liu Zhuang Mine. The study demonstrates that the model established in this paper integrates multiple models to further enhance the adaptability and robustness of the model, thus enabling it to cope with the complexity of different coal mines. This, in turn, realizes the intelligent prediction of spontaneous coal combustion temperature. Subsequent optimization and incorporation of additional machine learning algorithms will enhance its universality and prediction accuracy, thus facilitating enhanced early warning and safety management in various industrial contexts.