Identifying the Location and Size of an Underground Mine Fire with Simulated Ventilation Data and Random Forest Model.

Abstract

Underground mine fires are a threat to the safety and health of mine workers. The timely determination of the location and size of an underground fire is of great importance in developing firefighting strategies and reducing the risk of any injuries. Machine learning was used in this paper to develop a predictive model for fire location and fire size in an underground mine. The ventilation data were obtained by simulating different mine fire scenarios with MFire. The ventilation data of all airways were used as features to predict the fire location. Based on the feature importance, five airways were selected to monitor, and the airflow data of the selected airways were used to predict the fire location and fire size. An accuracy score of 0.920 was obtained for the prediction of fire location. In addition, in-depth analyses were conducted to characterize the wrong predictions with the purpose of improving the performance of the random forest model. The results show that the occurrence of fire at closely connected airways at some locations can generate misleading ventilation data for each other and the model performance can be further improved to 0.962 by grouping them. Fire size is another factor affecting the model performance and the model accuracy increases with increasing fire size. The result from this study can help mine safety personnel make informed decisions during a mine fire emergency.