Coal mine safety production has long faced various risks such as gas exceeding the limit, roof collapse, and equipment failure. The monitoring systems used in the past relied heavily on fixed threshold alarms and regular manual inspections, which have several significant drawbacks: slow response, frequent false alarms, and limited adaptability in complex and changing underground environments. To address the above issues, this article has developed a coal mine safety production monitoring system based on hybrid machine learning algorithms. This system integrates multiple technological advantages, capturing temporal data features through convolutional neural networks, modeling long-term dependencies using long short-term memory networks, and combining XGBoost algorithm to evaluate feature importance, thus constructing a complete closed-loop process covering "multi-source data collection, intelligent feature extraction, dynamic risk warning, and visual decision support". This study selected real sensor data from a coal mine for three consecutive years as experimental samples, covering a total of 12 monitoring parameters such as gas concentration, roof pressure, carbon monoxide concentration, and wind speed. Experiments have shown that compared with traditional threshold alarm methods and single machine learning models such as support vector machines and random forests, this system achieved an F1 score of 0.96 in anomaly detection tasks, with a performance improvement of about 22%; The warning response time has been shortened to 0.8 seconds, and the real-time performance has been improved by 75%; The false alarm rate has decreased to 3.2%, a decrease of 68%.
Liu et al. (Thu,) studied this question.