Enhanced Gas Explosion Prediction in Coal Mines Using Intelligent Mining Systems

By Soham NandiReviewed by Susha Cheriyedath, M.Sc.Nov 5 2023

In an article published in the journal PloS One, researchers enhanced gas explosion disaster prediction in coal mines with the help of real-time data from an intelligent mining system. They broke down the coal mine disaster system into sub-systems, creating an early warning index system.

A random forest-based model was trained and optimized, achieving 100% accuracy in gas explosion disaster prediction. Comparatively, a support vector machine (SVM) model achieved 75% accuracy. The optimized random forest model demonstrated superior performance, showcasing a novel approach that combines intelligent mining and multidimensional data analysis for enhanced coal mine safety management.

Background

Coal mines are crucial for energy supply but are fraught with high-risk environments, especially with the threat of coal mine explosion accidents. Researchers are actively seeking methods to enhance early warning capabilities in this context. Previous research has made strides in this area, employing machine learning, neural networks, Bayesian networks, and IoT technologies. However, gaps exist in terms of comprehensive data utilization and the need to further improve early warning system accuracy and reliability, addressing false alarms and missed warnings.

The present study aimed to address these gaps by leveraging intelligent mining systems and multivariate data analysis to develop a robust early warning model for coal mine explosions. The convergence of intelligent mining systems and multivariate data analysis offers promising avenues for improving coal mine explosion early warning. These systems can extract valuable insights from large volumes of data through techniques like data mining and machine learning, resulting in more accurate predictions of explosion probabilities and risks compared to traditional methods.

About the Study

Gas explosions in coal mines hinge on the convergence of gas concentration, oxygen levels, and high-temperature ignition sources. A sophisticated early warning system is developed by breaking down the coal mine disaster system into distinct sub-systems, encompassing disaster-causing factors, disaster-prone environments, and vulnerable populations.

In this study, a multi-level evaluation criteria system was introduced to appraise the risk of coal mine gas explosion disasters. This system factored in the peril posed by risk elements, the stability of potential disaster environments, and the susceptibility of disaster-exposed entities. The authors employed the Random Forest algorithm, utilizing decision trees and ensemble learning techniques, to establish a classification model. By training the model with coal mine data, importance scores were assigned to evaluation indicators, elevating the precision of risk assessment.

Parameter optimization held significant importance in refining the Random Forest model. Two pivotal parameters, Mtry and Ntree, were fine-tuned to prevent overfitting and enhance model precision. Mtry dictated the number of features chosen for each decision tree split, affecting model randomness, while Ntree governed the number of decision trees in the ensemble, influencing model complexity. The authors optimized these parameters to augment the model's performance.

In experimental validation, the study utilized the "Gas sensor array under dynamic gas mixtures" dataset containing 13,910 samples with 17 features, including gas sensor readings and temperature measurements. The model was established with 100 decision trees, a minimum sample size of 30, and specific parameter configurations. The parameter optimization results demonstrated that the best parameter combination is Mtry = 6 and Ntree = 200, yielding the highest accuracy and lowest error.

Results

The experimental results highlighted the effectiveness and superiority of the optimized classification model based on the Random Forest algorithm for gas explosion early warning in coal mines. When tested with 87 samples under unified parameter settings, the optimized model achieved a remarkable accuracy of 100%, surpassing the SVM model's accuracy. Furthermore, the misclassification rate for the optimized model was impressively low at 0.1309, while the SVM model exhibited a relative error of over two percent in sample classification.

The model error of the optimized model was notably lower, 0.0846, compared to the SVM model, emphasizing the superior performance of the Random Forest algorithm in gas explosion risk evaluation, primarily due to the parameter optimization of Mtry and Ntree. This study's findings aligned with previous research in the field of coal mine safety, indicating the potential of the Random Forest algorithm in enhancing safety measures.

Comparative experiments evaluating the gas explosion disaster sub-systems further emphasized the effectiveness of the optimized Random Forest model. When assessing the hazard level of disaster-causing factors, the optimized model provided lower evaluation values compared to the SVM model. These results validated the utility of the optimized Random Forest classification model in gas explosion early warning.

Conclusion

To summarize, the authors developed an enhanced early warning model for coal mine gas explosions by leveraging intelligent mining systems and advanced multi-dimensional data analysis techniques. The research involved constructing a comprehensive early warning index system using data collected from intelligent mining systems and implementing a coal mine gas explosion disaster early warning model based on the Random Forest classification algorithm. The optimized model achieved a remarkable 100% accuracy, while the SVM model's accuracy was 75%.

Furthermore, it exhibited low model error when evaluating the risk factors, disaster-prone environment stability, and disaster-affected body vulnerability. The study demonstrates the potential to significantly enhance coal mine safety by combining intelligent mining systems with data analysis techniques.