A new hierarchical prediction model has been established to improve the accuracy of coal spontaneous combustion risk assessments. This innovative model leverages the multi-objective genetic NSGA-II optimized random forest algorithm to classify the temperature stages of coal spontaneous combustion based on the concentrations of key gases such as carbon monoxide (CO), carbon dioxide (CO2), methane (CH4), ethane (C2H6), and acetylene (C2H2).
The prevalence of mine fires remains alarmingly high, with estimates indicating over 90% of such incidents are initiated by coal spontaneous combustion. Given the dangerous implications for workers and production efficiency, accurately predicting and classifying these combustion risks is of utmost importance for modern mining operations. Recent advancements have emphasized the need to refine methodologies for assessing spontaneous combustion hazards.
Prior studies have advanced the field by exploring the correlations between coal temperature and the evolution of gas emissions during spontaneous combustion processes. Notably, research has demonstrated how various gases produced under specific thermal conditions can act as indicators of imminent combustion risks. Despite these advances, traditional prediction models often struggled with precision due to the inherent variability of coal types and mining conditions.
This study introduces the dynamic NSGA-II-RF model, developed through coal natural ignition experiments conducted on samples from the Zhalainuoer Coal Field, China. By classifying spontaneous combustion processes across seven distinct phases, the model connects gaseous concentrations with corresponding temperature parameters. The research identified key flows among the gases and their transitions over temperature changes, allowing for effective risk classification.
Through extensive data analysis, the predictive performance of the NSGA-II-RF model was benchmarked against several existing methods, including particle swarm optimization approaches applied to random forests and support vector machines. Results indicated average accuracy improvements of 3%, 6%, and 9% over the PSO-RF, PSO-SVM, and unoptimized random forest models, respectively.
The innovation does not merely reside in enhanced accuracy; its design also permits the early detection of potential ignition hazards, fostering more effective mine safety protocols. The model's ability to adapt to the specifications of different coal types signifies great potential for broad application across the mining industry. By projecting the risks based on real-time gas concentrations, miners can develop preemptive strategies to avert catastrophic incidents.
Overall, this research not only advances predictive modeling techniques but also drives home the necessity for safety improvements within the sector. With this new tool readily available, mining operations can reduce hazards and improve their efficiency, making strides toward safer coal production environments.
Looking forward, the team anticipates the integration of more refined datasets and advanced cross-validation methods to continue refining the model’s predictive capabilities. Future iterations hold the promise of increasing specificity and providing even greater insights, aligning operational practices with the highest safety standards.