Analyzing lump coal gas release as a source of coal mine methane: Using GC and machine learning

Abstract

Recent advancements in coalbed methane (CBM) applications have facilitated methane capture and utilization as an energy source from coal seams before exploitation. CBM applications reduce the amount of coal mine methane (CMM) released underground during mining operations. However, during coal exploitation, the fractured coal releases lump coal gas, which often contains a high methane concentration. This lump coal gas cannot be prevented with CBM practices and contributes to the formation of CMM along with seam and strata gases. In this study, the mechanism of lump coal gas components CH₄, CO₂, and CO were evaluated by analyzing the particle size, time, and location of the coal lump. In this sense, coal samples of varying sizes and from diverse locations were sealed in airtight canisters, and changes in gas concentrations over time were analyzed at regular intervals using gas chromatography (GC). The changes in the air gap in the canisters and coal samples lead to manipulations in the direct analysis of GC results. Therefore, initially, within the scope of feature engineering, an release index feature was developed, representing the ratio of the amount of each released gas to the weight of the sample, enabling an investigation among varying sample sizes. Using the developed index, time-dependent polynomial models for methane release were developed for various coal sizes. The results of these models were then analyzed using the k-means clustering algorithm to investigate the changes due to coal location. The analyses indicate that from the moment coal is first fractured, CO and CO₂ gases are consistently higher in larger coal particles. However, methane (CH₄) release per unit mass are initially higher from smaller particles, but over time, methane releases from larger particles become more significant than from smaller particles. On the other hand, initially, the change in methane release based on particle size is not very pronounced, and only clustering can be achieved with a 0.45 F1 score. However, over time, the size-dependent change becomes more significant, with smaller samples releasing less methane compared to larger samples, leading to an increase in the F1 score up to 0.94.