Gas emission forecasting is an important basis for building new mines and new ventilation plans in the mining panel and gas prevention and management. It plays an important role in reducing gas explosion accidents and ensuring safety in coal mine production (Zhang, 2014). Therefore, the prediction accuracy directly influences the economic-technological indicators of mine, especially in large mines (Harpalani and Chen, 1995). It is very essential to calculate the gas content of coal seams and predict the gas emission from the surrounding rocks in the coal mine. Gas emission has a complex system in which the exact amount of influential variables is not yet known, and a specific model cannot be determined to accurately predict gas emission (He et al., 2008).
As a basic safety practice, one must estimate the gas content of coal seam and proceed to predict gas emission into the mine. Gas may be discharged from gas sources or from the main seam before or during the extraction. The emission of large and irregular amounts of gas in the longwall mining method has provided the need for optimizing the methods used to predict this amount of gas and the ventilation required to dilute the gas.
In the 1950s, some methods and processes were presented to measure the gas content of coal seams, and the statistical methods were used to calculate and predict the gas content and gas emission in coal mines. In the 1980s, various methods were proposed to predict gas emission. Since then, the analog method, geological mathematical model, speed method and other methods of predicting gas emission have been proposed and used. All new methods and technologies provided a scientific basis for the mine design and reconstruction (Li, el at., 2014). Some new methods for predicting gas emission in coal seam are: multivariate linear regression (Shi and Wu, 2008), grey system theory (Wu, Tian, Song and et al., 2005), neural network (Zhou, Chang and Zhang, 2007), support vector machine (Fu and Shi, 2013), evidence theory (Cheng, Zhang and Xiaokun, 2012), chaos theory (Shi, Song, He and et al., 2006), fractal theory (He, Pan and Nie, 2006), and rough set theory (Shao, 2009), statistical method, plot headstream method and gas geological map method (Zeng, 2004; Zhu, 2012; Yang, et al., 2018).
Li (2014) used techniques based on self-organizing data method to predict gas emission in coal mine accurately. The results have shown that such methods can automatically analyze nonlinear relationships between gas emission and the effective factors. In the exploitation stages in this coal mine, numerous safety and disaster control factors have been considered in the framework of a comprehensive surveillance plan for effective control of gas-driven disasters. To simplify the analysis, the gas content, coal bed thickness, advance rate, coal production, gas content of the adjacent layer, layer thickness, and distance to adjacent layer were selected for gas emission prediction. Among these seven factors, the adjacent layer thickness exhibited a relatively weaker effect than the other factors. The modeling results were generally in agreement with real data (Li, el at., 2014).
An artificial neural network (ANN) learned on known experimental data was devised to predict the gas emission (Lei. et al, 2014). Gas content, coal bed depth and thickness, coal bed pitch, working face thickness, working face length, advance rate, recovery factor, gas content of the adjacent layer, adjacent layer thickness, layer spacing, petrology of the interbeddings, and exploitation intensity were considered as independent variables while gas emission was taken as the only dependent variable in this study. A total of 18 sets of data were analyzed using a Regression Neural Network (GRNN) and a Multilayer Feedfoward Neural Network, and the results showed that the GRNN could better predict the gas emission at an error of 0.50 (Lei. et al., 2014).
Numerous researchers have reported the use of Monte Carlo simulations for modeling methane gas emission, with the results compared to those of stochastic modeling to reflect the acceptability of the simulation results (Zhou, 2015). In these studies, gas methane emission prediction has been based on the data on the coal structure, coal thickness, coal quality, and gas content (Zhou, 2015).
A gas emission prediction model using a modified neural network algorithm was presented by Fu (2014) and historical data from a real mine were provided to test the model and analyze the results. The findings showed that, upon optimization with the so-called ant colony clustering algorithm, the Elman neural network model had its generalizability and prediction accuracy improved and provided for dynamic prediction of gas emission (Fu, et al., 2014).
Methane gas emission was predicted by a grey-gas geology method. The main factors considered in this study included gas content, coal bed depth, coal bed thickness, per-day coal production rate, geographic setting, and thickness of the adjacent layers, which were evaluated using the theory of gas geology. Results of the test showed that the model provided for high accuracy as its predictions were highly reliable (Wang, 2018).
Currently, researchers have conducted numerous studies to predict the emission of gas in coal. However, as coal gas is a complex phenomenon, the research to predict gas content and gas emissions has not yet reached a consensus, and no mature theory or accurate calculation method has been recognized so far.
Since the gas content is among the most important parameters affecting the accurate assessment and prediction of the coal gas emission, the uncertainty-based estimation of gas content tends to end up with more reliable outcomes. As such, one should evaluate the uncertainty along with the gas emission estimation. The review of literature shows that despite valuable research, no work has been done to predict gas emission based on the gas content uncertainty. In fact, all models have assumed that the parameters affecting the gas emission level are certain, while these parameters are uncertain.
In the present work, the in-place gas content and the associated uncertainty were evaluated in a coal mine. The respective analysis improved the design and led to the application of a system for capturing and control of methane. For this purpose, we used two methods, namely the kriging and Sequential Gaussian simulation, and their results were compared in terms of variance and distribution. The kriging method is recommended when the final objective function is to minimize a single prediction error or to develop a uniform exploratory map of an unknown attribute. The simulation, on the other hand, is advised for the cases where the main objective is to properly evaluate the confidence intervals or model some spatial continuity (Olea, 2009). However, there is no clear criterion for ruling out the simulation or kriging in the geostatistical method (Olea, 2009). The difference of simulation approach with the kriging method is the higher capability for reproducing the spatial continuity patterns and realistic uncertainty models. This difference is based on the inherent preference between the both methods. Accordingly, in the present work, the SGSIM was used as a preferable analysis technique. Then, six main factors are selected to predict methane gas emission, and an uncertainty model is constructed to predict and evaluate methane gas emission using the Monte Carlo simulation method. Seam gas content, coal seam thickness, production rate, advance rate, gas content in adjacent seams and thickness of adjacent seams are the considered independent variables, and methane emission level is the considered dependent variable. This study aims to combine two simulation concepts: geostatistical simulation, to capture gas content uncertainty, and Monte Carlo simulation, to predict coal gas emission based on gas content uncertainty. Therefore, the present article demonstrates an attempt to present this method and build a gas emission prediction model to help prevent and control gas-driven disasters while improving the control level significantly.