Prediction of Methane Levels in Underground Coal Mines using Artificial Neural Networks

Abstract

Underground coal mines are known for having extremely volatile environments that are gas prone, especially to methane that can unexpectedly erupt or spike. In this paper, a methane levels forecast system is proposed for underground coal mining areas. The system trains an artificial neural network (ANN) using past data to predict methane levels. Supervised learning method is used to train the ANN with four algorithms (Levenberg-Marquardt, Gradient descent, Scaled Conjugate Gradient, and Resilient back propagation) were compared. The Levenberg-Marquardt (LM) proved to be the most reliable algorithm among the considered methods. The prediction performance is also compared by considering the mean square error (MSE) and root mean square error (RMSE), where the LM training method performs the best. The selection of hidden layer neuron numbers is also discussed to balance the training performance and overfitting problems. The study shows that ANN techniques can be used to predict methane levels with acceptable accuracy. The paper concludes that the accuracy of ANN model is dependent on the combination of input parameters and the training algorithm.