Justification of Artificial Neural Network Structure and Parameters for Predicting Concentration of Methane in Coal Face

Abstract

A problem of continuous forecasting a concentration of methane in a working face of a coal mine in order to increase a safety of operations in a longwall face and increase a volume of produced coal by reducing the downtime of the longwall face on a gas factor is considered. It is proposed to solve this problem using neural networks. To justify the rational structure of a neural network that provides the maximum prediction accuracy, four types of neural networks were selected: NARX network; Elman network; Feed-Forward Network and neural network with time delay. It is shown that in order to predict the level of methane concentration in the face, it is necessary to measure the methane concentration at least at three points of the exhaust air and the parameters of the operation mode of the combine and its location in the longwall face. The result of forecasting the level of methane concentration for a neural network of the type NARX as the best in terms of the minimum value of the mean square error is presented. Analysis of the obtained graphs indicates the acceptability of the result, the forecasting accuracy is 93%. The introduction of the proposed technology to predict the level of the methane concentration in the longwall face will increase the safety of operations in the longwall face of the coal mine, as well as increase the volume of the produced coal by reducing a longwall face downtime due by the gas factor.