If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact us at firstname.lastname@example.org.
Methane concentration forward prediction using machine learning from measurements in underground mines
AuthorDias de Almeida, Tulio
AdvisorDanko, George G
AltmetricsView Usage Statistics
Unmanaged gases inside the mine airways are hazards to health and explosions, mainly methane (CH4) in coal mines. Temperature rise caused by heat release from the strata and machinery is another factor that may harm the health and safety of workers in underground mines. Control of methane and other gas components and the high temperature near a working face require overall localized ventilation management and adequate mine cooling systems. Continuously monitoring the in-situ atmospheric conditions and the number of contaminant gases, especially methane, are important factors for predicting the necessary actions for keeping the mine a safe and healthy place for workers. Studies are reported for predicting methane concentration variations inside underground mines using a long-short-term memory (LSTM) artificial recurrent neural network. Results will be compared to a simple time-series regression predictor (time-series filter). Different combinations of the variables and techniques are tested in the LSTM model to find the best results for accuracy and applicability. Forward time step variations are tested to explore the best prediction outcome. The results show that the LSTM model is limited to one-step-ahead prediction for reasonable accuracy. Furthermore, increasing the number of variables or the training window size does not seem to increase the accuracy of the LSTM predictions. Comparing the results using artificial data and the measured data from the mine, it is observed that the LSTM performs better if the data has a specific pattern and is as smooth as possible.