Methane Concentration Prediction Method Based on Deep Learning and Classical Time Series Analysis

Methane is one of the most dangerous gases encountered in mining industry. During operations, methane can be broadly classified into three states: excavation, stoppage safety check, and abnormal concentration, which usually a precursor to gas accident, such as coal outburst. Consequently, it vital accurately predict concentrations. Herein, we apply deep learning methods—a recurrent neural network (RNN), long short-term memory (LSTM), gated unit (GRU)—to problem concentration prediction evaluate their efficacy. In addition, propose novel method that combines classical time series analysis with these models. The results revealed GRU has least root mean square error (RMSE) loss RMSE further reduced by approximately 35% using proposed combined approach, models are also less likely result overfitting. Therefore, combining methods provide accurate improve safety.