Multichannel Spatiotemporal Prediction of Coal Mine Gas Concentration Based on Data Decomposition and Deep Learning Algorithms

ABSTRACT Coal mine gas is a byproduct of coal and a clean energy source. The accurate prediction of gas concentration is of critical importance to prevent coal mine gas disasters and improve gas utilization efficiency. To fully use monitoring data and accurately predict gas concentration in the working face, the multichannel SpatioTemporal Graph Convolution Network prediction model based on data decomposition and Attention mechanism (Mc‐ASTGCN) was proposed. Gas concentration time‐series datasets were collected from three monitoring points in the same mining face, and their spatiotemporal distribution patterns were analyzed. The improved multivariate variational mode decomposition method was applied to decompose the data into multimodal components. The multichannel spatiotemporal prediction model was constructed by integrating spatiotemporal convolution modules and attention mechanism. Experimental results demonstrate that the proposed model maintains stable performance in short‐ and medium‐term prediction tasks, with an overall RMSE and R 2 of 0.0130 and 0.9856, respectively. Compared with benchmark models, the prediction error under extreme conditions is reduced by 9–57%, and the prediction interval coverage probability exceeds 0.978 at the 99% confidence level. The results indicate that the Mc‐ASTGCN model exhibits practical engineering applicability and provides effective technical support for coal mine gas monitoring and safety early warning.