Abstract: Accurate prediction of gas concentration is of great significance for the prevention and control of coal mine gas disasters. However, achieving long-term prediction of gas concentration at the working face during deep coal mining under complex geological conditions remains highly challenging. This study proposes a multivariate information fusion prediction method for gas concentration in a fully mechanized mining face based on Long Short-Term Memory (LSTM) networks. This method is based on an indicator system of real-time monitoring data and non-time-series monitoring data, using LSTM nonlinear fitting to fuse multiple indicators. It can achieve up to 10 step prediction of gas concentration in the working face, and the average absolute error of the prediction results is less than 0.04. The average absolute error of gas concentration prediction value (%) is reduced from 0.034 3 to 0.027 3. The results show that the multivariate fusion index system established can accurately predict the changes in gas concentration in the working face, improve the accuracy of the prediction results, and provide an important foundation for gas disaster control.