Dynamic Prediction of Gas Concentration in Mining Face Based on Long Short-term Memory Neural Network

To improve the prediction accuracy of gas concentration in mining face, considering that gas concentration is restricted by historical state, a dynamic prediction model of gas concentration in mining face based on long short-term memory neural network (LSTMNN) is proposed. Using the measured data of gas concentration in a coal mining face in Shanxi Province, the model learning training samples were constructed and the prediction effect was tested. The research shows that LSTMNN algorithm filters the information of gas concentration in the past period through forgetting and memory process, overcomes the short board of traditional prediction method which regards the output value independently, and improves the accuracy and reliability of mine gas concentration prediction; by comparing the prediction results of LSTMNN algorithm with the measured values, the average absolute error, average relative error, root mean square error and Nash model efficiency index of the prediction model were respectively 0.004 319, 0.800 6%, 0.005 714 and 0.436 3.