Prevention and Control of Fires in Goaf Areas Under Mines Based on Neural Networks

Accurate prediction of CO concentration in underground mines is of great significance for underground fire prevention and early warning. In view of the high time-varying nature of CO concentration in underground mines and the limitations of single model prediction, a combined neural network model of backpropagation neural network-long short-term memory network (BP-LSTM) was proposed. Firstly, the BP neural network is used for training, and the training set data, the prediction output and the residuals are input into the LSTM neural network optimized by the Adam algorithm, which is used to train the LSTM model. Combined with the advantages of BP network structure, short prediction time and high LSTM accuracy, the model is built and verified by the deep learning framework Tensorflow, and the prediction of underground CO concentration based on BP-LSTM combined neural network model is realized. The results show that the prediction effect of the combined neural network is better than that of the two single models of BP and LSTM, and the prediction accuracy is improved.