Abstract:

The roof accident caused by the unbalanced pressure of the mine is one of the major mine accidents, so it is of great significance to perceive the change of the mine pressure in the fully mechanized mining face in advance to ensure the safety and efficiency of the coal seam. To improve the accuracy of mine pressure prediction, a spatiotemporal hybrid prediction model based on LSTM network is proposed in this study. Two independent LSTM networks are used to extract the pressure characteristics on the goaf side and the support moving side respectively, and the obtained data are merged through the fully connected layer, so as to realize the common prediction of mine pressure. Two indicators, MSE and MAE, are used to evaluate the prediction effect of the spatio-temporal mixture model based on LSTM. The experimental results show that MSE and MAE are reduced by 24.49% and 35.24%, respectively. The spatio-temporal mixture prediction model based on LSTM is excellent when compared with the traditional LSTM prediction model, and the prediction method of the spatio-temporal hybrid model has higher reliability and accuracy than the traditional model, and can predict the changes in the mine pressure during the advancement of the working face.

Key words: fully-mechanized mining face, hydraulic support, mine pressure analysis, deep learning, mine pressure prediction, time series prediction, LSTM neural network, hybrid LSTM