Research on Early Warning of Coal and Gas Outburst Based on HPO-BiLSTM
Peng Ji, Shi Shi-liang, Xingyu Shi
Abstract
Aiming at the problem that coal and gas outburst is difficult to early warning, an early warning index system for coal and gas outburst was analyzed and established in this paper. On this basis, an early warning method based on Bi-directional Long Short-term Memory neural network algorithm (BiLSTM) was proposed. The local optimal solution was optimized to obtain the maximum or minimum value of the fitness function. An early warning model for coal and gas outburst based on the combination of the Hunter-prey Optimization algorithm and BiLSTM (HPO-BiLSTM) was established. The refined parameters of BiLSTM were adjusted using the dynamic optimization of the positions of the hunter and they prey in the HPO algorithm. By verifying the measured data on coal mine, it was found that the model can effectively give warnings about coal and gas outburst, with results consistent with onsite danger levels. This hybrid deep learning algorithm significantly outperforms CNN, LSTM, and BiLSTM by margins ranging from 36-56% for MAE, 21-55% for MAPE, 33-67% for RMSE; averaging at -1.57% for RE. These findings demonstrate that the proposed hybrid deep learning algorithm has excellent early warning capabilities with practical performance making it an effective guidance tool in coal mining operations.