Attention-GRU Based Method for Predicting Coal Mine Water Surge Analysis

Accurate prediction of mine water surges is vital to the prevention of mine overflow accidents. Traditional prediction methods are difficult to meet the needs of real-time prediction of water surges. In this paper, a hybrid neural network model combining Attention Mechanism (Attention) and Gated Recursive Unit (GRU) neural network is proposed for predicting water surges. Firstly, the water influx dataset is pre-processed to facilitate model training. Secondly, the dependencies between data are learned through Attention, which allows the model to focus on specific information. The results are then fed into a two-layer GRU module for sequence learning. More data patterns are explored by increasing the depth of the hidden layer, while the Adam optimism is introduced to optimise the model parameters and improve the prediction speed. Finally, root mean square error (RMSE) and coefficient of determination (R
²
) are used as model accuracy evaluation metrics. The model is evaluated by experimental comparison. The experimental results show that the Attention-GRU algorithm model reduces RMSE by 1.56 and improves R
²
by 6% compared to the other three models. The prediction accuracy of the model was improved by the improvement of the GRU algorithm.