In China, mine water disasters occur frequently, and the accurate prediction of aquifer water level has become an important guarantee for the safe and efficient exploitation of coal resources. In order to improve the accuracy of mine aquifer water level prediction, according to the characteristics of fluctuation and time sequence of aquifer water level in mining face, the vmd-lstm model, which has controllable parameters, strong robustness and is suitable for non-stationary signals, and the long-term and short-term memory network (LSTM), which has stronger capture ability in long-term dependence, is combined to study its effect on aquifer water level prediction in mining face, and compared with the control model. The results show that the vmd-lstm model has the best evaluation results, and its MAE, RMSE, MAPE and R2 indexes are 0.028, 0.035 m/d, 0.007% and 0.96 respectively, each metric achieved improvements of at least 28.2, 41.7, 30.0, and 7.9% compared to other models, which has the highest prediction accuracy and has obvious advantages in predicting large changes; The indicators of vmd-lstm model, vmd-gru model and vmd-rnn model are better than their single LSTM model, GRU model and RNN model, indicating that the decomposition and extraction of aquifer water level characteristic information by VMD method plays a great role in improving the accuracy of aquifer water level prediction; Compared with Gru model and RNN model, the indicators of LSTM model show significant advantages, indicating that LSTM model is more suitable for capturing aquifer water level time series information in long-term dependence. In conclusion, vmd-lstm model has high prediction accuracy and simple structure, which can be used as a convenient aquifer water level prediction model for mine water disaster prevention and control, and plays a certain role in ensuring mine safety production.
Zhang et al. (Thu,) studied this question.