Multiple Sequence Long and Short Memory Network Model for Corner Gas Concentration Prediction on Coal Mine Workings.

Area of Science:

• Mine safety engineering
• Computational intelligence
• Time series analysis

Background:

• Accurate prediction of gas concentration in mine tunnels is crucial for safety.
• Recurrent neural networks show potential but require improvements in accuracy and spatial consideration.

Purpose of the Study:

• To develop an improved gas concentration prediction model for mine tunnel upper corners.
• To incorporate spatial correlations between different mine airway locations.

Main Methods:

• Utilized a multiple sequence long and short memory network (MSLSTM).
• Applied data preprocessing techniques including white noise and smoothness tests, dataset splitting, and windowing.
• Employed grid search and time series decomposition for parameter optimization and analysis.

Main Results:

• A spatially fused MSLSTM model was established with specific parameters (1 layer, 32 neurons, Adam optimizer, 0.001 learning rate, 32 batch size).
• The model achieved a mean square error (MSE) of 0.0013 on the test set.
• Experimental comparisons demonstrated the model's superior performance over other methods.

Conclusions:

• The proposed MSLSTM model effectively predicts gas concentration by integrating spatial information.
• The study provides a reliable method and valuable insights for future mine gas prediction research.