Pressure Prediction Study of Coal Mining Working Face Based on Nadam-LSTM

Abstract

Coal mine roof accidents pose a significant threat to the safety of personnel and equipment. To mitigate these risks, this study presents an improved time-series prediction model, called Nadam-LSTM, for mining pressure data at the working face in coal mines. The Nadam-LSTM model combines long and short-term memory neural networks (LSTM) with the Nadam optimization algorithm to accurately forecast mining pressure and enhance safety measures. The model incorporates multiple parameters by utilizing factor analysis to identify the key variables influenced by pressure and correlation analysis to quantify their relationships. The Nadam-LSTM model takes into account various input factors, including mining height, coal thickness, burial depth, coal seam dip angle, working face length, roof lithology, and mining speed. The model provides predictions for important output variables, such as first pressure distance, first pressure intensity, periodic pressure distance, and periodic pressure intensity. To further enhance performance, the Nadam optimization algorithm and Dropout regularization are integrated into the model, enabling the construction of a control group prediction model and a mining pressure prediction optimization tool.The performance of the Nadam-LSTM model is evaluated using root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R), and R-squared (R^2). Furthermore, comparative analyses are conducted using Boxplot, Scatterplot, Violin plot, and Taylor diagram to assess and showcase the model’s performance. These evaluations provide comprehensive insights into the model’s accuracy, robustness, and predictive capabilities. Evaluation results demonstrate that the Nadam-LSTM model outperforms a single LSTM model, with significant changes observed in multiple indicators. Such as MAE and RMSE. Also, the inclusion of R and R^2 values provide a measure of the model’s goodness-of-fit. Comparative visualizations, including Boxplot, Scatterplot, Violin plot, and Taylor diagram, further highlight the improved performance and effectiveness of the Nadam-LSTM model in accurately forecasting mining pressure.

Keywords

• Long and short-term memory neural networks
• Nadam optimization algorithm
• pressure at the working surface
• time series prediction