Research on prediction of coal-gas compound dynamic disaster based on ICSA-CNN

Abstract

With the continuous extension of coal mining into deeper layers, the physical properties and structural characteristics of coal seams and surrounding rocks undergo changes, leading to alterations in corresponding indicator thresholds. This increases the potential risk of coal-gas compound dynamic disasters. In this study, based on data from a specific mine, the iForest and MICE methods were applied for data cleaning. Then, the GRA was used to select influential indicators, and an indicator system for the prediction of coal-gas compound dynamic disasters was established based on iForest-MICE-GRA. Subsequently, A coal-gas compound dynamic disaster prediction model based on ICSA-CNN is constructed by employing CNN from deep learning to establish the model framework and optimizing model hyperparameters through the ICSA. Other models were established for comparative validation, and the ICSA-CNN prediction model developed in this study exhibited the highest accuracy, demonstrating excellent robustness and generalization capabilities. By using the method of ICSA-CNN, this research provides a more reliable decision reference for the prediction and prevention of coal-gas compound dynamic disasters.