A High-Precision Gas Emission Prediction Model for Deep Mining Working Faces

Abstract

In the context of deep coal mining, the highly nonlinear and uncertain characteristics of gas emission intensity pose severe safety threats to underground production operations. This study proposes a novel gas emission prediction model integrating Kernel Principal Component Analysis (KPCA), an Improved Crow Search Algorithm (ICSA) embedded with adaptive neighborhood search mechanism, and Support Vector Regression (SVR). First, multi-source data preprocessing procedures including outlier elimination and missing value imputation are implemented to construct a standardized, high-quality dataset. On this basis, KPCA is adopted to extract core nonlinear features from the complex influencing factor system of gas emission, which effectively eliminates information redundancy and significantly improves subsequent computational efficiency. Furthermore, the ICSA with enhanced global exploration and local exploitation balance is introduced to perform adaptive optimization on the hyperparameters of SVR, which fundamentally avoids the local optimum trap in traditional parameter tuning processes and remarkably promotes the generalization performance of the final established KPCA−ICSA−SVR prediction model. Comparative experimental results show that the proposed model outperforms all benchmark models in prediction accuracy and robustness, with the Root Mean Square Error (RMSE) measured at 0.17898 for the training set and 0.3071 for the testing set, fully verifying its outstanding applicability and reliability in dynamic gas emission prediction scenarios of deep working faces.