关键词: 矿井充水水源　, 核主成分分析　, 差分进化算法　, 灰狼优化算法　, 支持向量机　, 特征融合

Abstract: To address the problems of insufficient feature extraction,suboptimal model parameter tuning,and poor nonlin ear data processing capability in existing mine water inrush source identification models,this study proposes a mine water in rush source discriminant model based on a Differential Evolution-Grey Wolf Optimizer (DE-GWO) algorithm and feature fu sion.The model first employs Kernel Principal Component Analysis (KPCA) method to reduce the dimensionality and fuse the high-dimensional nonlinear hydrochemical features of mine water inrush sources,eliminating feature redundancy while preser ving critical information.Then,DE-GWO is used to adaptively determine the penalty factor and kernel function parameters of the Support Vector Machine (SVM) through an intelligent optimization process,avoiding the limitations of manual parameter tuning and enabling deep nonlinear fusion and classification of features in the kernel space.Finally,the dimension-reduced fused features are fed into the DE-GWO-optimized SVM model to construct the mine water inrush source discriminant model. Test results show that in the application at the Qianyingzi Coal Mine,the model achieves a discrimination accuracy of 99.4% and a minimum discrimination time of 0.97 s.Moreover,under disturbance conditions of 30 dB Gaussian noise,20% missing samples,and ±10% feature perturbation,the model′s root mean square errors are 0.031,0.032,and 0.031,respectively,all below 0.1,demonstrating excellent anti-interference capability and stability.The model can accurately identify four types of wa ter sources:atmospheric precipitation,surface water,groundwater and goaf water.The study reflects that the model can rapidly and accurately discriminate mine water inrush sources,providing a reliable technical means for rapid identification of mine wa ter inrush and prevention decision-making.