Development of an Intelligent Identification Model for Mine Water Inrush Sources in Karst Mining Areas Based on Multi-Source Data Fusion and a KPCA-ISSA-SVM Framework

To address the challenges of identifying mine water inrush sources and the low efficiency of risk control under complex karst hydrogeological conditions in the Beiya Gold Mine, Yunnan, this study proposes an intelligent identification model integrating nonlinear feature extraction and intelligent parameter optimization. Utilizing 42 sets of measured water samples (comprising karst springs, surface water, and solution caves), a coupling identification model was constructed based on 11-dimensional features including hydrochemical indices and hydrogen–oxygen isotopes. The model employs Kernel Principal Component Analysis (KPCA) to extract discriminative low-dimensional features from nonlinear data, while the critical parameters of the Support Vector Machine (SVM) are optimized via an Improved Sparrow Search Algorithm (ISSA) to enhance generalization performance. The results demonstrate that the following: (1) the proposed model achieves an identification accuracy of 91.7% on the independent test set, significantly outperforming benchmark models such as RF and standard SVM; (2) three sets of comparative experiments indicate that the fusion of multi-source features yields superior identification performance compared to single-source inputs; and (3) SHAP (shapley additive explanation) interpretability analysis reveals that HCO3−, Mg2+, Ca2+, and F− are the core discriminative factors, with their contribution patterns aligning closely with the hydrogeochemical evolution mechanisms of the mining area. This model achieves a synergy between high-precision identification and mechanical interpretability, providing reliable technical support for water disaster prevention in karst mining areas.