Abstract: Applying UV-Vis spectral analysis to the identification of water inrush sources in coal mines offers a rapid alternative to conventional hydrochemical methods. To address issues such as limited samples, weak model generalization, and poor geological interpretability in machine learning-based approaches, this study proposes an unsupervised identification model combining UV-Vis spectroscopy with hierarchical clustering. A total of 28 water samples from the Daqing limestone aquifer, Ordovician limestone aquifer, and roadway inrush water were collected from the Niuerzhuang, Sunzhuang, and Xinan mines in the Fengfeng mining area. After denoising and principal component analysis (PCA), hierarchical clustering revealed strong spectral similarities between Daqing and Ordovician aquifers, suggesting potential hydraulic connectivity. The inrush water was primarily classified into the Ordovician aquifer group, with partial contribution from the Daqing aquifer. Compared with hydrochemical-based clustering, UV-Vis spectral data showed superior performance in distinguishing water samples with similar chemical signatures. Linear discriminant analysis (LDA) validated the classification, with an average discriminant score of 0.9998 for the Ordovician group—higher than other aquifers—demonstrating model reliability and agreement with field observations. Furthermore, water level correlation analysis showed a strong positive relationship (maximum coefficient 0.94) and a ～20-day lag between the two aquifers, supporting the existence of hydraulic connectivity. This study offers a practical and interpretable method for identifying inrush sources under data-limited conditions and contributes to understanding karst groundwater flow in North China-type coalfields.