Identification of water sources is a key issue of water inrush. This study applied a mixing model based on hydrochemical data to identify water sources and proportions. This study highlighted (1) the importance of model scale and reaction evaluation before using the mixing model, (2) a newly proposed criterion based on eigenvalue analysis to identify the number of end-members, and (3) linear mixing model based on PCA (principal component analysis). 2.5 km² area was an appropriate scale to mixing model because tectonics and lithology were simple. Ion activity, ion exchange, and cycle time of water were evaluated, indicating that groundwater components were dominated by the mixing process. Tracers, such as K, Na, Ca, Mg, Cl, SO₄, δ¹⁸O, δD, EC, TH, and TDS, were used as tracers in the mixing model. Five end-members (representing seawater, Quaternary water, freshwater, Ca-rich water, and Mg-rich water) were identified based on eigenvalue analysis and hydrochemical evolution analysis. A linear mixing algorithm was programmed using Matlab to compute the ratio of each end-member. The results showed that seawater was the dominated water sources (70% at most) threatening the mining operations, especially at the deep levels. Quaternary water mainly recharged the middle level and made up 50% at − 420 m level. Freshwater recharged the shallow level and made up to 40% at − 150 m level. Ca-rich water and Mg-rich water decreased with time. Finally, cross test and extension test of this method showed a high precision in reconstructing ion concentrations, low sensitivity to noise data, and highly extendible to future data.

识别水源是突水的关键问题。这项研究应用了基于水化学数据的混合模型来识别水源和比例。这项研究强调了（1）使用混合模型之前模型规模和反应评估的重要性；（2）基于特征值分析的新提出的标准以识别末端成员的数量；（3）基于PCA的线性混合模型（主要成分分析）。2.5 km ^2的面积是构造模型的合适尺度，因为构造和岩性很简单。对水的离子活性，离子交换和循环时间进行了评估，表明混合过程中地下水成分占主导地位。示踪剂，如钾，钠，钙，镁，氯，SO ₄，δ ¹⁸O，δD，EC，TH和TDS用作混合模型中的示踪剂。基于特征值分析和水化学演化分析，确定了五个最终成员（分别代表海水，第四纪水，淡水，富钙水和富镁水）。使用Matlab对线性混合算法进行编程，以计算每个末端成员的比例。结果表明，海水是威胁采矿作业的主要水源（最多占70％），尤其是在深层。第四纪水主要补给中层水，在-420 m处占50％。淡水补给了浅水位，在-150 m处补给了40％。富钙水和富镁水随时间减少。最后，此方法的交叉测试和扩展测试表明，重构离子浓度的精度很高，对噪声数据的敏感性较低，