Abstract: In order to improve the accuracy of mine water inrush source identification, a KPCA-GWO-SVM-based mine water inrush source identification model is proposed. The algorithm uses kernel principal component analysis（KPCA） for feature dimension reduction to speed up water source identification, and searches for the optimal parameters of support vector machine（SVM） through graywolf optimization（GWO） algorithm to make water source identification more accurate. Taking Zhaogezhuang Mine as the research object, analyzing the main hydro-chemical types of each aquifer, selecting 6 ion indicators, extracting 3 principal components by KPCA, and randomly selecting 70% of the total sample size as the training set （47 groups in total）, 30% as a prediction set（20 groups in total）, the KPCA-GWO-SVM model was constructed and compared with the KPCA-PSO-SVM, KPCA-WOA-SVM and KPCA-SVM models. The results show that the water source prediction results of KPCA-GWO-SVM are consistent with the actual results, which is 10% higher than the prediction accuracy of the model without KPCA processing, and the optimization speed is faster. Compared with other models, the model proposed in this paper has the highest accuracy rate and has superiority.