Abstract: In order to discriminate the source of water inrush accurately and prevent water inrush accidents,a KPCA-MPSO-ELM based model was built for discriminating mine water inrush source.For building the model,KPCA was used to reduce the attributes of data and to optimize the initial weights and thresholds of ELM by MPSO.Hydrochemical characteristics of each aquifer in the mine were considered,seven indicators of Ca²⁺,Mg²⁺,K⁺+Na⁺,HCO^-₃,SO^2-₄,Cl^- and total hardness were selected as a basis for judging water inrush.The 45 groups of measured data from Xinzhuangzi mine was selected as an example for analysis.Thirty-three groups of data were used as training data to train the model and other 12 groups of data were used as prediction samples.A comparison was made between the discrimination results by KPCA-MPSO-ELM based model and those by MPSO-ELM and KPCA-PSO-ELM models.The results show that KPCA analysis reduces information overlap between indicator data,that optimizing ELM parameters through MPSO improves the overall search performance and convergence speed of the model,and that the prediction accuracy of KPCA-MPSO-ELM model is higher than that of other two models.

Key words: mine water inrush, water source discrimination, kernel principal component analysis (KPCA), modified particle swarm optimization (MPSO), extreme learning machine (ELM)