An improved LSSVM discrimination model based on factor analysis and moth flame optimization algorithm for identifying water inrush sources across multiple aquifers in mines

To accurately and swiftly identifying the source of water inrush in mines, a discrimination model based on factor analysis (FA) and the moth flame optimization (MFO) algorithm coupled with least squares support vector machine (LSSVM) is proposed. Drawing from the hydrogeological conditions of the Yuanyi Mine in the Huaibei Mining Area, water samples from three aquifers were collected, and ten hydrochemical indicators were selected for the purpose of identifying the water inrush sources. Firstly, FA was performed on these ten indicators to extract five new components that can comprehensively reflect most of the information of all indicators, eliminating redundant information between the original indicators. Then, the extracted new components was used as inputs to the LSSVM model. Finally, the MFO algorithm was used to automatically optimize the two pivotal parameters of the penalty factor C and the kernel function parameter g of LSSVM, and a discrimination model based on FA-MFO-LSSVM was established. Furthermore, a comparative analysis of the discrimination performance of the FA-MFO-LSSVM model against five other models was carried out. The results unequivocally indicate that the FA-MFO-LSSVM model has high discrimination accuracy for both training and testing samples, and compared to the other five models, this model exhibits stronger discriminative performance.