|
The hydrogeological conditions of mining areas in western China are complicated, and roof flooding accidents often occur. However, Conventional inorganic water chemical parameters can not accurately distinguish the water source with similar composition in some well fields. Therefore, organic indicators including TOC, UV254, and dissolved organic matter (DOM) were added based on inorganic indicators including K++Na+, Ca2+, Mg2+, Cl-, SO42-, HCO3- and TDS, and the water source discrimination model was established in combination with random forest algorithm(RF). The main components and fluorescence intensity of DOM were obtained by fluorescence map and PARAFAC. After principal component analysis (PCA) of the inorganic data set and inorganic-organic data set, the discriminant model of inorganic indicators and the integrated discriminant model of inorganic-organic indicators were constructed by RF. The results show that the performance of the integrated discriminant model is significantly better than that of the inorganic discriminant model. To further improve the accuracy of the model, the artificial fish swarm algorithm(AFSA) is used to optimize the number of trees and the depth of trees in RF. To avoid local optimization, the adaptive speed, adaptive step, and weight attenuation mechanism are introduced into AFSA, and the water source identification model based on PCA-AFSA-RF was established by using the inorganic-organic data set. The results show that the precision, accuracy, recall, and f1_score of PCA-AFSA-RF model reach 93.67%percnt;, 91.93%, 95.19% and 94.13%, respectively, which are 8.48%, 6.75%, 9.18% and 10.48% higher than RF. And the model also accurately discriminates 13 unknown types of water samples. Therefore, it can be considered that the inorganic-organic comprehensive indicators can significantly improve the identification accuracy of coal seam roof inrush water sources, and the RF algorithm improved by AFSA has better global searching ability and convergence.
|
