Prediction of Coal Seam Floor Water Inrush Based on DBO-XGBoost Under Small-Sample Data

Abstract

Coal seam floor water inrush is a major geological hazard restricting the safe production of coal mines, directly threatening personnel life and engineering property safety. Accurate prediction of its occurrence risk is of great engineering significance. To address the problems of insufficient generalization ability, easy missed judgments and misjudgments of traditional prediction models caused by scarce water inrush samples and unbalanced data categories in actual mining, this study proposes a prediction model (DBO-XGBoost) integrating the improved SMOTE algorithm and Dung Beetle Optimizer (DBO) with eXtreme Gradient Boosting (XGBoost). A total of 50 sets of water inrush case data from Ordovician limestone nationwide were collected, and 6 core characteristic indicators including water pressure and aquiclude thickness were selected. The improved adaptive SMOTE algorithm was used to balance the data categories, and the 8:2 training-test set split ratio was determined through ten-fold five-cross validation. The global optimization ability of DBO simulating the natural behavior of dung beetles was utilized to optimize the key hyperparameters of XGBoost, fully excavating the nonlinear coupling relationships among features. Comparative verification with 7 models such as XGBoost and PSO-XGBoost showed that the accuracy, precision, recall, and F1-score of the proposed model reached 0.88, 0.89, 0.88, and 0.87 respectively, with an AUC value of 0.928. Compared with the traditional XGBoost, the true positive rate increased by 13.3% and the false negative rate decreased by 40%. The model was applied to the first mining area of Dongda Coal Mine to realize the visual evaluation of water inrush risk. It exhibits excellent accuracy and stability under small sample and complex geological scenarios, providing reliable technical support for the prevention and control of coal mine water inrush disasters.