Quantitative prediction model of water inrush quantities from coal mine roofs based on multi-factor analysis

Accurately and effectively predicting the quantity of water inrush from the roof of coal mines is important for the safety of coal mine production. There is a complex and nonlinear relationship between the water inrush quantities from the coal roof and its influencing factors. To improve the precision and reliability in predicting the water inrush quantity, this paper establishes a water inrush quantity quantitative prediction model for coal seam roof aquifers based on the partial least squares regression (PLSR) and radial basis function (RBF) neural network coupling methods. First, the influencing factors of the coal roof water inrush quantity in the study area are determined, and then PLSR is used to reduce the dimensions of the original data by extracting the principal components with the best interpretation function for the system. The principal components are then used as input to the RBF neural network to model and predict the coal roof water inrush quantity, which effectively overcomes the multicollinearity problem between variables, optimizes the network structure, and improves the learning efficiency and robustness of the network. Finally, the reliability of the method is verified through simulation testing and comparison with other prediction methods. The results show that: compared with the PLSR model, the multiple linear regression (MLR) model, the RBF neural network model, the SVM model, and the FA-RBF neural network model, the fitting and prediction capabilities of the coal roof water inrush quantity prediction model based on the PLSR and the RBF neural network are better than the other models. The average absolute error of fitting of this model is 6.07E-4 m3/h, and the average relative error of fitting is 6.07E-3%; the average absolute error and the average relative error of prediction of this model for new samples are 1.9967 m3/h and 9.8730% respectively. The model combines the unique advantages of the PLSR and the RBF neural network and can deal with the correlation and nonlinear problems between variables, which is very practicable and provides a new way for predicting water inrush quantities from coal roofs.