Combined model of neural network combined with machine learning for prediction of coal and gas protrusion and hazard class

Accurate prediction of coal and gas protrusion risk can effectively prevent coal and gas protrusion accidents and ensure the safe and efficient production of coal mines. In order to improve the accuracy and universality of the coal and gas herniation prediction model, the hidden layer of the last step of the BP neural network is extracted as the input feature of the random forest, and a combined model of BP neural network combined with random forest （BP-RF model） is constructed. The model is quantitatively evaluated using 60 sets of coal and gas herniation engineering datasets as samples with mean error, mean square error, hazard class prediction accuracy and correlation coefficient. The results show that the accuracy of the established BP-RF model in predicting the hazard class of coal and gas protrusion is 99.9%, and the accuracy of predicting the amount of coal and gas protrusion is 94.87%. The performance of the established BP-RF model is better than that of the BP, RF, and IFOA-GRNN models with higher accuracy. Meanwhile, the sensitivity of all the features is analyzed according to the established model, and the study concluded that the depth of the coal seam, thickness of the coal seam, changes in geological structure, changes in the thickness of the coal seam, changes in the inclination angle of the coal seam, changes in the thickness of the soft layer, the phenomenon of soft collapse of the coal seam, changes in the coefficient of solidity of the coal seam, the phenomenon of drilling dynamics, and the initial speed of gas release are the most sensitive to the predicted results of the model, and they have to be paid great attention to during the actual mining process of the coal mines.