Abstract: As deep mining becomes prevalent in China's coal mining industry, coal-gas compound dynamic disasters pose increasing threat to the safety production of coal mines. This paper adopts the field data of Pingmei No. 8 coal mine for analysis, with the attempt to predict coal-gas compound dynamic disaster through convolutional neural network. Following the routine of the big data processing, we first employed Box-plot analysis and multiple interpolation method(MI)to clean the data. Combined with grey relation analysis(GRA), we established a coal-gas compound dynamic disaster index system. Then, principal component analysis(PCA)is used for dimensionality reduction of the data. Combined with the convolution neural network(CNN)in deep learning, we established the coal-gas compound dynamic disaster prediction model based on BMGP-CNN. The field data is used to compare and verify this model with BP, random forest(RF), support vector machine(SVM)and artificial neural network(ANN). It is found that BMGP-CNN model yields prediction results with satisfactory accuracy and quick convergence. The results offer implications for the prediction and prevention of coal-gas compound dynamic disasters.