In order to predict the density of explosive column in real time and improve the prediction accuracy, an improved genetic algorithm was used to optimize the BP network (improved genetic algorithm backpropagation neural network, IGA-BPNN) model for predicting explosive density. By dynamically adjusting the crossover probability and mutation probability of GA, the optimal values of BPNN weights and thresholds were determined, and the IGA-BP prediction model was constructed to predict the explosive density based on the collected process parameters. The experimental results show that the improved GA makes a better adjustment to the crossover rate and mutation rate, can quickly search the optimal weight and threshold of BPNN, and improve the prediction accuracy of explosive pressing density.