Abstract: Blast vibration is a typical harmful effect in rock blast excavation projects.Accurate prediction of blast vibration is of great significance for risk control during blasting.To achieve this goal, the weights and thresholds of the back propagation neural network(BPNN)were optimized with the firefly algorithm(FA),and an FA-BPNN prediction model was established.Taking multiple design parameters and blast center distance of bench blasting in an open-pit mine as the model input parameters, the peak particle velocity(PPV)of blast vibration was predicted, and the prediction results of FA-BP neural network model, statistical prediction equation, BPNN model and random forest method were compared.The sensitivity of each parameter to the prediction results of the FA-BPNN model was analyzed by normalized mutual information.The results indicated that the FA-BPNN model was effective in predicting the PPV,the corresponding root mean square error(RMSE),mean absolute error(MAE),and coefficient of determination were 1.445,1.182,and 0.973 respectively, the FA-BPNN model was more effective than the other three methods.The maximum charge of a single section, the distance between the blasting center, the unit consumption of explosives, and the ratio of step height to the length of resistance line have great influences on the prediction results of the PPV.