Prediction of Blasting Vibration Velocity in Open-pit Mine based on MD-PCA-BP Model

In order to address the problem of predicting blasting vibration in complex geological conditions at open-pit mines, an improved BP neural network prediction model based on Mahalanobis distance discrimination(MD) and principal component analysis(PCA), namely MD-PCA-BP model, is proposed. By combining the monitoring data of blasting vibration at Changtan open-pit mine in Inner Mongolia, outliers in the monitoring data are eliminated using the Mahalanobis distance discrimination method. Then, the principal component analysis method is employed to reduce the dimensionality of factors affecting blasting vibration and obtain three principal component factors. The scores of each principal component factor are calculated, and finally a nonlinear relationship between blasting vibration and principal component scores is constructed through BP neural network to establish the prediction model based on MD-PCA-BP. The results show that the fitting degree between predicted values and measured values of blasting vibration velocity prediction model established based on MD-PCA-BP reaches 0.94, indicating high prediction accuracy of this model. When compared with Sadovsky empirical formula, two improved elevation empirical formulas, MD-BP model, PCA-BP model, and BP model, most of the prediction errors of MD-PCA-BP model are within 10%, demonstrating higher reliability and accuracy compared to empirical formulas and unimproved BP prediction models. The blast vibration prediction model based on MD-PCA-BP exhibits good predictive performance for blast vibration velocity in complex terrains.