Artificial Neural Network Optimized by Modified Particle Swarm Optimization for Predicting Peak Particle Velocity Induced by Blasting Operations in Open Pit Mines

Abstract

Blasting is an indispensable part of the open pit mining operations. It plays a vital role in
preparing the rock mass for subsequent operations, such as loading/unloading, transporting, crushing, and
dumping. However, adverse effects, especially blast-induced ground vibrations, are considered one of the
most dangerous problems. In this study, artificial intelligence was supposed to predict the intensity of
blast-induced ground vibration, which is represented by the peak particle velocity (PPV). Accordingly, an
artificial neural network was designed to predict PPV at the Coc Sau open pit coal mine with 137 blasting
events were collected. Aiming to optimize the ANN model, the modified version of the particle swarm
optimization (MPSO) algorithm was applied to optimize the ANN model for predicting PPV, called the
MPSO-ANN model. For the comparison purposes, two forms of empirical equations, namely United
States Bureau of Mining (USBM) and U Langefors - Kihlstrom, were also developed to predict PPV and
compared with the proposed MPSO-ANN model. The results showed that the proposed MPSO-ANN
model provided a better performance with a mean absolute error (MAE) of 1.217, root-mean-squared error
(RMSE) of 1.456, and coefficient of determination (R2) of 0.956. Meanwhile, the empirical models only
provided poorer performances with an MAE of 1.830 and 2.012, RMSE of 2.268 and 2.464, and R2 of
0.874 and 0.852 for the USBM and U Langefors – Kihlstrom empirical models, respectively.