Predicting peak particle velocity in pre-splitting of gas-producing devices using improved particle swarm optimization algorithm

Advance prediction and control of vibration velocity during blasting construction is essential to protect the tunnel and surrounding buildings. In this study, we propose an Improved Particle Swarm Optimization (IPSO) Backpropagation (BP) Neural Network Model (IPSO-BP) considering the effect of frequency (f) to provide a highly accurate prediction tool for predicting the peak particle velocity (PPV) generated from pre-fracture construction of gas-producing tools.The validity and consistency of this model are compared with the traditional particle swarm algorithm optimized BP neural network (PSO-BP), standard BP neural network, empirical formula, genetic-adaptive particle swarm hybrid algorithm optimized BP neural network model (GA-APSO-BP), Gray Wolf Algorithm Optimized Support Vector Regression model (GWO-SVR), Moth Flame Algorithm Optimized BP neural network model (MFO-BP), and the Rungekuta algorithm optimized extreme gradient boosting tree model (RUN-XGBoost) for multi-dimensional comparative analysis. The results show that the IPSO-BP model outperforms other models in prediction performance, both in the training set, testability, and overall. This study proves the reliability and accuracy of the IPSO-BP model in predicting the PPVs, which can contribute to improving the safety of the blasting construction.