Blasting-Induced Ground Vibration Modeling in Tunnel Excavation: A Comparative Study of ANN, Hybrid ANNs, and Empirical Models

Blasting is the most widely adopted tunneling method worldwide because it is economical, inexpensive, and adaptable to all geological terrains compared with mechanical excavation methods. However, the use of explosives and explosive materials for rock mass excavation in tunnels can have deleterious environmental consequences, including ground vibration (PPV), which can damage the peripheral rock mass surrounding the tunnel. This study aims to develop predictive models for estimating PPV in tunnel excavations using 221 field datasets from 5 transportation tunnels. Three machine learning models—Levenberg–Marquardt trained artificial neural network (ANN-LM), grasshopper optimization algorithm-optimized ANN (ANN-GOA), and multiverse optimization algorithm-optimized ANN (ANN-MVO)—were developed to estimate PPV based on seven parameters: charge per delay, number of holes, monitoring distance, hole depth, rock mass rating, total charge, and tunnel cross-sectional area. For comparison, multivariate regression and conventional empirical models were also developed. Based on the aforementioned comparison, the ANN-GOA 7-17-1 model is the most reliable for predicting blasting-induced ground vibrations with the highest coefficient of determination and lowest error indices for the testing dataset. The ANN-GOA 7-17-1 and other developed models were converted into easy-to-use explicit equations to ensure practical application. A sensitivity analysis revealed that the total charge and rock mass rating were PPV’s most influential parameters. The proposed models provide a reliable, cost-effective alternative to traditional field-based methods for predicting tunnel blasting-induced vibrations. This study offers accurate tools for practitioners, reducing the need for extensive and expensive field investigations.