Peak Particle Velocity Prediction Model Based on I-GWO-BP Network and SHAP Method

Accurate prediction for Peak Particle Velocity (PPV) induced by blasting vibrations is crucial for controlling engineering risks and optimizing blasting parameter design. In order to address the practical challenges of numerous influencing factors and limited available research samples for PPV, a PPV prediction model based on a Backpropagation neural network (BP) optimized by the Improved Grey Wolf Optimizer (I-GWO), termed the I-GWO-BP network, is developed. The I-GWO algorithm was employed to efficiently search for the optimal combination of weights and bias parameters that need to be optimized in the BP neural network. The results indicate that the proposed integrated prediction model forecasts PPV with greater accuracy and reliability than comparative models including the Grey Wolf Optimizer-enhanced BP model (GWO-BP) and Extreme Learning models (ELM). Beyond demonstrating optimal prediction performance, explainability analysis used the SHapley Additive exPlanations (SHAP)method further enhances the I-GWO-BP model’s interpretability, revealing that the distance from the blasting center is the most significant influencing factor. The application potential of the I-GWO-enhanced neural network for predicting PPV under similar geological and operational conditions is highlighted, providing a methodological reference for related prediction work.