Prediction Method of Tunnel-Blasting Ground Vibration in Sand–Mudstone Interbedded Area Using IPSO-BPNN Algorithm

Addressing the high risks associated with blasting construction in sand–mudstone interbedded areas and the challenges in accurately predicting surface vibration velocity, this study proposes a prediction method for surface vibration of tunnel blasting applicable to interlayered layered rock masses of sandstone and mudstone. Based on this, a prediction model based on an improved particle swarm optimization backpropagation (BP) neural network is proposed to predict surface vibration velocity caused by sand–mudstone interlayer stratified rock mass blasting. In order to further improve the accuracy of the model, rock parameters with sand–mudstone interbed formation characteristics are extracted as the characteristic parameters of the model, and the inertia weight of the model is improved. The improved model is called improved particle swarm optimization—backpropagation neural network (IPSO-BPNN). The results indicate that the number of sandstone and mudstone alternations (N2, N3) and explosive quantity (Q) are key factors influencing vibration velocity, with an impact degree exceeding 0.61. The IPSO-BPNN model has high prediction accuracy with a root mean square error of 1.141. The prediction model that accounts for the macrolevel characteristics of sand–mudstone interbeds demonstrates an average increase of 9.12% in R2 in contrast to the model without considering these characteristics. Compared with the BPNN and the GA-BPNN model, the IPSO-BPNN reduces the average prediction error by 7%. The average prediction error between the vibration velocity of the measurement points in the blasting simulation test and the field test points is 9.44%, and the numerical simulation results are credible. The maximum average error between the predicted surface vibration velocity of the IPSO-BPNN blasting vibration model in a sand–mudstone interbedded area and the measured vibration velocity at different blast sources and monitoring points in blasting simulation experiments is 7.18%, demonstrating high prediction accuracy. These findings provide significant theoretical guidance for blasting vibration prediction in layered structures with variable rock layers.