Prediction method of blast load on underground structure surface based on neural network

The dynamic load in the soil directly leads to damage of underground structures upon explosions. In this study, a method predict blast on structure surface based neural network was developed study distribution under close-in detonation. First, taking utility tunnel as experimental structure, 52 groups field tests were conducted mechanism, and data samples obtained. Second, key influencing parameters reflected obtained through dimensional analysis method, backpropagation model constructed test using Levenberg–Marquardt algorithm train optimize network. Finally, accuracy prediction results compared evaluated among network, empirical formula, nonlinear regression (NRA) methods. It is found that input combined variables can further improve with single physical variables. Compared formula NRA provided most accurate prediction. typical conditions calculated by showed explosive setting impact uneven shape surface. increase equivalent depth reduces nonuniformity distribution, while decrease explosion distance increases distribution.