Construction Method of Shock Wave Field Proxy Model Based on Physical Information Neural Network

This study is dedicated to constructing a surrogate model of the shock wave field using physics-informed neural networks, aiming to achieve accurate reconstruction of the shock wave field with a low amount of data. The key to this method lies in the integration of the physical laws of shock wave propagation, which aims to overcome the insufficient reconstruction accuracy caused by sparse measurement points in traditional testing. This effectively reduces the safety risk of non-combat casualties. Through experimental validation, the method proposed in this paper achieves an average absolute error of less than 0.5541 kPa when dealing with small datasets; using 56 training data points for testing, the improved PINN's relative L2 Norm error is reduced by 25.876% compared to the BP neural network, and the average absolute error decreases by 9.5%.