Aerodynamic characteristic analysis and prediction for projectiles using particle swarm optimization—back propagation neural network

High-precision projectile aerodynamic parameters are essential for solving external ballistic equations, directly influencing firing accuracy and trajectory optimization. A large-caliber projectile was selected as the research object, and projectile aerodynamic characteristics under various conditions were obtained by Computational Fluid Dynamics (CFD). The accuracy of CFD simulations was confirmed through a comparison with wind tunnel test data from the Army Navy Basic Finner model, showing very good agreement across the entire Mach range. Subsequently, a Backpropagation Neural Network (BPNN) model was developed to predict projectile aerodynamic parameters, with Genetic Algorithm (GA), Gray Wolf Optimizer (GWO), and Particle Swarm Optimization (PSO) applied to optimize its hyperparameters. Based on the PSO-BPNN prediction model, which shows superior performance compared to GA-BPNN and GWO-BPNN, the sensitivity analysis of aerodynamic parameters was performed. Simulation results demonstrate that the prediction performance of the BPNN improves as the sample size increases, and the PSO-BPNN trained on 200 samples outperforms the pure BPNN model using 500 samples across all evaluation metrics, reducing computational cost by 60%. All R2 for projectile aerodynamic parameters exceed 0.99, validating the PSO-BPNN model as an efficient and accurate method for dynamic aerodynamic parameters prediction.