For the unsteady aerodynamic characteristics during high angle of attack flight, this study proposes a modeling method (PSO-LSTM) that integrates the Particle Swarm Optimization (PSO) algorithm with a Long Short-Term Memory (LSTM) neural network. The research constructs an unsteady aerodynamic model based on wind tunnel test data of a specific aircraft, where the PSO algorithm optimizes the model's hyperparameters to enhance prediction accuracy. Comparative results with traditional dynamic derivative models demonstrate that the proposed PSO-LSTM model achieves higher predictive precision. The model is further applied to spin simulation, effectively validating its engineering practicality. Additionally, single-parameter variable analysis is conducted to explore the influence mechanisms of different entry methods and moments of inertia on spin characteristics. The results indicate that entry methods significantly affect the aircraft’s sink rate and rotation speed, while the moment of inertia primarily influences longitudinal characteristics.