Research on Adaptive Control Method for Variable-Sweep Wing Aircraft Based on Deep Reinforcement Learning

This study develops an integrated framework for optimizing variable-sweep wing aerodynamics. CFD simulations under steady conditions reveal that a small sweep angle enhances lift in subsonic flow, while a large sweep angle improves the lift-to-drag ratio at transonic and supersonic speeds. A Kriging model, validated with a low cross-validation error of 1.94%, is trained for rapid coefficient prediction. Using this model, a deep reinforcement learning framework based on the PPO algorithm is implemented. The agent successfully learns to adaptively control the sweep angle and angle of attack in unsteady flows, demonstrating the method's effectiveness for intelligent aerodynamic optimization.