Enhanced Control System for Morphing Hypersonic Aircraft Based on an Improved Proximal Policy Optimization Algorithm

This paper focuses on the application of deep reinforcement learning, specifically an improved PPO algorithm, to the design of an attitude control system for a hypersonic aircraft with a retractable wingspan. The derivation and enhancement of the PPO algorithm are thoroughly explained prior to the control system design. Following this, an Actor network with superior control performance is trained offline and utilized as the controller. Three-channel attitude tracking tests are conducted under three conditions: no noise, small noise, and large noise. Results demonstrate that the attitude controller based on the improved PPO algorithm achieves high accuracy and robust anti-jamming performance. This highlights the unique advantages of deep reinforcement learning in addressing highly nonlinear, uncertain, and time-varying control problems.