Distributed Electric Propulsion Aircraft Attitude Control Based on Assisted Training Multi-Agent Reinforcement Learning

Distributed Electric Propulsion (DEP) aircraft generate differential thrust by rotational speed differential between distributed electric propulsors. This creates additional control torque, which offers the potential for improved flight performance. However, DEP aircraft is a highly coupled and complex system with multiple inputs and outputs. It is crucial to ensure that the different controlled actuators (distributed electric propulsors, rudder, elevator, and ailerons) work in a coordinated manner. To this end, this paper proposes a Memorable Fuzzy Assisted Training Twin Delayed Deep Deterministic Policy Gradient (MFAT-TD3) algorithm. The MFAT-TD3 algorithm reduces the training time and increases the rate of convergence of the reinforcement learning algorithm through a method of decoupling first and then coupling. This also enhances the generalization performance of individual agents, thus achieving the coordination and cooperation between different controlled actuators of the DEP aircraft, so that the aircraft can fly with high agility. The results show that the proposed algorithm can reduce training time, improve flight efficiency and fault tolerance. Even when there are multiple propulsors failed, it is still able to accomplish the combined power yaw control.