A Control Method for Aero-engine Based on Reinforcement Learning

摘要

An aero-engine is a complex system with strong nonlinearity. To improve the performance of aero-engine control systems, this paper proposed to apply reinforcement learning algorithms in the field of the aero-engine control. Aiming at the problem of the aero-engine rotational speed control in the cruise phase, we proposed a deep deterministic policy gradient(DDPG) aero-engine control method. This method achieves the optimal control effect through training and learning four neural networks. The simulations prove that the intelligent control method can improve the response speeds, overshoots, and anti- interference ability compared with the PID controller.