Design and Realization of Intelligent Aero-engine DDPG Controller

Abstract

Abstract As the artificial intelligence technology advances, intellectualization has become an important development trend for the future aero-engine industry. The intelligent aero-engine will bring a new round of technology revolution in aviation industry. Since an aero-engine is a complex system with strong nonlinearity and coupling, it is difficult to achieve the optimal control performance with traditional PI control. Therefore, the DDPG (Deep Deterministic Policy Gradient, DDPG) algorithm belonging to the family of Deep Reinforcement Learning is hereby proposed for designing aero-engine controller. Based on the non-linear polynomial state-space mathematical model of JT9D turbofan engine, the intelligent DDPG controller is designed and then compared with the performance of PI controller. The proposed controller can achieve the optimal control effect for aero-engine through the training the neural network. The simulation results demonstrates that compared with PI control, the control method proposed in this paper achieves superior control performance in response speed, overshoot and anti-interference ability.