A Virtual Reinforcement Learning Method for Aero-engine Intelligent Control

Abstract

The aero-engine is a highly intricate thermal mechanical system characterized by significant nonlinearity, uncertainty, and time-varying behavior. As aerospace technology continues to advance, there is an increasing demand for aero-engines to deliver higher levels of performance. Against this background, traditional control methods have shown limitations in achieving optimal outcomes. Intelligent aero-engine technology has emerged as a significant and promising research area. Therefore, a virtual reinforcement learning method for aero-engine intelligent control is proposed in this paper. Firstly, this research establishes a data-driven virtual simulation environment for the aero-engine employing long short-term memory (LSTM) neural networks. Subsequently, the intelligent controller is trained within this environment utilizing the deep deterministic policy gradient (DDPG) algorithm. Finally, we verify the intelligent controller performance with JT9D engine model. Compared with traditional PID control, the intelligent controller has smaller overshoot and shorter setting time.