航空发动机虚拟自学习控制方法研究  

Research on virtual self-learning control method for aero-engine

作　　者：董建华 朱建铭 黎瀚涛 刘文烁 唐炜[1] DONG Jianhua;ZHU Jianming;LI Hantao;LIU Wenshuo;TANG Wei(School of Automation,Northwestern Polytechnical University,Xi’an 710129,China;School of Measuring and Optical Engineering,Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]西北工业大学自动化学院,西安710129 [2]南昌航空大学测试与光电工程学院,南昌330063

出　　处：《航空工程进展》2023年第6期81-90,共10页Advances in Aeronautical Science and Engineering

基　　金：先进航空动力创新工作站(依托中国航空发动机研究院设立)资助(HKCX2020-02-019)。

摘　　要：随着人工智能技术的发展,智能航空发动机逐渐成为当今航空领域研究的热点。传统的航空发动机控制对发动机模型的依赖性过强,而基于发动机气热动力学公式的机理建模会引入较大的建模误差,给控制器设计带来困难。对此,提出一种基于强化学习的航空发动机控制虚拟自学习方法,首先利用航空发动机的试验数据通过LSTM神经网络建立虚拟学习环境,然后采用深度强化学习TD3算法,在虚拟环境中训练智能控制器,最后采用JT9D发动机模型验证智能控制器的性能。结果表明:相比于传统PID控制,智能控制器产生的超调量更小,调节时间更短。With the development of artificial intelligence technology,intelligent aircraft engines have gradually become a hot spot in the field of aviation today.Traditional aero-engine control heavily relies on the engine model,and the theoretical modeling approach based on aerothermodynamic formula introduces modeling error that may degrade the performance of controller.In this paper a virtual self-learning control method for aero-engine intelligent controller design is proposed.Firstly,a virtual environment is established from the testing data of the aero-engine via LSTM neural network.Secondly,the reinforcement learning algorithm based on TD3 is employed for intelligent controller training in the virtual environment.Finally,the JT9D aero-engine model is utilized for controller performance evaluation.The simulation comparisons between intelligent controller and traditional PID control show that the intelligent controller has remarkable performance due to the less overshoot and shorter setting time.

关 键 词：航空发动机 智能控制 强化学习 LSTM神经网络 TD3算法 

分 类 号：V233.7[航空宇航科学与技术—航空宇航推进理论与工程]