航空学报 > 2019, Vol. 40 Issue (3): 322286-322286   doi: 10.7527/S1000-6893.2018.22286

状态预测神经网络控制应用于小型可回收火箭

陈书钊1, 楚龙飞1, 杨秀梅2, 蔡德淮1   

  1. 1. 翎客航天科技有限公司, 北京 100176;
    2. 昆明理工大学 信息工程与自动化学院, 昆明 650500
  • 收稿日期:2018-05-07 修回日期:2018-07-09 出版日期:2019-03-15 发布日期:2018-10-31
  • 通讯作者: 陈书钊 E-mail:chens1905@163.com

Application of state prediction neural network control algorithm in small reusable rocket

CHEN Shuzhao1, CHU Longfei1, YANG Xiumei2, CAI Dehuai1   

  1. 1. LinkSpace Aerospace Technology Group, Beijing 100176, China;
    2. The Academic Institute of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China
  • Received:2018-05-07 Revised:2018-07-09 Online:2019-03-15 Published:2018-10-31

摘要: 随着商业航天的到来,可重复使用运载器的研究受到广泛关注,以SpaceX为代表的商业航天公司研发的部分可回收火箭表现出了前所未有的竞争力。为了研发可回收火箭技术,翎客航天利用民间工业力量研制了RLV-T3小型可回收火箭验证机,并在该验证机上通过数百次试验逐渐掌握了垂直起降(VTVL)技术。主要介绍了翎客航天在VTVL技术中的一项动力控制技术,提出了状态预测神经网络控制(SPNNC)算法。该算法具有鲁棒性强、适用范围广、控制参数易调整等优点。详细地描述了该算法的原理,并通过Simulink对SISO和MIMO 2种系统进行了仿真。同时详细地论述了将状态预测神经网络控制算法应用于RLV-T3小型可回收火箭的飞行及回收的试验,包括RLV-T3小型可回收火箭的基本特点、控制难点、存在的问题,飞行过程中各物理量的曲线和试验结论。经试验验证,状态预测神经网络控制算法具有良好的控制性能,基于该控制技术,即状态预测神经网络控制算法的RLV-T3小型可回收火箭验证机可以安全地实现垂直起飞、弹道飞行、空中悬停、软着陆回收全流程。

关键词: 神经网络控制, 智能控制, 状态预测, 可重复利用火箭, 火箭回收, 商业航天

Abstract: With the advent of commercial aerospace exploration, research on reusable vehicles has received extensive attention. Some reusable rockets developed by commercial space companies represented by SpaceX have exhibited unprecedented competitiveness. To develop reusable rocket technology, LinkSpace has developed a small verifying rocket called RLV-T3, and gradually mastered the Vertical Takeoff and Veratical Landing (VTVL) technology through hundreds of tests on this verification machine. This paper mainly introduces the State Prediction Neural Network Control (SPNNC) algorithm, a thrust control technique in VTVL technology. The algorithm has strong robustness, wide application range, and easy adjustment of control parameters. This paper describes the principle of the algorithm in detail and simulates both SISO and MIMO systems using Simulink. At the same time, this paper discusses in detail the test of the SPNNC applied to the RLV-T3, including the basic characteristics of the small reusable rocket, control difficulties, existing problems, the flight curve of the physical quantities, and test conclusions. It has been verified by experiments that the SPNNC has good control performance, and the small recyclable rocket verification machine named RLV-T3 based on SPNNC can safely implement the whole process of vertical takeoff, ballistic flight, air hovering, and soft landing recovery.

Key words: neural network control, intelligent control, state prediction, reusable rocket, rocket recovery, commercial aerospace

中图分类号: