中国科技核心期刊

中文核心期刊

CSCD来源期刊

空间控制技术与应用 ›› 2021, Vol. 47 ›› Issue (1): 29-39.doi: 10.3969/j.issn.1674-1579.2021.01.005

• 论文与报告 • 上一篇    下一篇

基于模糊强化学习的多柔性梁振动控制仿真

  

  • 出版日期:2021-02-25 发布日期:2021-03-11
  • 基金资助:
    国家自然科学基金资助项目(51775190)

Simulation of Multiple Flexible Beams Vibration Control Based on Fuzzy Reinforcement Learning

  • Online:2021-02-25 Published:2021-03-11

摘要: 空间机械结构趋向大型化、复杂化、柔性化特点,容易导致残余振动,且残余振动频率低,振动时间长.针对多柔性梁耦合结构残余振动问题,通过有限元法建立了动力学模型,分析了振动特性,呈现密频特性,拍频特征.对于残余振动问题,结合模糊强化学习控制算法,构建模糊规则表,使用ε贪婪法选择每条规则中的动作,进而生成最终的控制电压,与环境交互后获得回报,利用时序差分误差对状态动作价值进行反馈学习.经过训练,控制器收敛于一个模糊控制规则.仿真结果显示模糊强化学习控制器对多柔性梁残余振动快速抑制,验证了模糊强化学习控制算法的有效性.

关键词: 多柔性梁, 振动控制, 有限元建模, 模糊强化学习

Abstract: Space mechanical structures tend to be largescale, complex and flexible, which easily leads to residual vibration, and the frequency of residual vibration is low and the vibration time is long. Aiming at the residual vibration problem of multi flexible beams coupling structure, the dynamic model is established by finite element method, and the vibration characteristics are analyzed, showing the characteristics of dense frequency and beat frequency. For the residual vibration problem, combined with fuzzy reinforcement learning control algorithm, the fuzzy rule table is constructed. The action in each rule is selected by εgreedy method, and the control voltage is generated in the end. The feedback learning of stateaction value is carried out via temporal difference (TD) error. After training, the controller converges to a fuzzy control rule. The simulation results show that the fuzzy reinforcement learning controller can quickly suppress the residual vibration of multi flexible beams, and verify the effectiveness of the fuzzy reinforcement learning control algorithm.

Key words: multi flexible beams, vibration control, finite element modeling, fuzzy reinforcement learning

中图分类号: 

  • V476.1